\n

Ori entation for new MSE in Financial Mathematics students will be held August 12 – August 25.

\nA full orientation schedule for 2015 will be available shortly before the st art of orientation.

\n\n X-TAGS;LANGUAGE=en-US:Financial Mathematics END:VEVENT BEGIN:VEVENT UID:ai1ec-5853@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;VALUE=DATE:20150827 DTEND;VALUE=DATE:20150828 SEQUENCE:0 SUMMARY:Fall Classes Begin URL:https://engineering.jhu.edu/ams/events/fall-classes-begin/ X-TAGS;LANGUAGE=en-US:current students END:VEVENT BEGIN:VEVENT UID:ai1ec-5852@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES;LANGUAGE=en-US:PhD CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20150819T083000 DTEND;TZID=America/New_York:20150819T170000 LOCATION:Whitehead 304 SEQUENCE:0 SUMMARY:PhD Introductory Exam URL:https://engineering.jhu.edu/ams/events/phd-introductory-exam-2/ X-TAGS;LANGUAGE=en-US:Intro Exam\,Orientation\,PhD END:VEVENT BEGIN:VEVENT UID:ai1ec-5976@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20150826T143000 DTEND;TZID=America/New_York:20150826T163000 SEQUENCE:0 SUMMARY:Teaching Assistant Orientation: Meet in Hodson 316 URL:https://engineering.jhu.edu/ams/events/teaching-assistant-orientation-m eet-in-hodson-316/ END:VEVENT BEGIN:VEVENT UID:ai1ec-5974@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20150827T133000 DTEND;TZID=America/New_York:20150827T143000 SEQUENCE:0 SUMMARY:Seminar: Get to Know You @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-get-to-know-you/ END:VEVENT BEGIN:VEVENT UID:ai1ec-5959@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:A GENERAL THEORY FOR COMPUTING ATTRACTIVE REPRESENTATIONS ON NO NCONVEX OPTIMIZATION PROBLEMS\nMore than one mathematical representation c an accurately depict a decision problem. Success in obtaining optimal solu tions\, however\, often depends upon the formulation selected. Since chall enging nonconvex optimization problems are typically solved by using linea r programming relaxations as tools to compute bounds for eliminating infer ior solutions\, “attractive” representations tend to be characterized by t he accuracy of their relaxations. This importance of relaxation strength i s well documented within the Operations Research literature\, where numero us authors have suggested methods for acquiring strength. The posed method s are often problem dependent\, relying on the exploitation of specific st ructures.\nThis talk presents a general theory for deriving representation s with tight relaxations. The fundamental idea is to recast a given proble m into higher-dimensional spaces by automatically generating auxiliary var iables and constraints. Strength is garnered via suitable mathematical ide ntities. The talk begins with an introduction to the importance of relaxat ion strength\, and then highlights contributions and challenges relative t o the progressively more general families of mixed-binary\, mixed-discrete \, and general nonconvex programs. Ongoing research is discussed. DTSTART;TZID=America/New_York:20150903T133000 DTEND;TZID=America/New_York:20150903T143000 SEQUENCE:0 SUMMARY:Seminar: Warren Adams (Clemson University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-warren-adams-clemson-uni versity-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

A GENERAL THE ORY FOR COMPUTING ATTRACTIVE REPRESENTATIONS ON NONCONVEX OPTIMIZATION PRO BLEMS

\nMore than one mathematical representation can accurately dep ict a decision problem. Success in obtaining optimal solutions\, however\, often depends upon the formulation selected. Since challenging nonconvex optimization problems are typically solved by using linear programming rel axations as tools to compute bounds for eliminating inferior solutions\, “ attractive” representations tend to be characterized by the accuracy of th eir relaxations. This importance of relaxation strength is well documented within the Operations Research literature\, where numerous authors have s uggested methods for acquiring strength. The posed methods are often probl em dependent\, relying on the exploitation of specific structures.

\nThis talk presents a general theory for deriving representations with tig ht relaxations. The fundamental idea is to recast a given problem into hig her-dimensional spaces by automatically generating auxiliary variables and constraints. Strength is garnered via suitable mathematical identities. T he talk begins with an introduction to the importance of relaxation streng th\, and then highlights contributions and challenges relative to the prog ressively more general families of mixed-binary\, mixed-discrete\, and gen eral nonconvex programs. Ongoing research is discussed.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5941@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:SUBGROUP-BASED ADAPTIVE (SUBA) ENRICHMENT DESIGNS FOR MULT-ARM BIOMARKER TRIALS\nTargeted therapies based on biomarker profiling are beco ming a mainstream direction of cancer research and treatment. Depending on the expression of specific prognostic biomarkers\, targeted therapies ass ign different cancer drugs to subgroups of patients even if they are diagn osed with the same type of cancer by traditional means\, such as tumor loc ation. For example\, Herceptin is only indicated for the subgroup of patie nts with HER2+ breast cancer\, but not other types of breast cancer. Howev er\, subgroups like HER2+ breast cancer with effective targeted therapies are rare and most cancer drugs are still being applied to large patient po pulations that include many patients who might not respond or benefit. Als o\, the response to targeted agents in humans is usually unpredictable. To address these issues\, we propose SUBA\, subgroup-based adaptive designs that simultaneously search for prognostic subgroups and allocate patients adaptively to the best subgroup-specific treatments throughout the course of the trial. The main features of SUBA include the continuous reclassific ation of patient subgroups based on a random partition model and the adapt ive allocation of patients to the best treatment arm based on posterior pr edictive probabilities. We compare the SUBA design with three alternative designs including equal randomization\, outcome-adaptive randomization and a design based on a probit regression. In simulation studies we find that SUBA compares favorably against the alternatives. DTSTART;TZID=America/New_York:20150910T133000 DTEND;TZID=America/New_York:20150910T143000 SEQUENCE:0 SUMMARY:Seminar: Yanxun Xu (Johns Hopkins University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-yanxun-xu-johns-hopkins- university/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nSUBGROUP-BASE D ADAPTIVE (SUBA) ENRICHMENT DESIGNS FOR MULT-ARM BIOMARKER TRIALS

\nTargeted therapies based on biomarker profiling are becoming a mainstream direction of cancer research and treatment. Depending on the expression o f specific prognostic biomarkers\, targeted therapies assign different can cer drugs to subgroups of patients even if they are diagnosed with the sam e type of cancer by traditional means\, such as tumor location. For exampl e\, Herceptin is only indicated for the subgroup of patients with HER2+ br east cancer\, but not other types of breast cancer. However\, subgroups li ke HER2+ breast cancer with effective targeted therapies are rare and most cancer drugs are still being applied to large patient populations that in clude many patients who might not respond or benefit. Also\, the response to targeted agents in humans is usually unpredictable. To address these is sues\, we propose SUBA\, subgroup-based adaptive designs that simultaneous ly search for prognostic subgroups and allocate patients adaptively to the best subgroup-specific treatments throughout the course of the trial. The main features of SUBA include the continuous reclassification of patient subgroups based on a random partition model and the adaptive allocation of patients to the best treatment arm based on posterior predictive probabil ities. We compare the SUBA design with three alternative designs including equal randomization\, outcome-adaptive randomization and a design based o n a probit regression. In simulation studies we find that SUBA compares fa vorably against the alternatives.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5983@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:THE GENERAL SETTING FOR SHAPE DEFORMATION ANALYSIS\nI will defi ne a unified setting for shape registration and LDDMM methods for shape an alysis\, using optimal control theory\, and give the Hamiltonian geodesic equations associated to a smooth enough reproducing kernel. I will then gi ve several applications of this framework\, such as fibered shapes (for mu scles)\, and the addition of constraints for the simultaneous study of mul tiple interacting shapes. DTSTART;TZID=America/New_York:20150917T133000 DTEND;TZID=America/New_York:20150917T143000 SEQUENCE:0 SUMMARY:Seminar: Sylvain Arguillere (Johns Hopkins University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-sylvain-arguillere-johns -hopkins-university-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTHE GENERAL S ETTING FOR SHAPE DEFORMATION ANALYSIS

\nI will define a unified sett ing for shape registration and LDDMM methods for shape analysis\, using op timal control theory\, and give the Hamiltonian geodesic equations associa ted to a smooth enough reproducing kernel. I will then give several applic ations of this framework\, such as fibered shapes (for muscles)\, and the addition of constraints for the simultaneous study of multiple interacting shapes.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5855@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES;LANGUAGE=en-US:WSE CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20150918T120000 DTEND;TZID=America/New_York:20150918T140000 SEQUENCE:0 SUMMARY:WSE Annual Fall Picnic URL:https://engineering.jhu.edu/ams/events/wse-annual-fall-picnic-2/ X-TAGS;LANGUAGE=en-US:current students\,faculty\,staff END:VEVENT BEGIN:VEVENT UID:ai1ec-5938@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:CHALLENGES IN GRAPH-BASED MACHINE LEARNING AND ROBUSTIFYING DAT A GRAPHS WITH SCALABLE LOCAL SPECTRAL METHODS\nGraphs are very popular way s to model data in many data analysis and machine learning applications\, but they can be quite challenging to work with\, especially when they are very sparse\, as is typically the case. We will discuss challenges we hav e encountered in working with large sparse graphs in machine learning and data analysis applications and in particular in the construction of these graphs\, e.g.\, with various sorts of popular nearest neighbor rules appli ed to feature vectors. In our experience\, many properties of the constru cted graphs are very sensitive to seemingly-minor and often-ignored aspect s of the graph construction process. This should suggest caution in using popular algorithmic and statistical tools\, e.g.\, popular nonlinear dime nsionality reduction methods\, in trying to extract insight from those con structed graphs. We will also describe recent results on using local spec tral methods to robustify this graph construction process. Local spectral methods use locally-biased random walks\, they have had several remarkabl e successes in worst-case algorithm design as well as in analyzing the emp irical properties of large social and information networks\, and they are an example of a worst-case approximation algorithm that implicitly but exa ctly implements a form of statistical regularization. Informally\, the re ason for the successes of these methods in robustifying graph construction is that these local random walks provide a regularized or stable version of an eigenvector\, and initial results on using these ideas to robustify the graph construction process are promising. DTSTART;TZID=America/New_York:20150924T133000 DTEND;TZID=America/New_York:20150924T143000 SEQUENCE:0 SUMMARY:Seminar: Michael Mahoney (University of California\, Berkeley) @ Wh itehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-michael-mahoney-uc-berke ley/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nCHALLENGES IN GRAPH-BASED MACHINE LEARNING AND ROBUSTIFYING DATA GRAPHS WITH SCALABLE L OCAL SPECTRAL METHODS

\nGraphs are very popular ways to model data i n many data analysis and machine learning applications\, but they can be q uite challenging to work with\, especially when they are very sparse\, as is typically the case. We will discuss challenges we have encountered in working with large sparse graphs in machine learning and data analysis app lications and in particular in the construction of these graphs\, e.g.\, w ith various sorts of popular nearest neighbor rules applied to feature vec tors. In our experience\, many properties of the constructed graphs are v ery sensitive to seemingly-minor and often-ignored aspects of the graph co nstruction process. This should suggest caution in using popular algorith mic and statistical tools\, e.g.\, popular nonlinear dimensionality reduct ion methods\, in trying to extract insight from those constructed graphs. We will also describe recent results on using local spectral methods to r obustify this graph construction process. Local spectral methods use loca lly-biased random walks\, they have had several remarkable successes in wo rst-case algorithm design as well as in analyzing the empirical properties of large social and information networks\, and they are an example of a w orst-case approximation algorithm that implicitly but exactly implements a form of statistical regularization. Informally\, the reason for the succ esses of these methods in robustifying graph construction is that these lo cal random walks provide a regularized or stable version of an eigenvector \, and initial results on using these ideas to robustify the graph constru ction process are promising.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5702@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES;LANGUAGE=en-US:Seminar CONTACT: DESCRIPTION:GAUGE DUALITY AND LOW-RANK SPECTRAL OPTIMIZATION\n \nGauge func tions significantly generalize the notion of a norm\, and\ngauge optimizat ion is the class of problems for finding the element of\na convex set that is minimal with respect to a gauge. These\nconceptually simple problems a ppear in a remarkable array of\napplications. Their structure allows for a special kind of duality\nframework that can lead to new algorithmic appro aches to challenging\nproblems. Low-rank spectral optimization problems th at arise in two\nsignal-recovery application\, phase retrieval and blind d econvolution\,\nillustrate the benefits of the approach. DTSTART;TZID=America/New_York:20151001T133000 DTEND;TZID=America/New_York:20151001T143000 SEQUENCE:0 SUMMARY:Seminar: Michael Friedlander (University of California\, Davis) @ W hitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-michael-friedlander-univ ersity-of-california-davis/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nGAUGE DUALITY AND LOW-RANK SPECTRAL OPTIMIZATION

\n\n

Gauge functions sign ificantly generalize the notion of a norm\, and

\ngauge optimization is the class of problems for finding the element of

\na convex set that is minimal with respect to a gauge. These

\nconceptually simple problems appear in a remarkable array of

\napplications. Their stru cture allows for a special kind of duality

\nframework that can lead to new algorithmic approaches to challenging

\nproblems. Low-rank s pectral optimization problems that arise in two

\nsignal-recovery ap plication\, phase retrieval and blind deconvolution\,

\nillustrate t he benefits of the approach.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5937@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES;LANGUAGE=en-US:Seminar CONTACT: DESCRIPTION:Change Point Inference for Time-varying Erdos-Renyi Graphs\nWe investigate a model of an Erdos-Renyi graph\, where the edges can be in a present/absent state. The states of each edge evolve as a Markov chain ind ependently of the other edges\, and whose parameters exhibit a change-poin t behavior in time. We derive the maximum likelihood estimator for the cha nge-point and characterize its distribution. Depending on a measure of the signal-to-noise ratio present in the data\, different limiting regimes em erge. Nevertheless\, a unifying adaptive scheme can be used in practice th at covers all cases.We illustrate the model and its flexibility on US Cong ress voting patterns using roll call data. DTSTART;TZID=America/New_York:20151008T133000 DTEND;TZID=America/New_York:20151008T143000 SEQUENCE:0 SUMMARY:Seminar: George Michailidis (University of Florida) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-george-michailidis-unive rsity-of-florida/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nChange Point Inference for Time-varying Erdos-Renyi Graphs

\nWe investigate a mod el of an Erdos-Renyi graph\, where the edges can be in a present/absent st ate. The states of each edge evolve as a Markov chain independently of the other edges\, and whose parameters exhibit a change-point behavior in tim e. We derive the maximum likelihood estimator for the change-point and cha racterize its distribution. Depending on a measure of the signal-to-noise ratio present in the data\, different limiting regimes emerge. Nevertheles s\, a unifying adaptive scheme can be used in practice that covers all cas es.We illustrate the model and its flexibility on US Congress voting patte rns using roll call data.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6073@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES;LANGUAGE=en-US:HUSAM CONTACT: DESCRIPTION:Lies\, Deceit\, and Misrepresentation: The Distortion of Statis tics in America\nH.G. Wells once said “Statistical thinking will one day b e as necessary for efficient citizenship as the ability to read and write. ” The widespread use of statistics plays an influential role in persuading public opinion. As such\, statistical literacy is necessary for members o f society to critically evaluate the bombardment of charts\, polls\, graph s\, and data that are presented on a daily basis. However\, what often pas ses for “statistical” calculations and discoveries need to be taken with a grain of salt. This talk will examine the applications of statistics in A merican media and give examples of where statistics has been grossly misus ed.\nThe talk will begin at 7pm in Hodson 110\, with refreshments being se rved at 6:30. A flyer for the event is attached and a link to RSVP on the Facebook page is here: https://www.facebook.com/events/959982947374497/. DTSTART;TZID=America/New_York:20151021T183000 LOCATION:Hodson 110 SEQUENCE:0 SUMMARY:HUSAM: Dr. Talitha Williams\, Harvey Mudd College URL:https://engineering.jhu.edu/ams/events/husam-dr-talitha-williams-harvey -mudd-college/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nLies\, Deceit \, and Misrepresentation: The Distortion of Statistics in America

\nH.G. Wells once said “Statistical thinking will one day be as necessary fo r efficient citizenship as the ability to read and write.” The widespread use of statistics plays an influential role in persuading public opinion. As such\, statistical literacy is necessary for members of society to crit ically evaluate the bombardment of charts\, polls\, graphs\, and data that are presented on a daily basis. However\, what often passes for “statisti cal” calculations and discoveries need to be taken with a grain of salt. T his talk will examine the applications of statistics in American media and give examples of where statistics has been grossly misused.

\nThe t alk will begin at 7pm in Hodson 110\, with refreshments being served at 6: 30. A flyer for the event is attached and a link to RSVP on the Facebook p age is here: ht tps://www.facebook.com/events/959982947374497/.

\n X-TAGS;LANGUAGE=en-US:HUSAM X-INSTANT-EVENT:1 END:VEVENT BEGIN:VEVENT UID:ai1ec-5989@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title: Information Theoretic Intuitions about Some Estimation P roblems in Speech Recognition\nAbstract:\nAutomatic speech recognition (AS R) systems compose probabilistic models of numerous kinds to transcribe a spoken utterance into a sequence of written words. The so called acoustic model aims to compute the probability of each sound category label\, such as a phoneme or an allophone\, given a short segment of speech. The para meters of the acoustic model are typically estimated to maximize discrimin ation between the correct and incorrect sound categories on some labeled “ training” samples—specifically to maximize a mutual information. The inve ntory of sound categories (allophones) the acoustic model is trained to di scriminate is also determined from data. This is typically done using dec ision trees to recursively divide all acoustic samples of a phoneme\, base d on the phonetic context of the sample\, into maximally homogeneous subse ts—specifically\, subsets that minimize a conditional entropy.\nTwo recent advances\, one each in acoustic model estimation and in the creation of p honetic decision trees\, will be described\, beginning with the informatio n theoretic intuitions behind the changes we made to currently used method s. The first replaces the maximization of mutual information with minimiz ation of a related conditional entropy\, which turns out to be advantageou s for semi-supervised training of acoustic models\, i.e. when some samples have missing labels. The second investigates an alternative to random f orests by developing multiple decisions trees in a deterministic manner\; it maximizes diversity by minimizing mutual information between the leaves assigned by the multiple trees to each training sample.\nThe presentation will have a tutorial flavor for a non-speech-technology audience\, and wi ll not present any new results in mathematics or statistics\; just some ma thematical intuitions about estimation techniques that appear to improve s peech recognition performance on benchmark data sets.\nBiosketch:\nSanjeev Khudanpur (PhD 1997\, Electrical Engineering\, University of Maryland) is an Associate Professor in the Departments of Electrical and Computer Engi neering and of Computer Science\, the Acting Director of the Center for La nguage and Speech Processing\, and a founding affiliate of the Human Langu age Technology Center of Excellence\, all in The Johns Hopkins University. His interests are in the application of statistical methods to speech an d text processing\, and to other engineering problems involving time-serie s data. His office is in Hackerman Hall\, the Homewood campus building wi th the surgical robots! DTSTART;TZID=America/New_York:20151022T133000 DTEND;TZID=America/New_York:20151022T143000 SEQUENCE:0 SUMMARY:Seminar: Sanjeev Khudanpur (Johns Hopkins University) @ Whitehead 3 04 URL:https://engineering.jhu.edu/ams/events/seminar-sanjeev-khudanpu-johns-h opkins-university-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
**: *Information Theoretic Intuitions about Some Estimation Prob
lems in Speech Recognition*

**Abstract**:

Automatic speech recognition (ASR) systems compose probabilistic models o f numerous kinds to transcribe a spoken utterance into a sequence of writt en words. The so called acoustic model aims to compute the probability of each sound category label\, such as a phoneme or an allophone\, given a s hort segment of speech. The parameters of the acoustic model are typicall y estimated to maximize discrimination between the correct and incorrect s ound categories on some labeled “training” samples—specifically to maximiz e a mutual information. The inventory of sound categories (allophones) th e acoustic model is trained to discriminate is also determined from data. This is typically done using decision trees to recursively divide all aco ustic samples of a phoneme\, based on the phonetic context of the sample\, into maximally homogeneous subsets—specifically\, subsets that minimize a conditional entropy.

\nTwo recent advances\, one each in acoustic m odel estimation and in the creation of phonetic decision trees\, will be d escribed\, beginning with the information theoretic intuitions behind the changes we made to currently used methods. The first replaces the maximiz ation of mutual information with minimization of a related conditional ent ropy\, which turns out to be advantageous for semi-supervised training of acoustic models\, i.e. when some samples have missing labels. The second investigates an alternative to random forests by developing multiple deci sions trees in a deterministic manner\; it maximizes diversity by minimizi ng mutual information between the leaves assigned by the multiple trees to each training sample.

\nThe presentation will have a tutorial flavo r for a non-speech-technology audience\, and will not present any new resu lts in mathematics or statistics\; just some mathematical intuitions about estimation techniques that appear to improve speech recognition performan ce on benchmark data sets.

\n**Biosketch**:

San jeev Khudanpur (PhD 1997\, Electrical Engineering\, University of Maryland ) is an Associate Professor in the Departments of Electrical and Computer Engineering and of Computer Science\, the Acting Director of the Center fo r Language and Speech Processing\, and a founding affiliate of the Human L anguage Technology Center of Excellence\, all in The Johns Hopkins Univers ity. His interests are in the application of statistical methods to speec h and text processing\, and to other engineering problems involving time-s eries data. His office is in Hackerman Hall\, the Homewood campus buildin g with the surgical robots!

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5960@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Recent Results on Polynomial Optimization Problems\nPolynomial optimization problems\, as the name suggests\, are optimization problem wh ere the objective function as well as the constraints are described by pol ynomials. Such problems have acquired increased interest to some degree b ecause of applications in engineering and science\, where constraints aris e because of physics\, and also because of increased theoretical understan ding. In this talk I will focus on two topics where I am working\, the CD T (Celis Dennis Tapia) problem\, which concerns the solution of a system o f quadratic inequalities over R^n\, and mixed-integer polynomial optimizat ion problems over graphs with structural sparsity\, i.e. low treewidth. W e will describe our results\, but also we will discuss how these problems relate to classical problems in various branches of mathematics. DTSTART;TZID=America/New_York:20151029T133000 DTEND;TZID=America/New_York:20151029T143000 SEQUENCE:0 SUMMARY:Goldman Lecture Series: Daniel Bienstock (Columbia University) @ Kr ieger 205 URL:https://engineering.jhu.edu/ams/events/seminar-daniel-bienstock-columbi a-university-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nRecent Result s on Polynomial Optimization Problems

\nPolynomial optimization prob lems\, as the name suggests\, are optimization problem where the objective function as well as the constraints are described by polynomials. Such p roblems have acquired increased interest to some degree because of applica tions in engineering and science\, where constraints arise because of phys ics\, and also because of increased theoretical understanding. In this ta lk I will focus on two topics where I am working\, the CDT (Celis Dennis T apia) problem\, which concerns the solution of a system of quadratic inequ alities over R^n\, and mixed-integer polynomial optimization problems over graphs with structural sparsity\, i.e. low treewidth. We will describe o ur results\, but also we will discuss how these problems relate to classic al problems in various branches of mathematics.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5975@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Graduate Research Opportunities in AMS\nThis seminar will famil iarize Master’s and PhD students from AMS or other WSE departments with th e research performed by the AMS faculty. It will be composed of a researc h overview presented by Professor Laurent Younes and consisting of snap-sh ot descriptions of research projects currently underway as well as others that are ripe for students to tackle immediately\, followed by a discussio n session between faculty and students.\n \nResearch Opportunities in Appl ied Mathematics and Statistics DTSTART;TZID=America/New_York:20151105T133000 DTEND;TZID=America/New_York:20151105T143000 SEQUENCE:0 SUMMARY:Seminar: The AMS Faculty URL:https://engineering.jhu.edu/ams/events/seminar-laurent-younes-johns-hop kins-university/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nGraduate Rese arch Opportunities in AMS

\nThis seminar will familiarize Master’s a nd PhD students from AMS or other WSE departments with the research perfor med by the AMS faculty. It will be composed of a research overview presen ted by Professor Laurent Younes and consisting of snap-shot descriptions o f research projects currently underway as well as others that are ripe for students to tackle immediately\, followed by a discussion session between faculty and students.

\n\n

Research Opp ortunities in Applied Mathematics and Statistics

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5734@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES;LANGUAGE=en-US:Seminar CONTACT: DESCRIPTION:Scaling and Generalizing Variational Inference\n \nLatent varia ble models have become a key tool for the modern statistician\, letting us express complex assumptions about the hidden structures that underlie our data. Latent variable models have been successfully applied in numerous f ields.\nThe central computational problem in latent variable modeling is p osterior inference\, the problem of approximating the conditional distribu tion of the latent variables given the observations.\nPosterior inference is central to both exploratory tasks and predictive tasks. Approximate po sterior inference algorithms have revolutionized Bayesian statistics\, rev ealing its potential as a usable and general-purpose language for data ana lysis.\nBayesian statistics\, however\, has not yet reached this potential .\nFirst\, statisticians and scientists regularly encounter massive data s ets\, but existing approximate inference algorithms do not scale well.\nSe cond\, most approximate inference algorithms are not generic\; each must b e adapted to the specific model at hand.\nIn this talk I will discuss our recent research on addressing these two limitations. I will describe stoc hastic variational inference\, an approximate inference algorithm for hand ling massive data sets. I will demonstrate its application to probabilist ic topic models of text conditioned on millions of articles. Then I will d iscuss black box variational inference. Black box inference is a generic algorithm for approximating the posterior. We can easily apply it to many models with little model-specific derivation and few restrictions on thei r properties. I will demonstrate its use on a suite of nonconjugate model s of longitudinal healthcare data.\n \nBiography:\nDavid Blei is a Profess or of Statistics and Computer Science at Columbia University\, and a membe r of the Columbia Data Science Institute. His research is in statistical machine learning\, involving probabilistic topic models\, Bayesian nonpara metric methods\, and approximate posterior inference algorithms for massiv e data. He works on a variety of applications\, including text\, images\, music\, social networks\, user behavior\, and scientific data. David has received several awards for his research\, including a Sloan Fellowship ( 2010)\, Office of Naval Research Young Investigator Award (2011)\, Preside ntial Early Career Award for Scientists and Engineers (2011)\, Blavatnik F aculty Award (2013)\, and ACM-Infosys Foundation Award (2013).\n DTSTART;TZID=America/New_York:20151112T133000 DTEND;TZID=America/New_York:20151112T143000 SEQUENCE:0 SUMMARY:Seminar: David Blei (Columbia University) @ Krieger 205 URL:https://engineering.jhu.edu/ams/events/seminar-david-blei-columbia-univ ersity/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nScaling and G eneralizing Variational Inference

\n\n

Latent variable models have become a key tool for the modern statistician\, letting us express c omplex assumptions about the hidden structures that underlie our data. Lat ent variable models have been successfully applied in numerous fields.

\nThe central computational problem in latent variable modeling is post erior inference\, the problem of approximating the conditional distributio n of the latent variables given the observations.

\nPosterior infere nce is central to both exploratory tasks and predictive tasks. Approximat e posterior inference algorithms have revolutionized Bayesian statistics\, revealing its potential as a usable and general-purpose language for data analysis.

\nBayesian statistics\, however\, has not yet reached thi s potential.

\nFirst\, statisticians and scientists regularly encoun ter massive data sets\, but existing approximate inference algorithms do n ot scale well.

\nSecond\, most approximate inference algorithms are not generic\; each must be adapted to the specific model at hand.

\nIn this talk I will discuss our recent research on addressing these two li mitations. I will describe stochastic variational inference\, an approxim ate inference algorithm for handling massive data sets. I will demonstrat e its application to probabilistic topic models of text conditioned on mil lions of articles. Then I will discuss black box variational inference. B lack box inference is a generic algorithm for approximating the posterior. We can easily apply it to many models with little model-specific derivat ion and few restrictions on their properties. I will demonstrate its use on a suite of nonconjugate models of longitudinal healthcare data.

\n\n

Biography:

\nDavid Blei is a Professor of Statistics and Computer Science at Columbia University\, and a member of the Columbia Dat a Science Institute. His research is in statistical machine learning\, in volving probabilistic topic models\, Bayesian nonparametric methods\, and approximate posterior inference algorithms for massive data. He works on a variety of applications\, including text\, images\, music\, social netwo rks\, user behavior\, and scientific data. David has received several awa rds for his research\, including a Sloan Fellowship (2010)\, Office of Nav al Research Young Investigator Award (2011)\, Presidential Early Career Aw ard for Scientists and Engineers (2011)\, Blavatnik Faculty Award (2013)\, and ACM-Infosys Foundation Award (2013).

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-5645@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Algebraic and Geometric ideas in the theory of Linear Optimizat ion”\nAbstract: Linear optimization is undeniably a central tool of applie d mathematics with applications in a wide\nrange of topics\, from statisti cal regression to image processing. The theory of linear optimization has many\nbeautiful geometric and algebraic topics and it is still a source o f many fascinating mathematical open problems.\n \nIn this talk I will pre sent several advances from the past 10 years in the theory of linear opt imization.\nThese results include new results on the complexity of the sim plex method\, the structure of central\npaths of interior point methods\, and about the geometry of some less well-known iterative techniques.\nOne interesting feature of these new theorems is that they connect this very applied algorithmic field with\nseemingly far away “pure” topics like alg ebraic geometry\, differential geometry\, and combinatorial topology.\n \n This panoramic talk is geared for students and the non-expert faculty mem ber. I will summarize work by many\nauthors\, including results that are m y own joint work with subsets of the following people A. Basu\, J. Haddock \,\n\nJunod\, S. Klee\, B. Sturmfels\, and C. Vinzant. DTSTART;TZID=America/New_York:20151119T133000 DTEND;TZID=America/New_York:20151119T143000 SEQUENCE:0 SUMMARY:Seminar: Jesus De Loera (University of California\, Davis) @ Whiteh ead 304 URL:https://engineering.jhu.edu/ams/events/seminar-jesus-de-loera-universit y-of-california-davis/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Algebraic and Geometric ideas in the theory of Linear Optimization”

\nAbstract: L inear optimization is undeniably a central tool of applied mathematics wit h applications in a wide

\nrange of topics\, from statistical regres sion to image processing. The theory of linear optimization has many

\nbeautiful geometric and algebraic topics and it is still a source of many fascinating mathematical open problems.

\n\n

In this tal k I will present several advances from the past 10 years in the theory of linear optimization.

\nThese results include new results on the co mplexity of the simplex method\, the structure of central

\npaths o f interior point methods\, and about the geometry of some less well-known iterative techniques.

\nOne interesting feature of these new theorem s is that they connect this very applied algorithmic field with

\nse emingly far away “pure” topics like algebraic geometry\, differential geo metry\, and combinatorial topology.

\n\n

This panoramic talk is geared for students and the non-expert faculty member. I will summariz e work by many

\nauthors\, including results that are my own joint w ork with subsets of the following people A. Basu\, J. Haddock\,

\n-
\n
- Junod\, S. Klee\, B. Sturmfels\, and C. Vinzant. \n

~~Comparative Effectiveness Research of Environmental Exposures: Conne
cting the Dots with Big Data~~

\n

Comparative effect iveness research increasingly depends on the analysis of a rapidly expandi ng universe of observational data made possible by the growing integration of administrative claims data (e.g. Medicare or SEER-Medicare claims) wit h environmental health exposures (e.g. emissions from power plants\, air p ollution for monitoring stations)\, with survey and census data (e.g. popu lation demographics).

\nWe are interested in addressing questions th at attempt to connect the dots between environmental exposures and human h ealth\, such as: Can increased noise levels near airports cause higher rat es of cardiovascular disease or stroke? Do even moderate increases in air pollution from sources such as automobiles and industrial smokestacks have a measurable effect on a community’s death rate? What are the most likely causes of hospitalizations during heat waves?

\nDevelopment of stat istical methods is needed to be able to handle large\, messy data sets\, i ntegrate them\, and extract meaningful conclusions. In this talk we will r eview some of these tatistical methods aimed at making causal inferences o n the effectiveness of environmental interventions with such large observa tional data structures.

\n\n

**Biography- Dr. Francesca
Dominici\, PhD**

\n

**Academic Career**

Francesca Dominici is a Professor in the Department of Biostatis tics at the Harvard School of Public Health and the Senior Associate Dean for Research. Dr. Dominici received her Ph.D. in Statistics from the Unive rsity of Padua\, Italy in 1997. During her PhD\, she spent two years as a visiting PhD student at Duke University\, NC\, USA. In 1997 she went to t he Bloomberg School of Public Health as a post-doctoral fellow. In 1999 sh e was appointed Assistant Professor at the Bloomberg School of Public Heal th and in 2007 she was promoted to Full Professor with Tenure. In 2009 she moved to Harvard School of Public Health as a tenured Professor of Biosta tistics\, was appointed Associate Dean of Information Technology in 2010\, and Senior Associate Dean for Research in 2013.

\nDr. Dominici’s re search has focused on the development of statistical methods for the analy sis of large observational data with the ultimate goal of addressing impor tant questions in environmental health science\, health related impacts of climate change\, and comparative effectiveness research. She is an expert in Bayesian methods\, longitudinal data analysis\, confounding adjustment \, causal inference\, and Bayesian hierarchical models. She has extensive experience on the development of statistical methods and their applicatio ns to environmental epidemiology\, implementation science and health polic y\, outcome research and patient safety\, and comparative effectiveness re search.

\n\n

**Research**

Dr. Dominici h
as authored more than 120 peer-reviewed publications. She is the PI\, toge
ther with Dr. Xihong Lin\, of a NCI P01 project entitled “**Statisti
cal Informatics for Cancer Research**” (http://www.hsph.harvard.edu/statinfo
rmatics/index.html). She is the PI of a Project called “**A Nat
ional Study to Assess Susceptibility\, Vulnerability and Effect Modificati
on of Air Pollution Health Risks**” as part of the Harvard EPA Cent
er entitled **“Air Pollution Mixtures: Health Effects Across Life St
ages**” (PI: Dr. Koutrakis) She is also the PI of several EPA/NIH/H
EI funded projects aimed at developing statistical methods and conducting
nation-wide epidemiological studies on the health effects of air pollution
. Most recently\, she has become more involved in comparative effectivenes
s research collaborating with investigators at Dana Farber Cancer Institut
e. With her colleagues she is developing statistical methods for causal in
ference and propensity score matching to compare health care delivery syst
ems in end of life cancer\, with a special focus on glioblastoma and pancr
eatic cancer. Dr. Dominici also oversees the management and the analysis o
f several administrative databases\, including Part A CMS files and SEER-M
edicare\, which are linked to air pollution and weather and socioeconomic
data.

\n

**Education and Mentoring**

D r. Dominici is teaching the course Bio249 entitled “Bayesian Methodology i n Biostatistics” at HSPH. Previously she taught Analysis of Longitudinal D ata\, and Multilevel Statistical Models while a faculty member at Johns Ho pkins University. She has been the primary advisory of 9 PhD students and 13 post-doctoral fellows. She is a passionate mentor of junior faculty.

\n\n

**Diversity**

Dr. Dominici is committ
ed to diversity. Together with Dr. Linda P. Fried (now Dean of the Mailman
School of Public Health at Columbia University)\, she has co-chaired the
University Committee of the Status of Women at Johns Hopkins University. F
rom this experience she wrote a paper entitled “So Few Women Leaders” *
Academe*\, *July-August 2009*” (http://www.aaup.org/article/so-fe
w-women-leaders – .Ubx4SZWQma4). In 2009\, she was awarded the Diversi
ty Recognition Award by the President of Johns Hopkins University. Recentl
y\, she has been giving lectures and moderated panel discussions on work-f
amily balance across Harvard (see http://news.harvard.edu/gazette/
story/2012/11/having-it-all-at-harvard/). Currently she is the chair (
with Dr. Burleigh) of the University Committee for the Advancement of Wome
n Faculty at HSPH.

\n

**Administration**

In her role as Associate Dean of Information Technology\, Dr. Dominici has led new initiatives at HSPH regarding research computing. More specif ically\, she led a MOU between our school and research computing (RC) faci lity at the Faculty of Arts and Science (FASRC) http://rc.fas.harvard.edu/ enabling HSPH faculty to access the FAS computing facilities. HSPH faculty are treated equally to FAS facu lty in terms of priority of access\, ticket turn-around\, and access to sh ared facilities and shared licenses. See http://rc.fas.harvard.edu/hsph-overview/ for details .

\n\n

**Service**

Dr. Dominici has serv ed on a number of National Academies’ committees\, including the Committee on Research Direction in Human Biological Effects of Low Level Ionizing R adiation\; the Committee on Gulf War and Health: Review of the Medical Lit erature Relative to Gulf War Veterans’ Health\; the Committee to Review th e Federal Response to the Health Effects Associated with the Gulf of Mexic o Oil Spill\; the Committee on Secondhand Smoke Exposure and Acute Coronar y Events\; the Committee to Review ATSDR’s Great Lakes Report\; the Commit tee on Making Best Use of the Agent Orange Exposure Reconstruction Model\; the Committee on Gulf War and Health\; the Committee to Assess Potential Health Effects from Exposures to PAVE PAWS Low-Level Phased-array Radiofre quency Energy\; and the Committee on the Utility of Proximity-Based Herbic ide Exposure Assessment in Epidemiologic Studies of Vietnam Veterans.

\nDr. Dominici has received numerous recognitions\, including the Flore nce Nightingale David award\, sponsored jointly by the Committee of Presid ents of Statistical Societies and Caucus for Women in Statistics 2015\; Ma thematics for Planet Earth Award Lecture\, hosted by the Statistical and A pplied Mathematical Sciences Institute (SAMSI) 2013\; Diversity Recognitio n Award\, Johns Hopkins University\, 2009\; Myrto Lefkopoulou Distinguishe d Lectureship Award\, Department of Biostatistics\, Harvard School of Publ ic Health\, 2007\; Gertrude Cox Award\, Washington DC Chapter of the Ameri can Statistical Association and RTI International\, 2007\; Mortimer Spiege lman Award\, Statistics Section of the American

\nPublic Health Asso ciation\, 2006\; Dean’s Lecture\, Bloomberg School of Public Health\, 2007 \; and an Invitation to Address the Royal Statistical Society\, London\, U K\, 2002.

\nShe is a member of numerous professional societies\, inc luding the American Statistical Association\, the International Biometric Society\, and the International Society for Environmental Epidemiology. Sh e is the Senior Editor of Chapman & Hall/CRC Texts in Statistical Science Series and Associate Editor of the Journal of the Royal Statistical Societ y.

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6099@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title: Feature allocations\, probability functions\, and paintb oxes\nAbstract:\nClustering involves placing entities into mutually exclus ive categories. We wish to relax the requirement of mutual exclusivity\, a llowing objects to belong simultaneously to multiple classes\, a formulati on that we refer to as “feature allocation.” The first step is a theoretic al one. In the case of clustering the class of probability distributions o ver exchangeable partitions of a dataset has been characterized (via excha ngeable partition probability functions and the Kingman paintbox). These c haracterizations support an elegant nonparametric Bayesian framework for c lustering in which the number of clusters is not assumed to be known a pri ori. We establish an analogous characterization for feature allocation\; w e define notions of “exchangeable feature probability functions” and “feat ure paintboxes” that lead to a Bayesian framework that does not require th e number of features to be fixed a priori. The second step is a computatio nal one. Rather than appealing to Markov chain Monte Carlo for Bayesian in ference\, we develop a method to transform Bayesian methods for feature al location (and other latent structure problems) into optimization problems with objective functions analogous to K-means in the clustering setting. T hese yield approximations to Bayesian inference that are scalable to large inference problems. DTSTART;TZID=America/New_York:20160128T133000 DTEND;TZID=America/New_York:20160128T143000 SEQUENCE:0 SUMMARY:Seminar: Tamara Broderick (MIT) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-tamara-broderick-mit-whi tehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Title: Featur e allocations\, probability functions\, and paintboxes

\nAbstract:\n

Clustering involves placing entities into mutually exclusive catego ries. We wish to relax the requirement of mutual exclusivity\, allowing ob jects to belong simultaneously to multiple classes\, a formulation that we refer to as “feature allocation.” The first step is a theoretical one. In the case of clustering the class of probability distributions over exchan geable partitions of a dataset has been characterized (via exchangeable pa rtition probability functions and the Kingman paintbox). These characteriz ations support an elegant nonparametric Bayesian framework for clustering in which the number of clusters is not assumed to be known a priori. We es tablish an analogous characterization for feature allocation\; we define n otions of “exchangeable feature probability functions” and “feature paintb oxes” that lead to a Bayesian framework that does not require the number o f features to be fixed a priori. The second step is a computational one. R ather than appealing to Markov chain Monte Carlo for Bayesian inference\, we develop a method to transform Bayesian methods for feature allocation ( and other latent structure problems) into optimization problems with objec tive functions analogous to K-means in the clustering setting. These yield approximations to Bayesian inference that are scalable to large inference problems.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6197@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title : On the spectra of direct sums and Kronecker products of side length 2 hypermatrices and related algorithmic problems in data scie nce.\nAbstract:\nWe present elementary method for obtaining the spectral d ecomposition of hypermatrices generated by arbitrary combinations of Krone cker products and direct sums of cubic hypermatrices having side length 2. The method is based on a generalization of Parseval’s identity. We use t he general formulation of Parseval’s identity to introduce hypermatrix Fou rier transforms and discrete Fourier hypermatrices. We extend to hypermatr ices orthogonalization procedures and Sylvester’s classical Hadamard matri x construction. We conclude the talk with illustrations of spectral decomp ositions of adjacency hypermatrices of finite groups and a proof of a hype rmatrix Rayleigh quotient inequality.\nThis is a joint work with Yuval Fil mus. DTSTART;TZID=America/New_York:20160202T133000 DTEND;TZID=America/New_York:20160202T143000 SEQUENCE:0 SUMMARY:Seminar: Edinah Gnang (Purdue University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-edinah-gnang-purdue-univ ersity-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle : On th e spectra of direct sums and Kronecker products of side length 2 hypermatr ices and related algorithmic problems in data science.

\nAbstract:\n

We present elementary method for obtaining the spectral decompositi on of hypermatrices generated by arbitrary combinations of Kronecker produ cts and direct sums of cubic hypermatrices having side length 2. The metho d is based on a generalization of Parseval’s identity. We use the general formulation of Parseval’s identity to introduce hypermatrix Fourier trans forms and discrete Fourier hypermatrices. We extend to hypermatrices ortho gonalization procedures and Sylvester’s classical Hadamard matrix construc tion. We conclude the talk with illustrations of spectral decompositions o f adjacency hypermatrices of finite groups and a proof of a hypermatrix Ra yleigh quotient inequality.

\nThis is a joint work with Yuval Filmus .

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6098@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title: Using Integer Programming for Solving Nonconvex Quadrati c Programs with Box Constraints\n \nWe discuss effective computational tec hniques for solving nonconvex quadratic programs with box constraints (Box QP). Cutting planes obtained from the well-known Boolean Quadric Polytope may be applied in this context\, and we demonstrate the equivalence betwe en the Chvatal-Gomory closure of a natural linear relaxation of (BoxQP) an d the relaxation of the Boolean Quadric Polytope consisting of the odd-cyc le inequalities. By using these cutting planes effectively at nodes of th e branch-and-bound tree\, in conjunction with additional integrality-based branching and a strengthened convex quadratic relaxation\, we demonstrate that we can effectively solve a well-known family of test instances. Our new solver\, GuBoLi\, is orders of magnitude faster than existing commerc ial and open-source solvers. DTSTART;TZID=America/New_York:20160204T133000 DTEND;TZID=America/New_York:20160204T143000 SEQUENCE:0 SUMMARY:Seminar: Jeff Linderoth (University of Wisconsin- Madison) @ Whiteh ead 304 URL:https://engineering.jhu.edu/ams/events/seminar-jeff-linderoth-universit y-of-wisconsin-madison-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Using Integer Programming for Solving Nonconvex Quadratic Programs with Box Cons traints

\n\n

We discuss effective computational techniques fo r solving nonconvex quadratic programs with box constraints (BoxQP). Cutt ing planes obtained from the well-known Boolean Quadric Polytope may be ap plied in this context\, and we demonstrate the equivalence between the Chv atal-Gomory closure of a natural linear relaxation of (BoxQP) and the rela xation of the Boolean Quadric Polytope consisting of the odd-cycle inequal ities. By using these cutting planes effectively at nodes of the branch-a nd-bound tree\, in conjunction with additional integrality-based branching and a strengthened convex quadratic relaxation\, we demonstrate that we c an effectively solve a well-known family of test instances. Our new solve r\, GuBoLi\, is orders of magnitude faster than existing commercial and op en-source solvers.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6185@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title: Important Features PCA (IF-PCA) for Large-Scale Inferenc e\, with Applications in Gene Microarrays\n\n\nAbstract:\nIdentification o f sample labels is a major problem in statistics with many applications. I n the Big Data era\, it faces two main challenges: 1. the number of featur es is much larger than the sample size\; 2. the signals are sparse and wea k\, masked by large amount of noise.\n\n\nWe propose a new tuning-free clu stering procedure for high-dimensional data\, Important Features PCA (IF-P CA). IF-PCA consists of a feature selection step\, a PCA step\, and a k-me ans step. The first two steps reduce the data dimensions recursively\, whi le the main information is preserved. As a consequence\, IF-PCA is fast an d accurate\, producing competitive performance in application to 10 gene m icroarray data sets.\n\n\nWe also generalize IF-PCA for the signal recover y and hypothesis testing problems. With IF-PCA and two aggregation methods \, we find the statistical limits for these three problems. DTSTART;TZID=America/New_York:20160209T133000 DTEND;TZID=America/New_York:20160209T143000 SEQUENCE:0 SUMMARY:Seminar: Wanjie Wang (University of Pennsylvania) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-wanjie-wang-university-o f-pennsylvania/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Impo
rtant Features PCA (IF-PCA) for Large-Scale Inference\, with Applications
in Gene Microarrays

\n\n\nAbstract:

\nIdentification of sample labels is a major problem in statistics wi
th many applications. In the Big Data era\, it faces two main challenges:
1. the number of features is much larger than the sample size\; 2. the sig
nals are sparse and weak\, masked by large amount of noise.

\n\n\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n*\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-6191@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Title: Recent theoretic and algorithmic advances in graph matc
hing\nAbstract: Inference across multiple graphs arises naturally in disc
iplines as varied as neuroscience\, physics\, and sociology. In a number
of methodologies for joint inference across graphs\, however\, it is assum
ed that an explicit vertex correspondence is a priori known across the ver
tex sets of the graphs. While this assumption is often reasonable\, in pra
ctice these correspondences may be unobserved and/or errorfully observed\,
and graph matching—aligning a pair of graphs to minimize their edge disag
reements—is used to align the graphs before performing subsequent inferenc
e. Graph matching is a computationally challenging and well-studied probl
em\, but few existing algorithms have theoretical support for their perfor
mance. For tractability\, many algorithms begin by relaxing the problem’s
binary constraints\, thus rendering applicable gradient-descent methodolo
gies. We develop a state-of-the-art algorithm for solving an indefinite re
laxed graph matching problem\, and we show that under mild model assumptio
ns\, our indefinite relaxation (when solved exactly) almost always uncover
s the optimal permutation\, while the commonly used convex relaxation almo
st always fails to identify the optimal permutation. We highlight some of
the practical and theoretical implications of these results on real and s
ynthetic data\, and we discuss recent work towards formalizing the connect
ion between graph matching and pairwise mutual information.
DTSTART;TZID=America/New_York:20160225T133000
DTEND;TZID=America/New_York:20160225T143000
SEQUENCE:0
SUMMARY:Seminar: Vince Lyzinski (JHU) @ Whitehead 304
URL:https://engineering.jhu.edu/ams/events/seminar-vince-lyzinski-jhu/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n* \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n
\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-7296@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:http://www.math.jhu.edu/~data/
DESCRIPTION:Modeling the dynamics of interacting particles by means of stoc
hastic networks\n\nMaterial science have been rapidly developing in recent
years. A variety of particles interacting according to different kinds of
pair potentials has been produced in experimental works. Looking into the
future\, one can imagine controlled self-assembly of particles into clust
ers of desired structures leading to the creation of new types of material
s. Analytical studies of the self-assembly involve coping with difficultie
s associated with the huge numbers configurations\, high dimensionality\,
complex geometry\, and unacceptably large CPU times. A feasible approach t
o the study of self-assembly consists of mapping the collections of cluste
rs onto stochastic networks (continuous-time Markov chains) and analyzing
their dynamics. Vertices of the networks represent local minima of the pot
ential energy of the clusters\, while arcs connect only those pairs of ver
tices that correspond to local minima between which direct transitions are
physically possible. Transition rates along the arcs are the transition r
ates between the corresponding pairs of local minima. Such networks are ma
thematically tractable and\, at the same time\, preserve important feature
s of the underlying dynamics. Nevertheless\, their huge size and complexit
y render their analysis challenging and invoke the development of new math
ematical techniques. I will discuss some approaches to construction and an
alysis of such networks.
DTSTART;TZID=America/New_York:20161012T150000
DTEND;TZID=America/New_York:20161012T160000
SEQUENCE:0
SUMMARY:Data Seminar: Maria Cameron (University of Maryland College Park) @
Krieger 309
URL:https://engineering.jhu.edu/ams/events/data-seminar-maria-cameron-unive
rsity-maryland-college-park-shaffer-100/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n \\n\\n \\n\\n \\n\\n \\n\\n**\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-6965@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:An Introduction to Distance Preserving Projections of Smooth Ma
nifolds\n \nManifold-based image models are assumed in many engineering ap
plications involving imaging and image classification. In the setting of
image classification\, in particular\, proposed designs for small and chea
p cameras motivate compressive imaging applications involving manifolds.
Interesting mathematics results when one considers that the problem one ne
eds to solve in this setting ultimately involves questions concerning how
well one can embed a low-dimensional smooth sub-manifold of high-dimension
al Euclidean space into a much lower dimensional space without knowing any
of its detailed structure. We will motivate this problem and discuss how
one might accomplish this seemingly difficult task using random projectio
ns. Little if any prerequisites will be assumed beyond linear algebra and
some probability.
DTSTART;TZID=America/New_York:20161103T133000
DTEND;TZID=America/New_York:20161103T143000
SEQUENCE:0
SUMMARY:Seminar: Mark Iwen (Michigan State University) @ Whitehead 304
URL:https://engineering.jhu.edu/ams/events/seminar-mark-iwen-michigan-state
-university-whitehead-304/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n** \\n\\n \\n\\n \\n\\n\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-6997@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Slipping Through the Cracks: Detecting Manipulation in Regional
Commodity Markets\n \nReid B. Stevens[1] and Jeffery Y. Zhang[2]\n \nBetw
een 2010 and 2014\, the regional price of aluminum in the United States (M
idwest premium) increased 400 percent. We argue that the Midwest premium w
as likely manipulated during this period through the exercise of market po
wer in the aluminum storage market. We first use a difference-in-differenc
es model to show that there was a statistically significant increase of $0
.07 per pound in the regional price of aluminum relative to the regional p
rice of a production complement\, copper. We then use several instrumenta
l variables to show that this increase was driven by a single financial co
mpany’s accumulation of an unprecedented level of aluminum inventories in
Detroit. Since this scheme targeted the regional price of aluminum\, regu
lators who monitored only spot and futures prices would not have noticed a
nything peculiar. We therefore present an algorithm for real-time detectio
n of similar manipulation schemes in regional commodity markets. The algo
rithm confirms the existence of a structural break in the U.S. aluminum ma
rket in late 2011. Using the algorithm\, regulators could have detected th
e scheme as early as December 2012\, more than six months before it was pu
blicized by an article in The New York Times. We also apply the algorithm
to another suspected case of regional price manipulation in the European a
luminum market and find a similar break in 2011\, suggesting the scheme ma
y have been implemented beyond the United States.\n[1] Department of Agric
ulture Economics\, Texas A&M University\, stevens@tamu.edu\n[2] Department
of Economics\, Yale University and Harvard Law School\, jeffery.zhang@yal
e.edu
DTSTART;TZID=America/New_York:20161117T133000
DTEND;TZID=America/New_York:20161117T143000
SEQUENCE:0
SUMMARY:Seminar: Reid Stevens (Texas A&M University) @ Whitehead 304
URL:https://engineering.jhu.edu/ams/events/seminar-helyette-geman-jhu-white
head-304/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n\n \\n\\n \\n\\n \\n\\n \\n\\n\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-9286@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:http://www.math.jhu.edu/~data/
DESCRIPTION:
DTSTART;TZID=America/New_York:20170308T150000
DTEND;TZID=America/New_York:20170308T160000
SEQUENCE:0
SUMMARY:Data Seminar: Matthew Hirn (Michigan State University) @ Whitehead
304
URL:https://engineering.jhu.edu/ams/events/data-seminar-matthew-hirn-michig
an-state-university-whitehead-304/
X-COST-TYPE:free
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-9262@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Title: Mean Field Games: theory and applications\n \nAbstract:
We review the Mean Field Game paradigm introduced independently by Caines
-Huang-Malhame and Lasry-Lyons ten years ago\, and we illustrate their rel
evance to applications with a few practical of examples (bird flocking\, r
oom exit\, systemic risk\, cyber-security\, …. ). We then review the proba
bilistic approach based on Forward-Backward Stochastic Differential Equati
ons\, and we derive the Master Equation from a version of the chain rule
(Ito’s formula) for functions over flows of probability measures. Finally\
, motivated by the literature on economic models of bank runs\, we introdu
ce mean field games of timing and discuss new results\, and some of the ma
ny remaining challenges.
DTSTART;TZID=America/New_York:20170309T133000
DTEND;TZID=America/New_York:20170309T143000
SEQUENCE:0
SUMMARY:Duncan Lecture Series: Rene Carmona (Princeton) @ Krieger 205
URL:https://engineering.jhu.edu/ams/events/duncan-lecture-series-rene-carmo
na-princeton-tba/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-9302@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:
DTSTART;TZID=America/New_York:20170412T150000
DTEND;TZID=America/New_York:20170412T160000
SEQUENCE:0
SUMMARY:Data Seminar: Andrew Christlieb (Michigan State University) @ White
head 304
URL:https://engineering.jhu.edu/ams/events/data-seminar-andrew-christlieb-m
ichigan-state-university-whitehead-304/
X-COST-TYPE:free
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-9366@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Reciprocal Graphical Models for Integrative Gene Regulatory Net
work Analysis\nConstructing gene regulatory networks is a fundamental task
in systems biology. We introduce a Gaussian reciprocal graphical model fo
r inference about gene regulatory relationships by integrating mRNA gene e
xpression and DNA level information including copy number and methylation.
Data integration allows for inference on the directionality of certain re
gulatory relationships\, which would be otherwise indistinguishable due to
Markov equivalence. Efficient inference is developed based on simultaneou
s equation models. Bayesian model selection techniques are adopted to esti
mate the graph structure. We illustrate our approach by simulations and tw
o applications in ZODIAC pairwise gene interaction analysis and colon aden
ocarcinoma pathway analysis.
DTSTART;TZID=America/New_York:20170413T133000
DTEND;TZID=America/New_York:20170413T143000
SEQUENCE:0
SUMMARY:Seminar: Peter Mueller (University of Texas) @ Whitehead 304
URL:https://engineering.jhu.edu/ams/events/seminar-peter-mueller-university
-texas-whitehead-304/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n \\n\\n \\n\\n**\n** \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n**
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-10417@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Title: Frames — two case studies: ambiguity and uncertainty\nAb
stract: The theory of frames is an essential concept for dealing with sign
al representation in noisy environments. We shall examine the theory in th
e settings of the narrow band ambiguity function and of quantum informatio
n theory. For the ambiguity function\, best possible estimates are derived
for applicable constant amplitude zero autocorrelation (CAZAC) sequences
using Weil’s solution of the Riemann hypothesis for finite fields. In exte
nding the theory to the vector-valued case modelling multi-sensor environm
ents\, the definition of the ambiguity function is characterized by means
of group frames. For the uncertainty principle\, Andrew Gleason’s measure
theoretic theorem\, establishing the transition from the lattice interpret
ation of quantum mechanics to Born’s probabilistic interpretation\, is gen
eralized in terms of frames to deal with uncertainty principle inequalitie
s beyond Heisenberg’s. My collaborators are Travis Andrews\, Robert Benede
tto\, Jeffrey Donatelli\, Paul Koprowski\, and Joseph Woodworth. \nRelated
papers:\nSuper-resolution by means of Beurling minimal extrapolation\nGen
eralized Fourier frames in terms of balayage\nUncertainty principles and w
eighted norm inequalities\nA frame reconstruction algorithm with applicati
ons to magnetric resonance imaging\nFrame multiplication theory and a vect
or-valued DFT and ambiguity functions
DTSTART;TZID=America/New_York:20170927T150000
DTEND;TZID=America/New_York:20170927T160000
SEQUENCE:0
SUMMARY:Data Science Seminar: John Benedetto (University of Maryland Colleg
e Park and Norbert Wiener Center) @ Hodson 203
URL:https://engineering.jhu.edu/ams/events/data-science-seminar-john-benede
tto-university-maryland-college-park-norbert-wiener-center-hodson-203/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n** \\n\\n**\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-10477@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Title: An improved approach to calibrating misspecified mathem
atical models\nAbstract: We consider the problem of calibrating misspecifi
ed mathematical models using experimental data. To compensate for the miss
pecification of the model\, a discrepancy function is usually included and
modeled via a Gaussian stochastic process (GaSP)\, leading to better resu
lts of prediction. The calibration parameters in the model\, however\, som
etimes become unidentifiable and the calibrated model fits the experimenta
l data poorly as a consequence. In this work\, we propose the scaled Gauss
ian stochastic process (S-GaSP)\, a novel stochastic process for calibrati
on and prediction. This new approach bridges the gap between two predomina
nt methods\, namely the $L_2$ calibration and GaSP calibration. A computat
ionally feasible approach is introduced for this new model under the Bayes
ian paradigm. The S-GaSP model not only provides a general framework for c
alibration\, but also enables the calibrated mathematical model to predict
well regardless of the discrepancy function. Simulation examples are prov
ided and real examples using satellite images to calibrate the model for v
olcanic hazard are studied to illustrate the connections and differences b
etween this new model and other previous approaches. \n
DTSTART;TZID=America/New_York:20170928T133000
DTEND;TZID=America/New_York:20170928T143000
SEQUENCE:0
SUMMARY:AMS Seminar: Mengyang Gu (JHU) @ Whitehead 304
URL:https://engineering.jhu.edu/ams/events/ams-seminar-mengyang-gu-jhu-whit
ehead-304/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n** \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n~~\n~~ \\n\\n \\n\\n \\n\\n**\n** \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n**\n** \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n**Abstract:**\n \\n\\n \\n\\n**\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-10874@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Title: Random Walks on Secondary Structure and the Folding of R
NA\nAbstract:\nWhat is the correct characterization of the native structur
e of a protein or a non-coding RNA molecule? Is it the minimum energy stat
e\, a metastable state\, or a sample from thermal equilibrium? The problem
is unsettled and a topic of enduring debate. The “agnostic” approach is t
hrough molecular dynamics—make a proper accounting of the intramolecular f
orces and interactions with the surrounding liquid (mostly water)\, write
down the corresponding kinetic equation (e.g. a Langevin equation)\, and s
imulate folding. But this usually isn’t practical\, and approximations nee
d to be made. I will explore an approximation in which the bulk of the fol
ding process is replaced by a random walk\, with discrete moves from one s
econdary structure to another. An analytic result identifies conditions fo
r guaranteed accuracy. Simulation results\, running the approximation agai
nst an optimized integrator of the Langevin equation\, achieve the expecte
d accuracy\, but 100 times faster.
DTSTART;TZID=America/New_York:20180330T133000
DTEND;TZID=America/New_York:20180330T143000
SEQUENCE:0
SUMMARY:Duncan Lecture Series AMS Seminar: Stuart Geman (Brown University)
@ Shaffer 100
URL:https://engineering.jhu.edu/ams/events/ams-seminar-stuart-geman-brown-u
niversity-whitehead-304-2/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n** \\n\\n \\n\\n**\n** \\n\\n \\n\\n**\n\n**\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-11249@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:Date: Monday\, April 9th\, 2018\nTime: 7:00 pm: Enjoy refreshme
nts and snacks with students and faculty.\n7:30 pm: The Mathemagics perfor
mance begins!\nLocation: Johns Hopkins University\, Homewood Campus\; Hods
on Hall\, Room 110\nRSVP: Please fill out the form at: https://tinyurl.com
/artbenjhusam\nOr E-mail us at husam.jhu@gmail.com\n\n\n
DTSTART;TZID=America/New_York:20180409T190000
DTEND;TZID=America/New_York:20180409T210000
SEQUENCE:0
SUMMARY:HUSAM Presents: Mathemagics Night with Dr. Arthur Benjamin @Hodson
Hall rm 110
URL:https://engineering.jhu.edu/ams/events/husam-presents-mathemagics-night
-with-dr-arthur-benjamin-hodson-hall-rm-110/
X-COST-TYPE:free
X-WP-IMAGES-URL:thumbnail\;https://engineering.jhu.edu/ams/wp-content/uploa
ds/sites/44/2018/04/ArtBenjaminposter-230x300.jpg\;503\;656\,medium\;https
://engineering.jhu.edu/ams/wp-content/uploads/sites/44/2018/04/ArtBenjamin
poster-230x300.jpg\;503\;656\,large\;https://engineering.jhu.edu/ams/wp-co
ntent/uploads/sites/44/2018/04/ArtBenjaminposter-230x300.jpg\;503\;656\,fu
ll\;https://engineering.jhu.edu/ams/wp-content/uploads/sites/44/2018/04/Ar
tBenjaminposter-230x300.jpg\;503\;656
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n \\n\\n**\n<
p>**

\n\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n \\n\\n

We propose a new tuning-free clustering procedure
for high-dimensional data\, Important Features PCA (IF-PCA). IF-PCA consis
ts of a feature selection step\, a PCA step\, and a k-means step. The firs
t two steps reduce the data dimensions recursively\, while the main inform
ation is preserved. As a consequence\, IF-PCA is fast and accurate\, produ
cing competitive performance in application to 10 gene microarray data set
s.

\n\n\nWe also generalize IF-PCA for the
signal recovery and hypothesis testing problems. With IF-PCA and two aggr
egation methods\, we find the statistical limits for these three problems.

\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-6344@engineering.jhu.edu/ams
DTSTAMP:20180715T165110Z
CATEGORIES:
CONTACT:
DESCRIPTION:Stochastic evolutionary modeling of cancer development and resi
stance to treatment\nCancer is the result of a stochastic evolutionary pro
cess characterized by the accumulation of mutations that are responsible f
or tumor growth\, immune escape\, and drug resistance\, as well as mutatio
ns with no effect on the phenotype. Stochastic modeling can be used to des
cribe the dynamics of tumor cell populations and obtain insights into the
hidden evolutionary processes leading to cancer. I will present recent app
roaches that use branching process models of cancer evolution to quantify
intra-tumor heterogeneity and the development of drug resistance\, and the
ir implications for interpretation of cancer sequencing data and the desig
n of optimal treatment strategies.
DTSTART;TZID=America/New_York:20160210T133000
DTEND;TZID=America/New_York:20160210T143000
SEQUENCE:0
SUMMARY:Seminar: Ivana Bozic (Harvard University)
URL:https://engineering.jhu.edu/ams/events/seminar-ivana-bozic-harvard-univ
ersity/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nStochastic ev olutionary modeling of cancer development and resistance to treatment

\nCancer is the result of a stochastic evolutionary process characteriz ed by the accumulation of mutations that are responsible for tumor growth\ , immune escape\, and drug resistance\, as well as mutations with no effec t on the phenotype. Stochastic modeling can be used to describe the dynami cs of tumor cell populations and obtain insights into the hidden evolution ary processes leading to cancer. I will present recent approaches that use branching process models of cancer evolution to quantify intra-tumor hete rogeneity and the development of drug resistance\, and their implications for interpretation of cancer sequencing data and the design of optimal tre atment strategies.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6188@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title: Universality in numerical computations with random data \n \nAbstract: This talk will concern recent progress on the statistical analysis of numerical algorithms with random initial data. In particular\ , with appropriate randomness\, the fluctuations of the iteration count (h alting time) of numerous numerical algorithms have been demonstrated to be universal\, i.e.\, independent of the distribution on the initial data. T his phenomenon has given new insights into random matrix theory. Furthermo re\, estimates from random matrix theory allow for fluctuation limit theor ems for simple algorithms and halting time estimates for others. The univ ersality in the halting time is directly related to the experimental work of Bakhtin and Correll on neural computation and human decision-making tim es. DTSTART;TZID=America/New_York:20160211T133000 DTEND;TZID=America/New_York:20160211T143000 SEQUENCE:0 SUMMARY:Seminar: Tom Trogdon (New York University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-tom-trogdon-new-york-uni versity-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Univer sality in numerical computations with random data

\n\n

Abstra ct: This talk will concern recent progress on the statistical analysis of numerical algorithms with random initial data. In particular\, with appr opriate randomness\, the fluctuations of the iteration count (halting time ) of numerous numerical algorithms have been demonstrated to be universal\ , i.e.\, independent of the distribution on the initial data. This phenome non has given new insights into random matrix theory. Furthermore\, estima tes from random matrix theory allow for fluctuation limit theorems for sim ple algorithms and halting time estimates for others. The universality in the halting time is directly related to the experimental work of Bakhtin and Correll on neural computation and human decision-making times.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6348@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Bringing Moneyball to Campaigns\n \nOver the past decade\, an e ntire industry has grown up around the use of data to help campaigns be mo re efficient and effective. Whether it is trying to identify that last pe rsuadable voter or allocating resources to get your supporters out to the polls\, today’s campaigns often rely on a staff of data analysts\, statist icians and modelers. Together\, data and analytics help identify which vo ters to target and what actions to take to generate the votes where they a re needed.\nIn this talk I will introduce the tools and techniques involvi ng data\, analytics\, and experimentation used by campaigns. We will disc uss where many of these techniques came from and how they evolved in polit ics to culminate in President Obama’s 2012 campaign. This survey of the d ata-driven campaigns will include polling\, micro-targeting and random con trolled experiments.\n \nPlease join HUSAM in welcoming Dr. Ben Yuhas to t he Johns Hopkins University community! DTSTART;TZID=America/New_York:20160211T190000 DTEND;TZID=America/New_York:20160211T200000 SEQUENCE:0 SUMMARY:HUSAM: Dr. Ben Yuhas (Principal of the Yuhas Consulting Group\, LLC ) @ Gilman 50 URL:https://engineering.jhu.edu/ams/events/husam-dr-ben-yuhas-principal-of- the-yuhas-consulting-group-llc-gilman-50/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nBringing Mone yball to Campaigns

\n\n

Over the past decade\, an entire indu stry has grown up around the use of data to help campaigns be more efficie nt and effective. Whether it is trying to identify that last persuadable voter or allocating resources to get your supporters out to the polls\, to day’s campaigns often rely on a staff of data analysts\, statisticians and modelers. Together\, data and analytics help identify which voters to ta rget and what actions to take to generate the votes where they are needed.

\nIn this talk I will introduce the tools and techniques involving data\, analytics\, and experimentation used by campaigns. We will discuss where many of these techniques came from and how they evolved in politics to culminate in President Obama’s 2012 campaign. This survey of the data -driven campaigns will include polling\, micro-targeting and random contro lled experiments.

\n\n

Please join HUSAM in welcoming Dr. Ben Yuhas to the Johns Hopkins University community!

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6198@engineering.jhu.edu/ams DTSTAMP:20180715T165110Z CATEGORIES: CONTACT: DESCRIPTION:Title: Controlling a Thermal Fluid: Theoretical and Computation al Issues\nAbstract: We first discuss the problem of designing a feedbac k law which locally stabilizes a two dimensional thermal fluid modeled b y the Boussinesq equations. The problem was motivated by the design and op eration of low energy consumption buildings. The investigation of stabilit y for a fluid flow in the natural convection problem is important in the t heory of hydrodynamical stability. The challenge of stabilization of the Boussinesq equations arises from the stabilization of the Navier-Stokes eq uations and its coupling with the convection-diffusion equation for temper ature. In our current work\, we are interested in stabilizing a possible u nstable steady state solution to the Boussinesq equations on a bounded and connected domain. We show that a finite number of controls acting on a pa rt of the boundary through Neumann/Robin boundary conditions is sufficient to stabilize the full nonlinear equations in the neighborhood of this st eady state solution. Dirichlet boundary conditions are imposed on the rest of the boundary. Moreover\, we prove that a stabilizing feedback control law can be obtained based on the partial estimation of the system state by solving an extended Kalman filter problem for the linearized Boussines q equations. In particular\, a reduced order model is derived to constru ct a finite dimensional estimator. Numerical results are provided to illu strate the idea. In the end\, we discuss the problem of control design for the Boussinesq equations with zero diffusivity and its application to optimal mixing\, mass and energy transport during processing. DTSTART;TZID=America/New_York:20160216T133000 DTEND;TZID=America/New_York:20160216T143000 SEQUENCE:0 SUMMARY:Seminar: Weiwei Hu (University of Minnesota) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-weiwei-hu-university-of- minnesota-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Contro lling a Thermal Fluid: Theoretical and Computational Issues

\nAbstra ct: We first discuss the problem of designing a feedback law which local ly stabilizes a two dimensional thermal fluid modeled by the Boussinesq equations. The problem was motivated by the design and operation of low en ergy consumption buildings. The investigation of stability for a fluid flo w in the natural convection problem is important in the theory of hydrodyn amical stability. The challenge of stabilization of the Boussinesq equati ons arises from the stabilization of the Navier-Stokes equations and its c oupling with the convection-diffusion equation for temperature. In our cur rent work\, we are interested in stabilizing a possible unstable steady st ate solution to the Boussinesq equations on a bounded and connected domain . We show that a finite number of controls acting on a part of the boundar y through Neumann/Robin boundary conditions is sufficient to stabilize th e full nonlinear equations in the neighborhood of this steady state soluti on. Dirichlet boundary conditions are imposed on the rest of the boundary. Moreover\, we prove that a stabilizing feedback control law can be obtain ed based on the partial estimation of the system state by solving an ex tended Kalman filter problem for the linearized Boussinesq equations. In p articular\, a reduced order model is derived to construct a finite dimen sional estimator. Numerical results are provided to illustrate the idea. In the end\, we discuss the problem of control design for the Boussine sq equations with zero diffusivity and its application to optimal mixing\, mass and energy transport during processing.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6082@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Robust and efficient collocation methods for parameterized mode ls\nMonte Carlo (MC) methods for the construction of polynomial approximat ions are effective tools for building a computational surrogate of the par ametric variation for a model response. In this talk we investigate least- squares regularization of noisy data and compressive sampling recovery of sparse representations. We wish to minimize the number of samples required for a stable and accurate procedure. We propose an algorithm for a partic ular kind of weighted Monte Carlo approximation method based on sampling f rom the pluripotential equilibrium measure. Standard MC methods suffer fro m poor stability and accuracy for high-order approximations\, but the prop erties of the equilibrium measure allow us to derive quasi-optimal stateme nts of mathematical recoverability in both over- or undersampled regressio n problems. We also show that such an approach typically yields very stabl e\, high-order computational algorithms for parameterized PDE approximatio n. We present theoretical analysis to motivate the algorithm\, and numeric al results to illustrate that equilibrium measure-based approaches are sup erior to standard MC methods in many situations of interest\, notably in h igh-dimensional scenarios. DTSTART;TZID=America/New_York:20160218T133000 DTEND;TZID=America/New_York:20160218T143000 SEQUENCE:0 SUMMARY:Seminar: Akil Narayan (University of Utah) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-akil-narayan-university- of-utah-whitehead-304/ X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nRobust and ef ficient collocation methods for parameterized models

\nMonte Carlo ( MC) methods for the construction of polynomial approximations are effectiv e tools for building a computational surrogate of the parametric variation for a model response. In this talk we investigate least-squares regulariz ation of noisy data and compressive sampling recovery of sparse representa tions. We wish to minimize the number of samples required for a stable and accurate procedure. We propose an algorithm for a particular kind of weig hted Monte Carlo approximation method based on sampling from the pluripote ntial equilibrium measure. Standard MC methods suffer from poor stability and accuracy for high-order approximations\, but the properties of the equ ilibrium measure allow us to derive quasi-optimal statements of mathematic al recoverability in both over- or undersampled regression problems. We al so show that such an approach typically yields very stable\, high-order co mputational algorithms for parameterized PDE approximation. We present the oretical analysis to motivate the algorithm\, and numerical results to ill ustrate that equilibrium measure-based approaches are superior to standard MC methods in many situations of interest\, notably in high-dimensional s cenarios.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6199@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title:\nStructure-Enhancing Algorithms for Statistical Learning Problems\nAbstract:\nFor many problems in statistical machine learning an d data-driven decision-making\, massive datasets necessitate the use of sc alable algorithms that deliver sensible (interpretable) and statistically sound solutions. In this talk\, we discuss several scalable algorithms th at directly promote well-structured solutions in two related contexts: (i) sparse high-dimensional linear regression\, and (ii) low-rank matrix comp letion\, both of which are particularly relevant in modern machine learnin g. In the context of linear regression\, we study several boosting algori thms – which directly promote sparse solutions – from the perspective of m odern first-order methods in convex optimization. We use this perspective to derive the first-ever computational guarantees for existing boosting m ethods and to develop new algorithms with associated computational guarant ees as well. In the context of matrix completion\, we present an extensio n of the Frank-Wolfe method in convex optimization that is designed to ind uce near-optimal low-rank solutions for regularized matrix completion prob lems\, and we derive computational guarantees that trade off between low-r ank structure and data fidelity. For both problem contexts\, we present c omputational results using datasets from microarray and recommender system applications. DTSTART;TZID=America/New_York:20160223T133000 DTEND;TZID=America/New_York:20160223T143000 SEQUENCE:0 SUMMARY:Seminar: Paul Grigas (MIT) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-paul-grigas-mit-whitehea d-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:**

Structure-Enhancing Algorithms for Statistical Learning Problems

\n**Abstract:**

For many problems in
statistical machine learning and data-driven decision-making\, massive dat
asets necessitate the use of scalable algorithms that deliver sensible (in
terpretable) and statistically sound solutions. In this talk\, we discuss
several scalable algorithms that directly promote *well-structured solutions in two related contexts: (i) sparse high-dimensional linear re
gression\, and (ii) low-rank matrix completion\, both of which are particu
larly relevant in modern machine learning. In the context of linear regre
ssion\, we study several boosting algorithms – which directly promote spar
se solutions – from the perspective of modern first-order methods in conve
x optimization. We use this perspective to derive the first-ever computat
ional guarantees for existing boosting methods and to develop new algorith
ms with associated computational guarantees as well. In the context of ma
trix completion\, we present an extension of the Frank-Wolfe method in con
vex optimization that is designed to induce near-optimal low-rank solution
s for regularized matrix completion problems\, and we derive computational
guarantees that trade off between low-rank structure and data fidelity.
For both problem contexts\, we present computational results using dataset
s from microarray and recommender system applications.*

Title: Recen t theoretic and algorithmic advances in graph matching

\nAbstract: Inference across multiple graphs arises naturally in disciplines as varied as neuroscience\, physics\, and sociology. In a number of methodologies for joint inference across graphs\, however\, it is assumed that an explic it vertex correspondence is a priori known across the vertex sets of the g raphs. While this assumption is often reasonable\, in practice these corre spondences may be unobserved and/or errorfully observed\, and graph matchi ng—aligning a pair of graphs to minimize their edge disagreements—is used to align the graphs before performing subsequent inference. Graph matchin g is a computationally challenging and well-studied problem\, but few exis ting algorithms have theoretical support for their performance. For tract ability\, many algorithms begin by relaxing the problem’s binary constrain ts\, thus rendering applicable gradient-descent methodologies. We develop a state-of-the-art algorithm for solving an indefinite relaxed graph match ing problem\, and we show that under mild model assumptions\, our indefini te relaxation (when solved exactly) almost always uncovers the optimal per mutation\, while the commonly used convex relaxation almost always fails t o identify the optimal permutation. We highlight some of the practical an d theoretical implications of these results on real and synthetic data\, a nd we discuss recent work towards formalizing the connection between graph matching and pairwise mutual information.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6111@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Mediation: From Intuition to Data Analysis.\n \nModern causal i nference links the “top-down” representation of causal intuitions and “bot tom-up” data analysis with the aim of choosing policy. Two innovations tha t proved key for this synthesis were a formalization of Hume’s counterfact ual account of causation using potential outcomes (due to Jerzy Neyman)\, and viewing cause effect relationships via directed acyclic graphs (due to Sewall Wright). I will briefly review how a synthesis of these two ideas was instrumental in formally representing the notion of “causal effect” a s a parameter in the language of potential outcomes\, and discuss a comple te identification theory linking these types of causal parameters and obse rved data\, as well as approaches to estimation of the resulting statistic al parameters.\nI will then describe\, in more detail\, how my collaborato rs and I are applying the same approach to mediation\, the study of effect s along particular causal pathways. I consider mediated effects at their most general: I allow arbitrary models\, the presence of hidden variables\ , multiple outcomes\, longitudinal treatments\, and effects along arbitrar y sets of causal pathways. As was the case with causal effects\, there ar e three distinct but related problems to solve — a representation problem (what sort of potential outcome does an effect along a set of pathways cor respond to)\, an identification problem (can a causal parameter of interes t be expressed as a functional of observed data)\, and an estimation probl em (what are good ways of estimating the resulting statistical parameter). I report a complete solution to the first two problems\, and progress on the third. In particular\, my collaborators and I show that for some par ameters that arise in mediation settings\, triply robust estimators exist\ , which rely on an outcome model\, a mediator model\, and a treatment mode l\, and which remain consistent if any two of these three models are corre ct.\n \nSome of the reported results are a joint work with Eric Tchetgen\, Caleb Miles\, Phyllis Kanki\, and Seema Meloni. DTSTART;TZID=America/New_York:20160303T133000 DTEND;TZID=America/New_York:20160303T143000 SEQUENCE:0 SUMMARY:Seminar: Ilya Shpitser (Johns Hopkins University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-ilya-shpitzer-johns-hopk ins-university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nMediation: Fr om Intuition to Data Analysis.

\n\n

Modern causal inference l inks the “top-down” representation of causal intuitions and “bottom-up” da ta analysis with the aim of choosing policy. Two innovations that proved k ey for this synthesis were a formalization of Hume’s counterfactual accoun t of causation using potential outcomes (due to Jerzy Neyman)\, and viewin g cause effect relationships via directed acyclic graphs (due to Sewall Wr ight). I will briefly review how a synthesis of these two ideas was instr umental in formally representing the notion of “causal effect” as a parame ter in the language of potential outcomes\, and discuss a complete identif ication theory linking these types of causal parameters and observed data\ , as well as approaches to estimation of the resulting statistical paramet ers.

\nI will then describe\, in more detail\, how my collaborators and I are applying the same approach to mediation\, the study of effects a long particular causal pathways. I consider mediated effects at their mos t general: I allow arbitrary models\, the presence of hidden variables\, m ultiple outcomes\, longitudinal treatments\, and effects along arbitrary s ets of causal pathways. As was the case with causal effects\, there are t hree distinct but related problems to solve — a representation problem (wh at sort of potential outcome does an effect along a set of pathways corres pond to)\, an identification problem (can a causal parameter of interest b e expressed as a functional of observed data)\, and an estimation problem (what are good ways of estimating the resulting statistical parameter). I report a complete solution to the first two problems\, and progress on th e third. In particular\, my collaborators and I show that for some parame ters that arise in mediation settings\, triply robust estimators exist\, w hich rely on an outcome model\, a mediator model\, and a treatment model\, and which remain consistent if any two of these three models are correct.

\n\n

Some of the reported results are a joint work with Eric Tchetgen\, Caleb Miles\, Phyllis Kanki\, and Seema Meloni.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6462@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Arbitrage-Free Pricing of XVA.\n \nAbstract: We develop a framework for computing the total valuation adjustment (XVA) of a Europe an claim accounting for funding costs\, counterparty credit risk\, and col lateralization. Based on no-arbitrage arguments\, we derive backward stoch astic differential equations\n(BSDEs) associated with the replicating port folios of long and short positions in the claim. This leads to the definit ion of buyer’s and seller’s XVA\, which in turn identify a no-arbitrage in terval. In the case that borrowing and lending rates coincide\, we provide a fully explicit expression for the uniquely determined XVA\, expressed a s a percentage of the price of the traded claim\, and for the correspondin g replication strategies. In the general case of asymmetric funding\, repo and collateral rates\, we study the semi-linear partial differential equa tion (PDE) characterizing buyer’s and seller’s XVA and show the existence of a unique classical solution to it. To illustrate our results\, we condu ct a numerical study demonstrating how funding costs\, repo rates\, and co unterparty risk contribute to determine the total valuation adjustment. Th is talk is based on joint works with Agostino Capponi (Columbia) and Steph an Sturm (WPI). DTSTART;TZID=America/New_York:20160324T133000 DTEND;TZID=America/New_York:20160324T143000 SEQUENCE:0 SUMMARY:Seminar: Maxim Bichuch (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-maxim-bichuch-jhu-whiteh ead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Arbitr age-Free Pricing of XVA.

\n\n

Abstract: We develop a framewor k for computing the total valuation adjustment (XVA) of a European claim a ccounting for funding costs\, counterparty credit risk\, and collateraliza tion. Based on no-arbitrage arguments\, we derive backward stochastic diff erential equations

\n(BSDEs) associated with the replicating portfol ios of long and short positions in the claim. This leads to the definition of buyer’s and seller’s XVA\, which in turn identify a no-arbitrage inter val. In the case that borrowing and lending rates coincide\, we provide a fully explicit expression for the uniquely determined XVA\, expressed as a percentage of the price of the traded claim\, and for the corresponding r eplication strategies. In the general case of asymmetric funding\, repo an d collateral rates\, we study the semi-linear partial differential equatio n (PDE) characterizing buyer’s and seller’s XVA and show the existence of a unique classical solution to it. To illustrate our results\, we conduct a numerical study demonstrating how funding costs\, repo rates\, and count erparty risk contribute to determine the total valuation adjustment. This talk is based on joint works with Agostino Capponi (Columbia) and Stephan Sturm (WPI).

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6116@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Co-clustering of nonsmooth graphons\nAbstract:\nTheoreti cal results are becomming known for community detection and clustering of networks\; however\, these results assume an idealized generative model th at is unlikely to hold in many settings. Here we consider exploratory co-c lustering of a bipartite network\, where the rows and columns of the adjac ency matrix are assumed to be samples from an arbitrary population. This i s equivalent to assuming that the data is generated from a nonparametric m odel known as a graphon. We show that co-clusters found by any method can be extended to the row and column populations\, or equivalently that the e stimated blockmodel approximates a blocked version of the generative graph on\, with generalization error bounded by n^{-1/2}. Analogous results are also shown for degree-corrected co-blockmodels and random dot product bipa rtite graphs\, with error rates depending on the dimensionality of the lat ent variable space. DTSTART;TZID=America/New_York:20160331T133000 DTEND;TZID=America/New_York:20160331T143000 SEQUENCE:0 SUMMARY:Seminar: David Choi (CMU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-david-choi-cmu-whitehead -304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Co-clu stering of nonsmooth graphons

\nAbstract:

\nTheoretical result s are becomming known for community detection and clustering of networks\; however\, these results assume an idealized generative model that is unli kely to hold in many settings. Here we consider exploratory co-clustering of a bipartite network\, where the rows and columns of the adjacency matri x are assumed to be samples from an arbitrary population. This is equivale nt to assuming that the data is generated from a nonparametric model known as a graphon. We show that co-clusters found by any method can be extende d to the row and column populations\, or equivalently that the estimated b lockmodel approximates a blocked version of the generative graphon\, with generalization error bounded by n^{-1/2}. Analogous results are also shown for degree-corrected co-blockmodels and random dot product bipartite grap hs\, with error rates depending on the dimensionality of the latent variab le space.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6466@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Distributed proximal gradient methods for cooperative mu lti-agent consensus optimization\n \nAbstract:\nIn this talk\, I will dis cuss decentralized methods for solving cooperative multi-agent consensus o ptimization problems. Consider an undirected network of agents\, where onl y those agents connected by an edge can directly communicate with each oth er. The objective is to minimize the sum of agent-specific composite conve x functions\, i.e.\, each term in the sum is a private cost function belon ging to an agent. In the first part\, I will discuss the unconstrained cas e\, and in the second part I will focus on the constrained case\, where ea ch agent has a private conic constraint set. For the constrained case the optimal consensus decision should lie in the intersection of these private sets. This optimization model abstracts a number of applications in machi ne learning\, distributed control\, and estimation using sensor networks. I will discuss different types of distributed algorithms\; in particular\, I will describe methods based on inexact augmented Lagrangian\, and linea rized ADMM. I will provide convergence rates both in sub-optimality error and consensus violation\; and also examine the effect of underlying networ k topology on the convergence rates of the proposed decentralized algorith ms.\n \nJoint work with Ph.D. students Zi Wang\, Erfan Yazdandoost\, and S hiqian Ma from Chinese University of Hong Kong\, and Garud Iyengar from Co lumbia University. DTSTART;TZID=America/New_York:20160407T133000 DTEND;TZID=America/New_York:20160407T143000 SEQUENCE:0 SUMMARY:Seminar: Serhat Aybat (Penn State University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-serhat-aybat-penn-state- university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
: **Distributed proximal gradient methods for cooperative multi-age
nt consensus optimization

** **** **

**Abstract:**

In this talk\, I will discuss dece ntralized methods for solving cooperative multi-agent consensus optimizati on problems. Consider an undirected network of agents\, where only those a gents connected by an edge can directly communicate with each other. The o bjective is to minimize the sum of agent-specific composite convex functio ns\, i.e.\, each term in the sum is a private cost function belonging to a n agent. In the first part\, I will discuss the unconstrained case\, and i n the second part I will focus on the constrained case\, where each agent has a private conic constraint set. For the constrained case the optimal c onsensus decision should lie in the intersection of these private sets. Th is optimization model abstracts a number of applications in machine learni ng\, distributed control\, and estimation using sensor networks. I will di scuss different types of distributed algorithms\; in particular\, I will d escribe methods based on inexact augmented Lagrangian\, and linearized ADM M. I will provide convergence rates both in sub-optimality error and conse nsus violation\; and also examine the effect of underlying network topolog y on the convergence rates of the proposed decentralized algorithms.

\n\n

Joint work with Ph.D. students Zi Wang\, Erfan Yazdandoost\, and Shiqian Ma from Chinese University of Hong Kong\, and Garud Iyengar fr om Columbia University.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6084@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Scalable Bayesian Models of Interacting Time Series\n \n Abstract:\nData streams of increasing complexity and scale are being colle cted in a variety of fields ranging from neuroscience\, genomics\, and env ironmental monitoring to e-commerce. Modeling the intricate and possibly evolving relationships between the large collection of series can lead to increased predictive performance and domain-interpretable structures. For scalability\, it is crucial to discover and exploit sparse dependencies b etween the data streams. Such representational structures for independent data sources have been studied extensively\, but have received limited at tention in the context of time series. In this talk\, we present a series of Bayesian models for capturing such sparse dependencies via clustering\ , graphical models\, and low-dimensional embeddings of time series. We e xplore these methods in a variety of applications\, including house price modeling and inferring networks in the brain.\nWe then turn to observed in teraction data\, and briefly touch upon how to devise statistical network models that capture important network features like sparsity of edge conne ctivity. Within our Bayesian framework\, a key insight is to move to a co ntinuous-space representation of the graph\, rather than the typical discr ete adjacency matrix structure. We demonstrate our methods on a series of real-world networks with up to hundreds of thousands of nodes and million s of edges.\n \nBio:\nEmily Fox is currently the Amazon Professor of Machi ne Learning in the Statistics Department at the University of Washington. She received a S.B. in 2004 and Ph.D. in 2009 from the Department of Elec trical Engineering and Computer Science at MIT. She has been awarded a Sl oan Research Fellowship (2015)\, an ONR Young Investigator award (2015)\, an NSF CAREER award (2014)\, the Leonard J. Savage Thesis Award in Applied Methodology (2009)\, and the MIT EECS Jin-Au Kong Outstanding Doctoral Th esis Prize (2009). Her research interests are in large-scale Bayesian dyna mic modeling and computations. DTSTART;TZID=America/New_York:20160414T133000 DTEND;TZID=America/New_York:20160414T143000 SEQUENCE:0 SUMMARY:Seminar: Emily Fox (University of Washington) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-emily-fox-university-of- washington-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Scalab le Bayesian Models of Interacting Time Series

\n\n

Abstract:< /p>\n

Data streams of increasing complexity and scale are being collecte d in a variety of fields ranging from neuroscience\, genomics\, and enviro nmental monitoring to e-commerce. Modeling the intricate and possibly evo lving relationships between the large collection of series can lead to inc reased predictive performance and domain-interpretable structures. For sc alability\, it is crucial to discover and exploit sparse dependencies betw een the data streams. Such representational structures for independent da ta sources have been studied extensively\, but have received limited atten tion in the context of time series. In this talk\, we present a series of Bayesian models for capturing such sparse dependencies via clustering\, g raphical models\, and low-dimensional embeddings of time series. We expl ore these methods in a variety of applications\, including house price mod eling and inferring networks in the brain.

\nWe then turn to observe d interaction data\, and briefly touch upon how to devise statistical netw ork models that capture important network features like sparsity of edge c onnectivity. Within our Bayesian framework\, a key insight is to move to a continuous-space representation of the graph\, rather than the typical d iscrete adjacency matrix structure. We demonstrate our methods on a serie s of real-world networks with up to hundreds of thousands of nodes and mil lions of edges.

\n\n

Bio:

\nEmily Fox is currently the Amazon Professor of Machine Learning in the Statistics Department at the U niversity of Washington. She received a S.B. in 2004 and Ph.D. in 2009 fr om the Department of Electrical Engineering and Computer Science at MIT. She has been awarded a Sloan Research Fellowship (2015)\, an ONR Young Inv estigator award (2015)\, an NSF CAREER award (2014)\, the Leonard J. Savag e Thesis Award in Applied Methodology (2009)\, and the MIT EECS Jin-Au Kon g Outstanding Doctoral Thesis Prize (2009). Her research interests are in large-scale Bayesian dynamic modeling and computations.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6032@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Quickest detection in correlated and coupled systems\nAb stract:\nIn this works we consider the problem N-dimensional quickest dete ction in correlated and coupled systems. The objective is to detect the fi rst time that the system of N sensors undergoes a change with a one shot c ommunication to the central fusion center.\nIn both cases it is seen that the minimum of N – cumulative sum tests with appropriately chosen threshol ds is asymptotically optimal in managing the trade off between a small det ection delay and a large mean time to first False alarm as the mean time t o the first false alarm increases without bound. In the former case a Line ar penalty is used for detection delay while in the latter a Kulback- Leib ler distance of the measure before and after regime switching is used. DTSTART;TZID=America/New_York:20160421T133000 DTEND;TZID=America/New_York:20160421T143000 SEQUENCE:0 SUMMARY:Seminar: Olympia Hadjiliadis (City University of New York\, Brookly n) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-olympia-hadjiliadis-city -university-of-new-york-brooklyn/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Quicke st detection in correlated and coupled systems

\nAbstract:

\nIn
this works we consider the problem N-dimensional quickest detection in co
rrelated and coupled systems. The objective is to detect the first time th
at the system of N sensors undergoes a change with a one shot communicatio
n to the central fusion center.

\nIn both cases it is seen that the m
inimum of N – cumulative sum tests with appropriately chosen thresholds is
asymptotically optimal in managing the trade off between a small detectio
n delay and a large mean time to first False alarm as the mean time to the
first false alarm increases without bound. In the former case a Linear pe
nalty is used for detection delay while in the latter a Kulback- Leibler d
istance of the measure before and after regime switching is used.

**Movie
Reconstruction from Brain Signals: “Mind-Reading”**

In a t hrilling breakthrough at the intersection of neuroscience and statistics\, penalized Least Squares methods have been used to construct a “mind-readi ng” algorithm that reconstructs movies from fMRI brain signals. The story of this algorithm is a fascinating tale of the interdisciplinary research that led to the development of the system which was selected as one of Tim e Magazine’s 50 Best Inventions of 2011. Talk 1: Movie Reconstruction from Brain Signals: “Mind-Reading”

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6112@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Unveiling the mysteries in spatial gene expression\nGenome-wide data reveal an intricate landscape where gene activities are highly diffe rentiated across diverse spatial areas. These gene actions and interaction s play a critical role in the development and function of both normal and abnormal tissues. As a result\, understanding spatial heterogeneity of gen e networks is key to developing treatments for human diseases. Despite the abundance of recent spatial gene expression data\, extracting meaningful information remains a challenge for local gene interaction discoveries. In response\, we have developed staNMF\, a method that combines a powerful u nsupervised learning algorithm\, nonnegative matrix factorization (NMF)\, with a new stability criterion that selects the size of the dictionary. Us ing staNMF\, we generate biologically meaningful Principle Patterns (PP)\, which provide a novel and concise representation of Drosophila embryonic spatial expression patterns that correspond to pre-organ areas of the deve loping embryo. Furthermore\, we show how this new representation can be us ed to automatically predict manual annotations\, categorize gene expressio n patterns\, and reconstruct the local gap gene network with high accuracy . Finally\, we discuss on-going crispr/cas9 knock-out experiments on Droso phila to verify predicted local gene-gene interactions involving gap-genes . An open-source software is also being built based on SPARK and Fiji.\nTh is talk is based on collaborative work of a multi-disciplinary team (co-le ad Erwin Frise) from the Yu group (statistics) at UC Berkeley\, the Celnik er group (biology) at the Lawrence Berkeley National Lab (LBNL)\, and the Xu group (computer science) at Hsinghua Univ. DTSTART;TZID=America/New_York:20160428T133000 DTEND;TZID=America/New_York:20160428T143000 SEQUENCE:0 SUMMARY:Duncan Lecture Series: Bin Yu (University of California Berkeley) @ Mergenthaler 111 URL:https://engineering.jhu.edu/ams/events/duncan-lecture-series-bin-yu-uni versity-of-california-berkeley/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Unvei
ling the mysteries in spatial gene expression**

Genome-wide data reveal an intricate landscape where gene activities are highly diffe rentiated across diverse spatial areas. These gene actions and interaction s play a critical role in the development and function of both normal and abnormal tissues. As a result\, understanding spatial heterogeneity of gen e networks is key to developing treatments for human diseases. Despite the abundance of recent spatial gene expression data\, extracting meaningful information remains a challenge for local gene interaction discoveries. In response\, we have developed staNMF\, a method that combines a powerful u nsupervised learning algorithm\, nonnegative matrix factorization (NMF)\, with a new stability criterion that selects the size of the dictionary. Us ing staNMF\, we generate biologically meaningful Principle Patterns (PP)\, which provide a novel and concise representation of Drosophila embryonic spatial expression patterns that correspond to pre-organ areas of the deve loping embryo. Furthermore\, we show how this new representation can be us ed to automatically predict manual annotations\, categorize gene expressio n patterns\, and reconstruct the local gap gene network with high accuracy . Finally\, we discuss on-going crispr/cas9 knock-out experiments on Droso phila to verify predicted local gene-gene interactions involving gap-genes . An open-source software is also being built based on SPARK and Fiji.

\nThis talk is based on collaborative work of a multi-disciplinary team (co-lead Erwin Frise) from the Yu group (statistics) at UC Berkeley\, the Celniker group (biology) at the Lawrence Berkeley National Lab (LBNL)\, a nd the Xu group (computer science) at Hsinghua Univ.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6970@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20160901T133000 DTEND;TZID=America/New_York:20160901T143000 SEQUENCE:0 SUMMARY:Seminar: Getting to Know You @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-getting-to-know-you-whit ehead-304/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-7268@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION:Statistics of the Stability Bounds in the Phase Retrieval Probl em\nIn this talk we present a local-global Lipschitz analysis of the phase retrieval problem. Additionally we present tentative estimates of the tai l-bound for the distribution of the global Lipschitz constants. Specifical ly it is known that if the frame {f1\,…\,fm} for Cn is phase retrievable t hen there are constants a0 and b0 so that for every x\,y∈Cn: a0∣∣xx*-yy*∣∣ 12≤∑k=1m∣∣〈x\,fk〉∣2-∣〈y\,fk〉∣2∣2≤b0∣∣xx*-yy*∣∣12. Assumef1\,…\,fm are inde pendent realizations with entries from CN(0\,1). In this talk we establish estimates for the probability P(a0>a). DTSTART;TZID=America/New_York:20160907T150000 DTEND;TZID=America/New_York:20160907T160000 SEQUENCE:0 SUMMARY:Data Seminar: Radu Balan (University of Maryland College Park) @ Sh affer 100 URL:https://engineering.jhu.edu/ams/events/data-seminar-radu-balan-universi ty-maryland-college-park-shaffer-100/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nStatistics of
the Stability Bounds in the Phase Retrieval Problem

\nIn this talk w
e present a local-global Lipschitz analysis of the phase retrieval problem
. Additionally we present tentative estimates of the tail-bound for the di
stribution of the global Lipschitz constants. Specifically it is known tha
t if the frame {f1\,…\,fm} for Cn is phase retrievable then there are cons
tants a0 and b0 so that for every x\,y∈Cn: a0∣∣xx*-yy*∣∣12≤∑k=1m∣∣〈x\,fk〉∣
2-∣〈y\,fk〉∣2∣2≤b0∣∣xx*-yy*∣∣12. Assumef1\,…\,fm are independent realizatio
ns with entries from CN(0\,1). In this talk we establish estimates for the
probability P(a0>a).

Title: To Rep lace or Not to Replace in Finite Population Sampling

\nAbstract:

\nWe revisit the classical result in finite population sampling which s
tates that in *equally-likely *“simple” random sampling the sample
mean is more reliable when we do not replace after each draw. In this tal
k\, we review a classical result for the equally likely sampling case. The
n we investigate if and when the same is true for samples where it may no
longer be true that each member of the population has an equal chance of b
eing selected\, and when the population mean is estimated using the Horvit
z-Thompson inverse probability weighing to produce an unbiased estimator.
For a certain class of sampling schemes\, we are able to obtain convenien
t expressions for the variance of the sample mean and surprisingly\, we fi
nd that for some selection distributions a more reliable estimate of the p
opulation mean will happen by replacing after each draw. We show for sele
ction distributions lying in a certain polytope the classical result preva
ils.

\n

This is joint work with Fred Torcaso.

\n< /HTML> END:VEVENT BEGIN:VEVENT UID:ai1ec-7272@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION:Finite-Sample Bounds for Geometric Multiresolution Analysis\n DTSTART;TZID=America/New_York:20160914T150000 DTEND;TZID=America/New_York:20160914T160000 SEQUENCE:0 SUMMARY:Data Seminar: Nate Strawn (Georgetown University) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-nate-strawn-georget own-university-krieger-309/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nFinite-Sample Bounds for Geometric Multiresolution Analysis

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-6996@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Stochastic Newton Methods for Machine Learning\nOptimization me thods play a crucial role in supervised learning where they are employed t o solve problems in very high dimensional parameter spaces. The optimizati on problems are inherently stochastic and often involve huge data sets. Th ere has recently been much interest in sub-sampled Newton methods for thes e types of applications. The methods use approximations to the gradient an d Hessian in a way that strikes a balance between computational effort and speed of convergence. We provide a review and analysis of sub-sampled New ton methods\, which include Newton-sketch and non-uniform subsampling tech niques\, and illustrate their effectiveness on some large-scale machine le arning applications.\n \nJorge Nocedal is the David and Karen Sachs Profes sor of Industrial Engineering and Management Sciences at Northwestern Univ ersity. He received his PhD in Mathematical Sciences from Rice University and was a postdoctoral fellow at the Courant Institute. His research is in nonlinear optimization with applications to machine learning. Over the ye ars\, his work has spanned algorithms\, analysis and software. He is a SIA M Fellow\, has been an invited speaker at the International Congress of Ma thematicians\, and was awarded the 2012 George B. Dantzig Prize.\n DTSTART;TZID=America/New_York:20160915T133000 DTEND;TZID=America/New_York:20160915T143000 SEQUENCE:0 SUMMARY:Goldman Lecture Series: Jorge Nocedal (Northwestern University)- Ma ryland 110 URL:https://engineering.jhu.edu/ams/events/goldman-lecture-series-jorge-noc edal-northwestern-university-location-tbd/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Stochastic Newton Methods fo
r Machine Learning**

Optimization methods play a crucial role in supervised learning where they are employed to solve problems in very high dimension al parameter spaces. The optimization problems are inherently stochastic a nd often involve huge data sets. There has recently been much interest in sub-sampled Newton methods for these types of applications. The methods us e approximations to the gradient and Hessian in a way that strikes a balan ce between computational effort and speed of convergence. We provide a rev iew and analysis of sub-sampled Newton methods\, which include Newton-sket ch and non-uniform subsampling techniques\, and illustrate their effective ness on some large-scale machine learning applications.

\n\n

Jorge Nocedal is the David and Karen Sachs Professor of Industrial Engineering and Managem ent Sciences at Northwestern University. He received his PhD in Mathematic al Sciences from Rice University and was a postdoctoral fellow at the Cour ant Institute. His research is in nonlinear optimization with applications to machine learning. Over the years\, his work has spanned algorithms\, a nalysis and software. He is a SIAM Fellow\, has been an invited speaker at the International Congress of Mathematicians\, and was awarded the 2012 G eorge B. Dantzig Prize.

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7352@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:The JHU Actuarial Club will host an event on September 16th. T he speaker JHU alum\, Matt Sedlock\, is currently working at Mass Mutual. He will share his experience working in the actuarial industry and discus s the recruiting process from Mass Mutual DTSTART;TZID=America/New_York:20160916T180000 DTEND;TZID=America/New_York:20160916T200000 SEQUENCE:0 SUMMARY:What is an Actuary? @ Arellano Theatre\, Levering Hall URL:https://engineering.jhu.edu/ams/events/actuary-arellano-theatre-leverin g-hall/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

The JHU Actua rial Club will host an event on September 16th. The speaker JHU alum\, Ma tt Sedlock\, is currently working at Mass Mutual. He will share his exper ience working in the actuarial industry and discuss the recruiting process from Mass Mutual

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7276@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20160921T150000 DTEND;TZID=America/New_York:20160921T160000 SEQUENCE:0 SUMMARY:Data Seminar: Charles Meneveau (JHU) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-charles-meneveau-jh u-shaffer-100/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-6964@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:High-Dimensional Analysis of Stochastic Algorithms for Convex a nd Nonconvex Optimization: Limiting Dynamics and Phase Transitions\n \nAbs tract\nWe consider efficient iterative methods (e.g.\, stochastic gradient descent\, randomized Kaczmarz algorithms\, iterative coordinate descent) for solving large-scale optimization problems\, whether convex or nonconve x. A flurry of recent work has focused on establishing their theoretical p erformance guarantees. This intense interest is spurred on by the remarkab ly impressive empirical performance achieved by these low-complexity and m emory-efficient methods.\nIn this talk\, we will present a framework for a nalyzing the exact dynamics of these methods in the high-dimensional limit . For concreteness\, we consider two prototypical problems: regularized li near regression (e.g. LASSO) and sparse principal component analysis. For each case\, we show that the time-varying estimates given by the algorithm s will converge weakly to a deterministic “limiting process” in the high-d imensional (scaling and mean-field) limit. Moreover\, this limiting proces s can be characterized as the unique solution of a nonlinear PDE\, and it provides exact information regarding the asymptotic performance of the alg orithms. For example\, performance metrics such as the MSE\, the cosine si milarity and the misclassification rate in sparse support recovery can all be obtained by examining the deterministic limiting process. A steady-sta te analysis of the nonlinear PDE also reveals interesting phase transition phenomenons related to the performance of the algorithms. Although our an alysis is asymptotic in nature\, numerical simulations show that the theor etical predictions are accurate for moderate signal dimensions.\nWhat make s our analysis tractable is the notion of exchangeability\, a fundamental property of symmetry that is inherent in many of the optimization problems encountered in signal processing and machine learning.\n \nBio\nYue M. Lu was born in Shanghai. After finishing undergraduate studies at Shanghai J iao Tong University\, he attended the University of Illinois at Urbana-Cha mpaign\, where he received the M.Sc. degree in mathematics and the Ph.D. d egree in electrical engineering\, both in 2007. He was a Research Assistan t at the University of Illinois at Urbana-Champaign\, and has worked for M icrosoft Research Asia\, Beijing\, and Siemens Corporate Research\, Prince ton\, NJ. Following his work as a postdoctoral researcher at the Audiovisu al Communications Laboratory at Ecole Polytechnique Fédérale de Lausanne ( EPFL)\, Switzerland\, he joined Harvard University in 2010\, where he is c urrently an Associate Professor of Electrical Engineering at the John A. P aulson School of Engineering and Applied Sciences.\nHe received the Most I nnovative Paper Award (with Minh N. Do) of IEEE International Conference o n Image Processing (ICIP) in 2006\, the Best Student Paper Award of IEEE I CIP in 2007\, and the Best Student Presentation Award at the 31st SIAM SEA S Conference in 2007. Student papers supervised and coauthored by him won the Best Student Paper Award (with Ivan Dokmanic and Martin Vetterli) of I EEE International Conference on Acoustics\, Speech and Signal Processing i n 2011 and the Best Student Paper Award (with Ameya Agaskar and Chuang Wan g) of IEEE Global Conference on Signal and Information Processing (GlobalS IP) in 2014.\nHe has been an Associate Editor of the IEEE Transactions on Image Processing since 2014\, an Elected Member of the IEEE Image\, Video\ , and Multidimensional Signal Processing Technical Committee since 2015\, and an Elected Member of the IEEE Signal Processing Theory and Methods Tec hnical Committee since 2016. He received the ECE Illinois Young Alumni Ach ievement Award in 2015. DTSTART;TZID=America/New_York:20160922T133000 DTEND;TZID=America/New_York:20160922T143000 SEQUENCE:0 SUMMARY:Seminar: Yue Lu (Harvard) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-yue-lu-harvard-whitehead -304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**High-
Dimensional Analysis of Stochastic Algorithms for Convex and Nonconvex Opt
imization: Limiting Dynamics and Phase Transitions**

\n

**Abstract**

We consider efficient iterative met hods (e.g.\, stochastic gradient descent\, randomized Kaczmarz algorithms\ , iterative coordinate descent) for solving large-scale optimization probl ems\, whether convex or nonconvex. A flurry of recent work has focused on establishing their theoretical performance guarantees. This intense intere st is spurred on by the remarkably impressive empirical performance achiev ed by these low-complexity and memory-efficient methods.

\nIn this t alk\, we will present a framework for analyzing the exact dynamics of thes e methods in the high-dimensional limit. For concreteness\, we consider tw o prototypical problems: regularized linear regression (e.g. LASSO) and sp arse principal component analysis. For each case\, we show that the time-v arying estimates given by the algorithms will converge weakly to a determi nistic “limiting process” in the high-dimensional (scaling and mean-field) limit. Moreover\, this limiting process can be characterized as the uniqu e solution of a nonlinear PDE\, and it provides exact information regardin g the asymptotic performance of the algorithms. For example\, performance metrics such as the MSE\, the cosine similarity and the misclassification rate in sparse support recovery can all be obtained by examining the deter ministic limiting process. A steady-state analysis of the nonlinear PDE al so reveals interesting phase transition phenomenons related to the perform ance of the algorithms. Although our analysis is asymptotic in nature\, nu merical simulations show that the theoretical predictions are accurate for moderate signal dimensions.

\nWhat makes our analysis tractable is the notion of exchangeability\, a fundamental property of symmetry that is inherent in many of the optimization problems encountered in signal proce ssing and machine learning.

\n\n

**Bio**

Yue M. Lu was born in Shanghai. After finishing undergraduate studies at Shanghai Jiao Tong University\, he attended the University of Illinois at Urbana-Champaign\, where he received the M.Sc. degree in mathematics and t he Ph.D. degree in electrical engineering\, both in 2007. He was a Researc h Assistant at the University of Illinois at Urbana-Champaign\, and has wo rked for Microsoft Research Asia\, Beijing\, and Siemens Corporate Researc h\, Princeton\, NJ. Following his work as a postdoctoral researcher at the Audiovisual Communications Laboratory at Ecole Polytechnique Fédérale de Lausanne (EPFL)\, Switzerland\, he joined Harvard University in 2010\, whe re he is currently an Associate Professor of Electrical Engineering at the John A. Paulson School of Engineering and Applied Sciences.

\nHe re ceived the Most Innovative Paper Award (with Minh N. Do) of IEEE Internati onal Conference on Image Processing (ICIP) in 2006\, the Best Student Pape r Award of IEEE ICIP in 2007\, and the Best Student Presentation Award at the 31st SIAM SEAS Conference in 2007. Student papers supervised and coaut hored by him won the Best Student Paper Award (with Ivan Dokmanic and Mart in Vetterli) of IEEE International Conference on Acoustics\, Speech and Si gnal Processing in 2011 and the Best Student Paper Award (with Ameya Agask ar and Chuang Wang) of IEEE Global Conference on Signal and Information Pr ocessing (GlobalSIP) in 2014.

\nHe has been an Associate Editor of t he IEEE Transactions on Image Processing since 2014\, an Elected Member of the IEEE Image\, Video\, and Multidimensional Signal Processing Technical Committee since 2015\, and an Elected Member of the IEEE Signal Processin g Theory and Methods Technical Committee since 2016. He received the ECE I llinois Young Alumni Achievement Award in 2015.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7280@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION:From Molecular Dynamics to Large Scale Inference\n\nMolecular m odels and data analytics problems give rise to very large systems of stoch astic differential equations (SDEs) whose paths are designed to ergodicall y sample multimodal probability distributions. An important challenge for the numerical analyst (or the data scientist\, for that matter) is the des ign of numerical procedures to generate these paths. One of the interestin g ideas is to construct stochastic numerical methods with close attention to the error in the invariant measure. Another is to redesign the underlyi ng stochastic dynamics to reduce bias or locally transform variables to en hance sampling efficiency. I will illustrate these ideas with various exam ples\, including a geodesic integrator for constrained Langevin dynamics [ 1] and an ensemble sampling strategy for distributed inference [2]. DTSTART;TZID=America/New_York:20160923T150000 DTEND;TZID=America/New_York:20160923T160000 SEQUENCE:0 SUMMARY:Data Seminar: Ben Leimkuhler (University of Edinburgh) @ TBD URL:https://engineering.jhu.edu/ams/events/data-seminar-ben-leimkuhler-univ ersity-edinburgh-tbd/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nFrom Molecular Dynamics to Large Scale Inference

\n

\nMolecular models and data analytics problems give rise to v
ery large systems of stochastic differential equations (SDEs) whose paths
are designed to ergodically sample multimodal probability distributions. A
n important challenge for the numerical analyst (or the data scientist\, f
or that matter) is the design of numerical procedures to generate these pa
ths. One of the interesting ideas is to construct stochastic numerical met
hods with close attention to the error in the invariant measure. Another i
s to redesign the underlying stochastic dynamics to reduce bias or locally
transform variables to enhance sampling efficiency. I will illustrate the
se ideas with various examples\, including a geodesic integrator for const
rained Langevin dynamics [1] and an ensemble sampling strategy for distrib
uted inference [2].

Title: Edge-c oloring Multigraphs

\nAbstract: Graph (vertex) coloring is a central area of discrete math\; however\, it is NP-hard even to approximate the c hromatic number. Edge-coloring can be seen as a special case of vertex co loring. As such\, we may hope that computing (or approximating) the edge chromatic number is easier\; in fact\, it is. We will survey a number of theorems and conjectures on edge-coloring. These include many instances w hen the edge-chromatic number satisfies a trivial lower bound with equalit y\, such as when it equals the graph’s maximum degree. I will also mentio n some of my recent work in this area\, and introduce one of the main tool s\, Tashkinov trees\, which rely on a beautiful double induction.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7439@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION:Geometric Methods for the Approximation of High-dimensional Dyn amical Systems\n\nI will discuss a geometry-based statistical learning fra mework for performing model reduction and modeling of stochastic high-dime nsional dynamical systems. I will consider two complementary settings. In the first one\, I am given long trajectories of a system\, e.g. from molec ular dynamics\, and I discuss techniques for estimating\, in a robust fash ion\, an effective number of degrees of freedom of the system\, which may vary in the state space of then system\, and a local scale where the dynam ics is well-approximated by a reduced dynamics with a small number of degr ees of freedom. I will then use these ideas to produce an approximation to the generator of the system and obtain\, via eigenfunctions of an empiric al Fokker-Planck question\, reaction coordinates for the system that captu re the large time behavior of the dynamics. I will present various example s from molecular dynamics illustrating these ideas. In the second setting I assume I only have access to a (large number of expensive) simulators th at can return short simulations of high-dimensional stochastic system\, an d introduce a novel statistical learning framework for learning automatica lly a family of local approximations to the system\, that can be (automati cally) pieced together to form a fast global reduced model for the system\ , called ATLAS. ATLAS is guaranteed to be accurate (in the sense of produc ing stochastic paths whose distribution is close to that of paths generate d by the original system) not only at small time scales\, but also at larg e time scales\, under suitable assumptions on the dynamics. I discuss appl ications to homogenization of rough diffusions in low and high dimensions\ , as well as relatively simple systems with separations of time scales\, a nd deterministic chaotic systems in high-dimensions\, that are well-approx imated by stochastic differential equations.\nNo knowledge of molecular dy namics is required\, and the techniques above are quite universal. Ideas i n the first part of the talk are based on what is called Diffusion Geometr y\, and have been used widely in data analysis\; ideas in the second part are applicable to MCMC. The talk will be accessible to students with a wid e variety of backgrounds and interests. DTSTART;TZID=America/New_York:20161005T150000 DTEND;TZID=America/New_York:20161005T160000 SEQUENCE:0 SUMMARY:Data Seminar: Mauro Maggioni (JHU) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-mauro-maggioni-jhu- krieger-309/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nGeometric Methods for the Approximation of High-dimensional Dynamical
Systems

\n

\nI will discuss a geometry-base
d statistical learning framework for performing model reduction and modeli
ng of stochastic high-dimensional dynamical systems. I will consider two c
omplementary settings. In the first one\, I am given long trajectories of
a system\, e.g. from molecular dynamics\, and I discuss techniques for est
imating\, in a robust fashion\, an effective number of degrees of freedom
of the system\, which may vary in the state space of then system\, and a l
ocal scale where the dynamics is well-approximated by a reduced dynamics w
ith a small number of degrees of freedom. I will then use these ideas to p
roduce an approximation to the generator of the system and obtain\, via ei
genfunctions of an empirical Fokker-Planck question\, reaction coordinates
for the system that capture the large time behavior of the dynamics. I wi
ll present various examples from molecular dynamics illustrating these ide
as. In the second setting I assume I only have access to a (large number o
f expensive) simulators that can return short simulations of high-dimensio
nal stochastic system\, and introduce a novel statistical learning framewo
rk for learning automatically a family of local approximations to the syst
em\, that can be (automatically) pieced together to form a fast global red
uced model for the system\, called ATLAS. ATLAS is guaranteed to be accura
te (in the sense of producing stochastic paths whose distribution is close
to that of paths generated by the original system) not only at small time
scales\, but also at large time scales\, under suitable assumptions on th
e dynamics. I discuss applications to homogenization of rough diffusions i
n low and high dimensions\, as well as relatively simple systems with sepa
rations of time scales\, and deterministic chaotic systems in high-dimensi
ons\, that are well-approximated by stochastic differential equations.

*No knowledge of molecular dynamics is required\, and the technique
s above are quite universal. Ideas in the first part of the talk are based
on what is called Diffusion Geometry\, and have been used widely in data
analysis\; ideas in the second part are applicable to MCMC. The talk will
be accessible to students with a wide variety of backgrounds and interests
.*

Title:

\n< p>Stochastic Search Methods for Simulation Optimization\n\n

Abstract:

\nA variety of systems arising in finance\, engineering de sign\, and manufacturing require the use of optimization techniques to imp rove their performance. Due to the complexity and stochastic dynamics of s uch systems\, their performance evaluation frequently requires computer si mulation\, which however often lacks structure needed by classical optimiz ation methods. We developed a gradient-based stochastic search approach\, based on the idea of converting the original (structure-lacking) problem t o a differentiable optimization problem on the parameter space of a sampli ng distribution that guides the search. A two-timescale updating scheme is further studied and incorporated to improve the algorithm efficiency. Con vergence properties of our approach are established through techniques fro m stochastic approximation\, and the performance of our algorithms is illu strated in comparison with some state-of-the-art simulation optimization m ethods. This is a joint work with Jiaqiao Hu (Stony Brook University) and Shalabh Bhartnagar (Indian Institute of Science).

\n\n

Biogra phy:

\nEnlu Zhou is currently an associate professor in the H. Milto n School of Industrial & Systems Engineering at Georgia Institute of Techn ology. Prior to joining Georgia Tech in 2013\, she was an assistant profe ssor in the Industrial & Enterprise Systems Engineering Department at the University of Illinois Urbana-Champaign from 2009-2013. She received the B .S. degree with highest honors in electrical engineering from Zhejiang Uni versity\, China\, in 2004\, and the Ph.D. degree in electrical engineering from the University of Maryland\, College Park\, in 2009. Her research in terests include stochastic control\, simulation optimization\, and Monte C arlo statistical methods. She is a recipient of the “Best Theoretical Pape r” award at the Winter Simulation Conference in 2009\, AFOSR Young Investi gator award in 2012\, and NSF CAREER award in 2015.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7463@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20161010T190000 DTEND;TZID=America/New_York:20161010T213000 SEQUENCE:0 SUMMARY:HUSAM Event: Global Hedge Fund Marshall Wace Recruiting Event URL:https://engineering.jhu.edu/ams/events/husam-event-global-hedge-fund-ma rshall-wace-recruiting-event/ X-COST-TYPE:free X-WP-IMAGES-URL:thumbnail\;http://engineering.jhu.edu/ams/wp-content/upload s/sites/44/2016/10/Husam-300x232.jpg\;730\;565\,medium\;http://engineering .jhu.edu/ams/wp-content/uploads/sites/44/2016/10/Husam-300x232.jpg\;730\;5 65\,large\;http://engineering.jhu.edu/ams/wp-content/uploads/sites/44/2016 /10/Husam-300x232.jpg\;730\;565\,full\;http://engineering.jhu.edu/ams/wp-c ontent/uploads/sites/44/2016/10/Husam-300x232.jpg\;730\;565 X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nModeling the dynamics of interacting particles by means of stochastic
networks

\n

\nMaterial science have been ra
pidly developing in recent years. A variety of particles interacting accor
ding to different kinds of pair potentials has been produced in experiment
al works. Looking into the future\, one can imagine controlled self-assemb
ly of particles into clusters of desired structures leading to the creatio
n of new types of materials. Analytical studies of the self-assembly invol
ve coping with difficulties associated with the huge numbers configuration
s\, high dimensionality\, complex geometry\, and unacceptably large CPU ti
mes. A feasible approach to the study of self-assembly consists of mapping
the collections of clusters onto stochastic networks (continuous-time Mar
kov chains) and analyzing their dynamics. Vertices of the networks represe
nt local minima of the potential energy of the clusters\, while arcs conne
ct only those pairs of vertices that correspond to local minima between wh
ich direct transitions are physically possible. Transition rates along the
arcs are the transition rates between the corresponding pairs of local mi
nima. Such networks are mathematically tractable and\, at the same time\,
preserve important features of the underlying dynamics. Nevertheless\, the
ir huge size and complexity render their analysis challenging and invoke t
he development of new mathematical techniques. I will discuss some approac
hes to construction and analysis of such networks.

Title:

\n< p>Leveraged Funds: Robust Replication and Performance Evaluation\nA bstract:

\nLeveraged and inverse ETFs seek a daily return equal to a multiple of an index’ return\, an objective that requires continuous port folio rebalancing. The resulting trading costs create a tradeoff between t racking error\, which controls the short-term correlation with the index\, and excess return (or tracking difference) – the long-term deviation from the leveraged index’ performance. With proportional trading costs\, the o ptimal replication policy is robust to the index’ dynamics. A summary of a fund’s performance is the implied spread\, equal to the product of tracki ng error and excess return\, rescaled for leverage and average volatility. The implied spread is insensitive to the benchmark’s risk premium and off ers a tool to compare the performance of funds tracking the same index wit h different factors and tracking errors.

\n\n

http://ssrn.com /abstract=2839852

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7300@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20161026T150000 DTEND;TZID=America/New_York:20161026T160000 SEQUENCE:0 SUMMARY:Data Seminar: Robert Pego (Carnegie Mellon University) @ Krieger 30 9 URL:https://engineering.jhu.edu/ams/events/data-seminar-robert-pego-carnegi e-mellon-university-krieger-309/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-6963@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Adaptive Contrast Weighted Learning and Tree-based Reinforcemen t Learning for Multi-Stage Multi-Treatment Decision-Making\nDynamic treatm ent regimes (DTRs) are sequential decision rules that focus simultaneously on treatment individualization and adaptation over time. We develop robus t and flexible semiparametric and machine learning methods for estimating optimal DTRs. In this talk\, we present a dynamic statistical learning met hod\, adaptive contrast weighted learning (ACWL)\, which combines doubly r obust semiparametric regression estimators with flexible machine learning methods. ACWL can handle multiple treatments at each stage and does not re quire prespecifying candidate DTRs. At each stage\, we develop robust semi parametric regression-based contrasts with the adaptation of treatment eff ect ordering for each patient\, and the adaptive contrasts simplify the pr oblem of optimization with multiple treatment comparisons to a weighted cl assification problem that can be solved with existing machine learning tec hniques. We further develop a tree-based reinforcement learning (T-RL) met hod to directly estimate optimal DTRs in a multi-stage multi-treatment set ting. At each stage\, T-RL builds an unsupervised decision tree that maint ains the nature of batch-mode reinforcement learning. Unlike ACWL\, T-RL h andles the optimization problem with multiple treatment comparisons direct ly through the purity measure constructed with augmented inverse probabili ty weighted estimators. By combining robust semiparametric regression with flexible tree-based learning\, T-RL is robust\, efficient and easy to int erpret for the identification of optimal DTRs. However\, ACWL seems more r obust to tree-type misspecification than T-RL when the true optimal DTR is non-tree-type. We illustrate the performances of both methods in simulati ons and case studies. DTSTART;TZID=America/New_York:20161027T133000 DTEND;TZID=America/New_York:20161027T143000 SEQUENCE:0 SUMMARY:Seminar: Lu Wang (University of Michigan) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-lu-wang-university-of-mi chigan-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Adapt
ive Contrast Weighted Learning and Tree-based Reinforcement Learning for M
ulti-Stage Multi-Treatment Decision-Making**

Dynamic treatm ent regimes (DTRs) are sequential decision rules that focus simultaneously on treatment individualization and adaptation over time. We develop robus t and flexible semiparametric and machine learning methods for estimating optimal DTRs. In this talk\, we present a dynamic statistical learning met hod\, adaptive contrast weighted learning (ACWL)\, which combines doubly r obust semiparametric regression estimators with flexible machine learning methods. ACWL can handle multiple treatments at each stage and does not re quire prespecifying candidate DTRs. At each stage\, we develop robust semi parametric regression-based contrasts with the adaptation of treatment eff ect ordering for each patient\, and the adaptive contrasts simplify the pr oblem of optimization with multiple treatment comparisons to a weighted cl assification problem that can be solved with existing machine learning tec hniques. We further develop a tree-based reinforcement learning (T-RL) met hod to directly estimate optimal DTRs in a multi-stage multi-treatment set ting. At each stage\, T-RL builds an unsupervised decision tree that maint ains the nature of batch-mode reinforcement learning. Unlike ACWL\, T-RL h andles the optimization problem with multiple treatment comparisons direct ly through the purity measure constructed with augmented inverse probabili ty weighted estimators. By combining robust semiparametric regression with flexible tree-based learning\, T-RL is robust\, efficient and easy to int erpret for the identification of optimal DTRs. However\, ACWL seems more r obust to tree-type misspecification than T-RL when the true optimal DTR is non-tree-type. We illustrate the performances of both methods in simulati ons and case studies.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7304@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION:Variational problems on graphs and their continuum limits\n\nWe will discuss variational problems arising in machine learning and their l imits as the number of data points goes to infinity. Consider point clouds obtained as random samples of an underlying “ground-truth” measure. Graph representing the point cloud is obtained by assigning weights to edges ba sed on the distance between the points. Many machine learning tasks\, such as clustering and classification\, can be posed as minimizing functionals on such graphs. We consider functionals involving graph cuts and graph la placians and their limits as the number of data points goes to infinity. I n particular we establish under what conditions the minimizers of discrete problems have a well defined continuum limit\, and characterize the limit . The talk is primarily based on joint work with Nicolas Garcia Trillos\, as well as on works with Xavier Bresson\, Moritz Gerlach\, Matthias Hein\, Thomas Laurent\, James von Brecht and Matt Thorpe. DTSTART;TZID=America/New_York:20161102T150000 DTEND;TZID=America/New_York:20161102T160000 SEQUENCE:0 SUMMARY:Data Seminar: Dejan Slepcev (Carnegie Mellon University) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-dejan-slepcev-carne gie-mellon-university-krieger-309/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nVariational problems on graphs and their continuum limits

\n

\nWe will discuss variational problems arising in ma
chine learning and their limits as the number of data points goes to infin
ity. Consider point clouds obtained as random samples of an underlying “gr
ound-truth” measure. Graph representing the point cloud is obtained by ass
igning weights to edges based on the distance between the points. Many mac
hine learning tasks\, such as clustering and classification\, can be posed
as minimizing functionals on such graphs. We consider functionals involvi
ng graph cuts and graph laplacians and their limits as the number of data
points goes to infinity. In particular we establish under what conditions
the minimizers of discrete problems have a well defined continuum limit\,
and characterize the limit. The talk is primarily based on joint work with
Nicolas Garcia Trillos\, as well as on works with Xavier Bresson\, Moritz
Gerlach\, Matthias Hein\, Thomas Laurent\, James von Brecht and Matt Thor
pe.

An Introducti on to Distance Preserving Projections of Smooth Manifolds

\n\n< p>Manifold-based image models are assumed in many engineering applications involving imaging and image classification. In the setting of image clas sification\, in particular\, proposed designs for small and cheap cameras motivate compressive imaging applications involving manifolds. Interestin g mathematics results when one considers that the problem one needs to sol ve in this setting ultimately involves questions concerning how well one c an embed a low-dimensional smooth sub-manifold of high-dimensional Euclide an space into a much lower dimensional space without knowing any of its de tailed structure. We will motivate this problem and discuss how one might accomplish this seemingly difficult task using random projections. Littl e if any prerequisites will be assumed beyond linear algebra and some prob ability.\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7308@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION:Scalable Information Inequalities for Uncertainty Quantificatio n in high dimensional probabilistic models\n\nIn this this talk we discuss new scalable information bounds for quantities of interest of complex sto chastic models. The scalability of inequalities allows us to (a) obtain un certainty quantification bounds for quantities of interest in high-dimensi onal systems and/or for long time stochastic dynamics\; (b) assess the imp act of large model perturbations such as in nonlinear response regimes in statistical mechanics\; (c) address model-form uncertainty\, i.e. compare different extended probabilistic models and corresponding quantities of in terest. We demonstrate these tools in fast sensitivity screening of chemic al reaction networks with a very large number of parameters\, and towards obtaining robust and tight uncertainty quantification bounds for phase dia grams in statistical mechanics models. DTSTART;TZID=America/New_York:20161109T150000 DTEND;TZID=America/New_York:20161109T160000 SEQUENCE:0 SUMMARY:Data Seminar: Markos Katsoulakis (University of Massachusetts Amher st) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-markos-katsoulakis- university-massachusetts-amherst-krieger-309/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Scalable Information Inequalities for Uncertainty Quantification in hi
gh dimensional probabilistic models

\n

\nIn
this this talk we discuss new scalable information bounds for quantities
of interest of complex stochastic models. The scalability of inequalities
allows us to (a) obtain uncertainty quantification bounds for quantities o
f interest in high-dimensional systems and/or for long time stochastic dyn
amics\; (b) assess the impact of large model perturbations such as in nonl
inear response regimes in statistical mechanics\; (c) address model-form u
ncertainty\, i.e. compare different extended probabilistic models and corr
esponding quantities of interest. We demonstrate these tools in fast sensi
tivity screening of chemical reaction networks with a very large number of
parameters\, and towards obtaining robust and tight uncertainty quantific
ation bounds for phase diagrams in statistical mechanics models.

**Slipp
ing Through the Cracks: Detecting Manipulation in Regional Commodity Marke
ts**

\n

Reid B. Stevens[1] and Jeffery Y. Zhang[2]\n

\n

Between 2010 and 2014\, the regional price of aluminum in
the United States (Midwest premium) increased 400 percent. We argue that
the Midwest premium was likely manipulated during this period through the
exercise of market power in the aluminum storage market. We first use a di
fference-in-differences model to show that there was a statistically signi
ficant increase of $0.07 per pound in the regional price of aluminum relat
ive to the regional price of a production complement\, copper. We then us
e several instrumental variables to show that this increase was driven by
a single financial company’s accumulation of an unprecedented level of alu
minum inventories in Detroit. Since this scheme targeted the regional pri
ce of aluminum\, regulators who monitored only spot and futures prices wou
ld not have noticed anything peculiar. We therefore present an algorithm f
or real-time detection of similar manipulation schemes in regional commodi
ty markets. The algorithm confirms the existence of a structural break in
the U.S. aluminum market in late 2011. Using the algorithm\, regulators c
ould have detected the scheme as early as December 2012\, more than six mo
nths before it was publicized by an article in *The New York Times*
. We also apply the algorithm to another suspected case of regional price
manipulation in the European aluminum market and find a similar break in 2
011\, suggesting the scheme may have been implemented beyond the United St
ates.

[1] Department of Agric ulture Economics\, Texas A&M University\, stevens@tamu.edu

\n[2] Department of Economics\, Yale Universit y and Harvard Law School\, jeffery.zhang@yale.edu

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7312@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20161130T150000 DTEND;TZID=America/New_York:20161130T160000 SEQUENCE:0 SUMMARY:Data Seminar: Youssef Marzouk (Massachusetts Institute of Technolog y) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-youssef-marzouk-mas sachusetts-institute-technology-shaffer-100/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-8216@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Spectral Clustering for Dynamic Stochastic Block Model\n \nAbstract: One of the most common and crucial aspects of many network da ta sets is the dependence of network link structure on time. In this work\ , we extend the existing (static) nonparametric latent variable model in t he context of time-varying networks\, and thereby propose a class of dyna mic network models. For some special cases of these models (namely the dyn amic stochastic block model and dynamic degree corrected block model)\, wh ich assume that there is a common clustering structure for all networks\, we consider the problem of identifying the common clustering structure. We propose two extensions of the (standard) spectral clustering method for t he dynamic network models\, and give theoretical guarantee that the spectr al clustering methods produce consistent community detection in case of bo th dynamic stochastic block model and dynamic degree-corrected block model . The methods are shown to work under sufficiently mild conditions on the number of time snapshots to detect both associative and dissociative comm unity structure\, even if all the individual networks are very sparse and most of the individual networks are below community detectability thresho ld. We reinforce the validity of the theoretical results via simulations t oo.\n(Joint work with Shirshendu Chatterjee\, CUNY) DTSTART;TZID=America/New_York:20161201T133000 DTEND;TZID=America/New_York:20161201T143000 SEQUENCE:0 SUMMARY:Seminar: Sharmodeep Bhattacharyya (Oregon State University) @ White head 304 URL:https://engineering.jhu.edu/ams/events/seminar-sharmodeep-bhattacharyya -oregon-state-university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Spectr al Clustering for Dynamic Stochastic Block Model

\n\n

Abstrac t: One of the most common and crucial aspects of many network data sets is the dependence of network link structure on time. In this work\, we exten d the existing (static) nonparametric latent variable model in the context of time-varying networks\, and thereby propose a class of dynamic networ k models. For some special cases of these models (namely the dynamic stoch astic block model and dynamic degree corrected block model)\, which assume that there is a common clustering structure for all networks\, we conside r the problem of identifying the common clustering structure. We propose t wo extensions of the (standard) spectral clustering method for the dynamic network models\, and give theoretical guarantee that the spectral cluster ing methods produce consistent community detection in case of both dynamic stochastic block model and dynamic degree-corrected block model. The met hods are shown to work under sufficiently mild conditions on the number of time snapshots to detect both associative and dissociative community stru cture\, even if all the individual networks are very sparse and most of t he individual networks are below community detectability threshold. We rei nforce the validity of the theoretical results via simulations too.

\n< p>(Joint work with Shirshendu Chatterjee\, CUNY)\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7316@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20161207T150000 DTEND;TZID=America/New_York:20161207T160000 SEQUENCE:0 SUMMARY:Data Seminar: Ben Adcock (Simons Fraser University) @ Krieger 309 URL:https://engineering.jhu.edu/ams/events/data-seminar-ben-adcock-simons-f raser-university-shaffer-100/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-8240@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Online and Random-order Load Balancing Simultaneously\nA bstract: We consider the problem of online load balancing under lp-norms: sequential jobs need to be assigned to one of the machines and the goal is to minimize the lp-norm of the machine loads. This generalizes the classi cal problem of scheduling for makespan minimization (case l_infty) and has been thoroughly studied. We provide algorithms with simultaneously optima l* guarantees for the worst-case model as well as for the random-order (i. e. secretary) model\, where an arbitrary set of jobs comes in random order .\nOne of the main components is a new algorithm with improved regret for Online Linear Optimization (OLO) over the non-negative vectors in the lq b all. Interestingly\, this OLO algorithm is also used to prove a purely pro babilistic inequality that controls the correlations arising in the random -order model\, a common source of difficulty for the analysis. A property that drives both our load balancing algorithms and our OLO algorithm is a smoothing of the the lp-norm that may be of independent interest. DTSTART;TZID=America/New_York:20161208T133000 DTEND;TZID=America/New_York:20161208T143000 SEQUENCE:0 SUMMARY:Seminar: Marco Molinaro (Pontifical Catholic University of Rio de J aneiro) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-marco-molinaro-universit y-california-davis-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Online and Random-order Load Balancing Simultaneously

\nAbstract: We consi der the problem of online load balancing under lp-norms: sequential jobs n eed to be assigned to one of the machines and the goal is to minimize the lp-norm of the machine loads. This generalizes the classical problem of sc heduling for makespan minimization (case l_infty) and has been thoroughly studied. We provide algorithms with simultaneously optimal* guarantees for the worst-case model as well as for the random-order (i.e. secretary) mod el\, where an arbitrary set of jobs comes in random order.

\nOne of the main components is a new algorithm with improved regret for Online Lin ear Optimization (OLO) over the non-negative vectors in the lq ball. Inter estingly\, this OLO algorithm is also used to prove a purely probabilistic inequality that controls the correlations arising in the random-order mod el\, a common source of difficulty for the analysis. A property that drive s both our load balancing algorithms and our OLO algorithm is a smoothing of the the lp-norm that may be of independent interest.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-8220@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Spatial-temporal modeling of the association between air pollut ion exposures and birth outcomes: identifying critical exposure windows\nE xposure to high levels of air pollution during pregnancy has been linked t o increased probability of adverse birth outcomes. We consider statistical models for evaluating associations between pollutants and birth outcomes\ , taking into account multipollutant exposures\, susceptible windows in pr egnancy\, and variability in exposure over space and time. We consider geo coded vital records data from Texas as well as data from the National Birt h Defects Prevention Study. DTSTART;TZID=America/New_York:20170202T133000 DTEND;TZID=America/New_York:20170202T143000 SEQUENCE:0 SUMMARY:Wierman Lecture Series: Amy Herring (University of North Carolina C hapel Hill) @ Hodson 110 URL:https://engineering.jhu.edu/ams/events/wierman-lecture-series-amy-herri ng-university-north-carolina-chapel-hill-tba/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Spati
al-temporal modeling of the association between air pollution exposures an
d birth outcomes: identifying critical exposure windows**

E xposure to high levels of air pollution during pregnancy has been linked t o increased probability of adverse birth outcomes. We consider statistical models for evaluating associations between pollutants and birth outcomes\ , taking into account multipollutant exposures\, susceptible windows in pr egnancy\, and variability in exposure over space and time. We consider geo coded vital records data from Texas as well as data from the National Birt h Defects Prevention Study.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9274@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20170208T150000 DTEND;TZID=America/New_York:20170208T160000 SEQUENCE:0 SUMMARY:Data Seminar: Kasso Okoudjou (University of Maryland\, College Park ) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/data-seminar-kasso-okoudjou-univ ersity-maryland-college-park-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7404@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Geometry\, Shapes and PDEs\n \nAbstract:\nThe interest i n right invariant metrics on the diffeomorphism group is fueled by its rel ations to hydrodynamics. Arnold noted in 1966 that Euler’s equations\, whi ch govern the motion of ideal\, incompressible fluids\, can be interpreted as geodesic equations on the group of volume preserving diffeomorphisms w ith respect to a suitable Riemannian metric. Since then other PDEs arising in physics have been interpreted as geodesic equations on manifold of map pings. Examples include Burgers’ equation\, the KdV and Camassa-Holm equat ions or the Hunter-Saxton equation.\nAnother important motivation for the study of Riemannian metrics on manifold of mappings can be found in its a ppearance in the field of shape analysis and in particular in the eminent role of the diffeomorphism group in computational anatomy: the space of me dical images is acted upon by the diffeomorphism group and differences bet ween images are encoded by diffeomorphisms in the spirit of Grenander’s pa ttern theory. The study of anatomical shapes can be thus reduced to the st udy of the diffeomorphism group.\nUsing these observations as a starting p oint\, I will consider Riemannian metrics on spaces of mappings. I will di scuss the local and global well-posedness of the corresponding geodesic eq uation\, study the induced geodesic distance and present selected numerica l examples of minimizing geodesics. DTSTART;TZID=America/New_York:20170209T133000 DTEND;TZID=America/New_York:20170209T143000 SEQUENCE:0 SUMMARY:Seminar: Martin Bauer (Florida State University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-martin-bauer-florida-sta te-university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Title: Geomet ry\, Shapes and PDEs

\n\n

Abstract:

\nThe interest in r ight invariant metrics on the diffeomorphism group is fueled by its relati ons to hydrodynamics. Arnold noted in 1966 that Euler’s equations\, which govern the motion of ideal\, incompressible fluids\, can be interpreted as geodesic equations on the group of volume preserving diffeomorphisms with respect to a suitable Riemannian metric. Since then other PDEs arising in physics have been interpreted as geodesic equations on manifold of mappin gs. Examples include Burgers’ equation\, the KdV and Camassa-Holm equation s or the Hunter-Saxton equation.

\nAnother important motivation for the study of Riemannian metrics on manifold of mappings can be found in i ts appearance in the field of shape analysis and in particular in the emin ent role of the diffeomorphism group in computational anatomy: the space o f medical images is acted upon by the diffeomorphism group and differences between images are encoded by diffeomorphisms in the spirit of Grenander’ s pattern theory. The study of anatomical shapes can be thus reduced to th e study of the diffeomorphism group.

\nUsing these observations as a starting point\, I will consider Riemannian metrics on spaces of mappings . I will discuss the local and global well-posedness of the corresponding geodesic equation\, study the induced geodesic distance and present select ed numerical examples of minimizing geodesics.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9278@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20170215T150000 DTEND;TZID=America/New_York:20170215T160000 SEQUENCE:0 SUMMARY:Data Seminar: Jerome Darbon (Brown University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/data-seminar-jerome-darbon-brown -university-whitehead-304/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-8440@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:TITLE:\nA local limit theorem for QuickSort key comparisons via multi-round smoothing\nABSTRACT:\nIt is a well-known result\, due indepen dently to Régnier (1989) and Rösler (1991)\, that the number of key compar isons required by the randomized sorting algorithm QuickSort to sort a lis t of n distinct items (keys) satisfies a global distributional limit the orem. We resolve an open problem of Fill and Janson from 2002 by using a multi-round smoothing technique to establish the corresponding local limit theorem. (in plain text\; note that only the “n” in “sort a list of n d istinct items” would be set in math mode in LaTeX)\nThis is joint work wit h Béla Bollobás and Oliver Riordan. DTSTART;TZID=America/New_York:20170216T133000 DTEND;TZID=America/New_York:20170216T143000 SEQUENCE:0 SUMMARY:Seminar: Jim Fill (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-jim-fill-jhu-whitehead-3 04/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**TITLE
:**

A local limit theorem for QuickSort key comparisons via multi-round smoothing

\n**ABSTRACT:**

It is a well-known result\, due independently to Régnier (1989) and Rösler (1991)\ , that the number of key comparisons required by the randomized sorting al gorithm QuickSort to sort a list of n distinct items (keys) satisfies a global distributional limit theorem. We resolve an open problem of Fill a nd Janson from 2002 by using a multi-round smoothing technique to establis h the corresponding local limit theorem. (in plain text\; note that only t he “n” in “sort a list of n distinct items” would be set in math mode in LaTeX)

\nThis is joint work with Béla Bollobás and Oliver Riordan.< /p>\n END:VEVENT BEGIN:VEVENT UID:ai1ec-8420@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20170223T133000 DTEND;TZID=America/New_York:20170223T143000 SEQUENCE:0 SUMMARY:Seminar: Laurent Younes (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-laurent-younes-jhu-white head-304/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-9442@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Link to the slides from Tom Loredo’s seminar- JHU17-HierBayesCo smicPopns DTSTART;TZID=America/New_York:20170302T133000 DTEND;TZID=America/New_York:20170302T143000 SEQUENCE:0 SUMMARY:Seminar: Tom Loredo (Cornell University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-tom-loredo-cornell-unive rsity-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Link to the s lides from Tom Loredo’s seminar- JHU17- HierBayesCosmicPopns

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9470@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20170307T130000 DTEND;TZID=America/New_York:20170307T140000 LOCATION:Hodson 3rd Floor Lobby SEQUENCE:0 SUMMARY:HUSAM Presents Coffee Chat URL:https://engineering.jhu.edu/ams/events/husam-presents-coffee-chat/ X-COST-TYPE:free X-WP-IMAGES-URL:thumbnail\;https://engineering.jhu.edu/ams/wp-content/uploa ds/sites/44/2017/03/Coffee_Chat-002-300x139.jpg\;300\;139\,medium\;https:/ /engineering.jhu.edu/ams/wp-content/uploads/sites/44/2017/03/Coffee_Chat-0 02-300x139.jpg\;300\;139\,large\;https://engineering.jhu.edu/ams/wp-conten t/uploads/sites/44/2017/03/Coffee_Chat-002-300x139.jpg\;300\;139\,full\;ht tps://engineering.jhu.edu/ams/wp-content/uploads/sites/44/2017/03/Coffee_C hat-002-300x139.jpg\;300\;139 X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Mean Field Games: theory and applications

\n\n

Abstract: We review the Mean Field Game paradigm introduced independently by Caines-Huang-Mal hame and Lasry-Lyons ten years ago\, and we illustrate their relevance to applications with a few practical of examples (bird flocking\, room exit\, systemic risk\, cyber-security\, …. ). We then review the probabilistic a pproach based on Forward-Backward Stochastic Differential Equations\, and we derive the Master Equation from a version of the chain rule (Ito’s for mula) for functions over flows of probability measures. Finally\, motivate d by the literature on economic models of bank runs\, we introduce mean fi eld games of timing and discuss new results\, and some of the many remaini ng challenges.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9266@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Mean Field Games with Major and Minor Players: Theory an d Numerics.\n \nAbstract: We present a (possibly) new formulation of the m ean field game problem in the presence of major and minor players\, and gi ve new existence results for linear quadratic models and models with finit e state spaces. We shall also provide numerical results illustrating the t heory and raising new challenging open problems. DTSTART;TZID=America/New_York:20170310T133000 DTEND;TZID=America/New_York:20170310T143000 SEQUENCE:0 SUMMARY:Duncan Lecture Series: Rene Carmona (Princeton) @ Krieger 205 URL:https://engineering.jhu.edu/ams/events/duncan-lecture-series-rene-carmo na-princeton-tba-2/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Mean F ield Games with Major and Minor Players: Theory and Numerics.

\n\n

Abstract: We present a (possibly) new formulation of the mean field game problem in the presence of major and minor players\, and give new exi stence results for linear quadratic models and models with finite state sp aces. We shall also provide numerical results illustrating the theory and raising new challenging open problems.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9483@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:HUSAM is hosting a professional event with Deloitte. The event will be an overview of consulting at Deloitte. A panel of Deloitte pract itioners will present on Deloitte’s BTA consulting track\, health analytic s\, and answer questions. DTSTART;TZID=America/New_York:20170313T183000 LOCATION:Arellano Theater in Levering Hall SEQUENCE:0 SUMMARY:HUSAM Deloitte Event URL:https://engineering.jhu.edu/ams/events/husam-deloitte-event/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nHUSAM is host ing a professional event with Deloitte. The event will be an overview of consulting at Deloitte. A panel of Deloitte practitioners will present on Deloitte’s BTA consulting track\, health analytics\, and answer questions .

\n X-INSTANT-EVENT:1 END:VEVENT BEGIN:VEVENT UID:ai1ec-9258@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Nuke the Clouds: Using nuclear norm optimization to remo ve clouds from satellite images\nAbstract: We discuss how to use the nucle ar norm and matrix factorization techniques to remove clouds from satellit e images. The talk will focus on discussing the key properties and variat ional inequalities that is commonly used in minimizing convex functions wi th a nuclear norm term. We will also contrast the convex formulations wit h the corresponding rank constrained problems that are highly non-convex\, but which are sometimes simpler to solve regardless. Finally we will sho w a lot of examples/demos of how this is working in practice. DTSTART;TZID=America/New_York:20170316T133000 DTEND;TZID=America/New_York:20170316T143000 SEQUENCE:0 SUMMARY:Seminar: Peder Olsen (IBM) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-peder-olsen-ibm-whitehea d-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Nuke t he Clouds: Using nuclear norm optimization to remove clouds from satellite images

\nAbstract: We discuss how to use the nuclear norm and matri x factorization techniques to remove clouds from satellite images. The ta lk will focus on discussing the key properties and variational inequalitie s that is commonly used in minimizing convex functions with a nuclear norm term. We will also contrast the convex formulations with the correspondi ng rank constrained problems that are highly non-convex\, but which are so metimes simpler to solve regardless. Finally we will show a lot of exampl es/demos of how this is working in practice.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9290@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20170329T150000 DTEND;TZID=America/New_York:20170329T160000 SEQUENCE:0 SUMMARY:Data Seminar: Jason Eisner (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/data-seminar-jason-eisner-jhu-wh itehead-304/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-8224@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Heuristics for Network Revenue Management\nWe consider a networ k revenue management problem with customer choice and exogenous prices. Su ch problems are central in several applications including airline ticket p ricing. Given the infeasibility of explicitly finding optimal policies\, w e study the performance of a class of heuristic policies. These heuristics periodically re-solve the deterministic linear program (DLP) that results when all future random variables are replaced by their average values and implement the solutions in a probabilistic manner. We provide an upper bo und for the expected revenue loss under such policies when compared to the optimal policy. Using this bound\, we construct a schedule of re-solving times such that the resulting expected revenue loss is bounded by a consta nt that is independent of the size of the problem.\nJoint work with Stefan us Jasin at University of Michigan.\n DTSTART;TZID=America/New_York:20170330T133000 DTEND;TZID=America/New_York:20170330T143000 SEQUENCE:0 SUMMARY:Seminar: Sunil Kumar (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-sunil-kumar-jhu-whitehea d-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nHeuristics fo r Network Revenue Management

\nWe consider a network revenue managem ent problem with customer choice and exogenous prices. Such problems are c entral in several applications including airline ticket pricing. Given the infeasibility of explicitly finding optimal policies\, we study the perfo rmance of a class of heuristic policies. These heuristics periodically re- solve the deterministic linear program (DLP) that results when all future random variables are replaced by their average values and implement the so lutions in a probabilistic manner. We provide an upper bound for the expec ted revenue loss under such policies when compared to the optimal policy. Using this bound\, we construct a schedule of re-solving times such that t he resulting expected revenue loss is bounded by a constant that is indepe ndent of the size of the problem.

\nJoint work with Stefanus Jasin a t University of Michigan.

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-7408@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Energy Prices & Dynamic Games with Stochastic Demand\nThe drama tic decline in oil prices\, from around $110 per barrel in June 2014 to ar ound $30 in January 2016 highlights the importance of competition between different energy producers. Indeed\, the price drop has been primarily at tributed to OPEC’s strategic decision (until very recently) not to curb it s oil production in the face of increased supply of shale gas and oil in t he US\, which was spurred by the development of fracking technology. Most dynamic Cournot models focus on supply-side factors\, such as increased sh ale oil\, and random discoveries. However declining and uncertain demand f rom China is a major factor driving oil price volatility. We study Cournot games in a stochastic demand environment\, and present asymptotic and num erical results\, as well as a modified Hotelling’s rule for games with sto chastic demand. DTSTART;TZID=America/New_York:20170403T103000 DTEND;TZID=America/New_York:20170403T113000 SEQUENCE:0 SUMMARY:Seminar: Ronnie Sircar (Princeton University) @ Ames 234 URL:https://engineering.jhu.edu/ams/events/seminar-ronnie-sircar-princeton- university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Energy Prices & Dynamic Games with Stochastic Demand

\nThe dramatic decline in oi l prices\, from around $110 per barrel in June 2014 to around $30 in Janua ry 2016 highlights the importance of competition between different energy producers. Indeed\, the price drop has been primarily attributed to OPEC’ s strategic decision (until very recently) not to curb its oil production in the face of increased supply of shale gas and oil in the US\, which was spurred by the development of fracking technology. Most dynamic Cournot m odels focus on supply-side factors\, such as increased shale oil\, and ran dom discoveries. However declining and uncertain demand from China is a ma jor factor driving oil price volatility. We study Cournot games in a stoch astic demand environment\, and present asymptotic and numerical results\, as well as a modified Hotelling’s rule for games with stochastic demand.\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9310@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20170405T150000 DTEND;TZID=America/New_York:20170405T160000 SEQUENCE:0 SUMMARY:Data Seminar: Afonso Bandeira (NYU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/data-seminar-afonso-bandeira-nyu -whitehead-304/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-9478@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: DTSTART;TZID=America/New_York:20170405T150000 DTEND;TZID=America/New_York:20170405T160000 SEQUENCE:0 SUMMARY:HUSAM Presents Coffee Chat URL:https://engineering.jhu.edu/ams/events/husam-presents-coffee-chat-2/ X-COST-TYPE:free X-WP-IMAGES-URL:thumbnail\;https://engineering.jhu.edu/ams/wp-content/uploa ds/sites/44/2017/03/Coffee_Chat-002-300x139.jpg\;300\;139\,medium\;https:/ /engineering.jhu.edu/ams/wp-content/uploads/sites/44/2017/03/Coffee_Chat-0 02-300x139.jpg\;300\;139\,large\;https://engineering.jhu.edu/ams/wp-conten t/uploads/sites/44/2017/03/Coffee_Chat-002-300x139.jpg\;300\;139\,full\;ht tps://engineering.jhu.edu/ams/wp-content/uploads/sites/44/2017/03/Coffee_C hat-002-300x139.jpg\;300\;139 X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Reciprocal Graphical Models for Integrative Gene Regulatory Network Analysis

\nConstructing gene regula tory networks is a fundamental task in systems biology. We introduce a Gau ssian reciprocal graphical model for inference about gene regulatory relat ionships by integrating mRNA gene expression and DNA level information inc luding copy number and methylation. Data integration allows for inference on the directionality of certain regulatory relationships\, which would be otherwise indistinguishable due to Markov equivalence. Efficient inferenc e is developed based on simultaneous equation models. Bayesian model selection techniques are adopted to estimate t he graph structure. We illustrate our approach by simulations and two appl ications in ZODIAC pairwise gene interaction analysis and colon adenocarci noma pathway analysis.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-8444@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:From solving PDEs to machine learning PDEs: \nAn odyssey in com putational mathematics\nGeorge Em Karniadakis\nDivision of Applied Mathema tics\, Brown University\nAbstract: In the last 30 years I have pursued the numerical solution of partial differential equations (PDEs) using spectra l and spectral elements methods for diverse applications\, starting from d eterministic PDEs in complex geometries\, to stochastic PDEs for uncertain ty quantification\, and to fractional PDEs that describe non-local behavio r in disordered media and viscoelastic materials. More recently\, I have b een working on solving PDEs in a fundamentally different way. I will prese nt a new paradigm in solving linear and nonlinear PDEs from noisy measurem ents without the use of the classical numerical discretization. Instead\, we infer the solution of PDEs from noisy data\, which can represent measur ements of variable fidelity. The key idea is to encode the structure of th e PDE into prior distributions and train Bayesian nonparametric regression models on available noisy data. The resulting posterior distributions can be used to predict the PDE solution with quantified uncertainty\, efficie ntly identify extrema via Bayesian optimization\, and acquire new data via active learning. Moreover\, I will present how we can use this new framew ork to learn PDEs from noisy measurements of the solution and the forcing terms.\n \nBio: George Karniadakis received his S.M. and Ph.D. from Massac husetts Institute of Technology. He was appointed Lecturer in the Departme nt of Mechanical Engineering at MIT in 1987 and subsequently he joined the Center for Turbulence Research at Stanford / Nasa Ames. He joined Princet on University as Assistant Professor in the Department of Mechanical and A erospace Engineering and as Associate Faculty in the Program of Applied an d Computational Mathematics. He was a Visiting Professor at Caltech in 199 3 in the Aeronautics Department and joined Brown University as Associate P rofessor of Applied Mathematics in the Center for Fluid Mechanics in 1994. After becoming a full professor in 1996\, he continues to be a Visiting P rofessor and Senior Lecturer of Ocean/Mechanical Engineering at MIT. He is a Fellow of the Society for Industrial and Applied Mathematics (SIAM\, 20 10-)\, Fellow of the American Physical Society (APS\, 2004-)\, Fellow of t he American Society of Mechanical Engineers (ASME\, 2003-) and Associate F ellow of the American Institute of Aeronautics and Astronautics (AIAA\, 20 06-). He received the Ralf E Kleinman award from SIAM (2015)\, the J. Tins ley Oden Medal (2013)\, and the CFD award (2007) by the US Association in Computational Mechanics. His h-index is 79 and he has been cited over 32\, 500 times.\n DTSTART;TZID=America/New_York:20170420T133000 DTEND;TZID=America/New_York:20170420T143000 SEQUENCE:0 SUMMARY:Seminar: George Karniadakis (Brown University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-george-karniadakis-brown -university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**From
solving PDEs to machine learning PDEs: **

**An odysse
y in computational mathematics**

*George Em Karn
iadakis*

*Division of Applied Mathematics\
, Brown University*

**Abstract:** In the
last 30 years I have pursued the numerical solution of partial differenti
al equations (PDEs) using spectral and spectral elements methods for diver
se applications\, starting from deterministic PDEs in complex geometries\,
to stochastic PDEs for uncertainty quantification\, and to fractional PDE
s that describe non-local behavior in disordered media and viscoelastic ma
terials. More recently\, I have been working on solving PDEs in a fundamen
tally different way. I will present a new paradigm in solving linear and n
onlinear PDEs from noisy measurements without the use of the classical num
erical discretization. Instead\, we infer the solution of PDEs from noisy
data\, which can represent measurements of variable fidelity. The key idea
is to encode the structure of the PDE into prior distributions and train
Bayesian nonparametric regression models on available noisy data. The resu
lting posterior distributions can be used to predict the PDE solution with
quantified uncertainty\, efficiently identify extrema via Bayesian optimi
zation\, and acquire new data via active learning. Moreover\, I will prese
nt how we can use this new framework to learn PDEs from noisy measurements
of the solution and the forcing terms.

\n

**Bio: George Karniadakis received his S.M. and Ph.D. from Massachusetts In
stitute of Technology. He was appointed Lecturer in the Department of Mech
anical Engineering at MIT in 1987 and subsequently he joined the Center fo
r Turbulence Research at Stanford / Nasa Ames. He joined Princeton Univers
ity as Assistant Professor in the Department of Mechanical and Aerospace E
ngineering and as Associate Faculty in the Program of Applied and Computat
ional Mathematics. He was a Visiting Professor at Caltech in 1993 in the A
eronautics Department and joined Brown University as Associate Professor o
f Applied Mathematics in the Center for Fluid Mechanics in 1994. After bec
oming a full professor in 1996\, he continues to be a Visiting Professor a
nd Senior Lecturer of Ocean/Mechanical Engineering at MIT. He is a Fellow
of the Society for Industrial and Applied Mathematics (SIAM\, 2010-)\, Fel
low of the American Physical Society (APS\, 2004-)\, Fellow of the America
n Society of Mechanical Engineers (ASME\, 2003-) and Associate Fellow of t
he American Institute of Aeronautics and Astronautics (AIAA\, 2006-). He r
eceived the Ralf E Kleinman award from SIAM (2015)\, the J. Tinsley Oden M
edal (2013)\, and the CFD award (2007) by the US Association in Computatio
nal Mechanics. His h-index is 79 and he has been cited over 32\,500 times.
**

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9298@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Solving Fredholm integrals from incomplete measurements\nWe pre sent an algorithm to solve Fredholm integrals of the first kind with tenso r product structures\, from a limited number of measurements with the goal of using this method to accelerate Nuclear Magnetic Resonance (NMR) acqui sition. This is done by incorporating compressive sampling type arguments to fill in the missing measurements using a priori knowledge of the struct ure of the data. In the first step\, we recover a compressed data matrix f rom measurements that form a tight frame\, and establish that these measur ements satisfy the restricted isometry property (RIP). In the second step\ , we solve the zeroth-order regularization minimization problem using the Venkataramanan-Song-Huerlimann algorithm. We demonstrate the performance o f this algorithm on simulated and real data and we compare it with other s ampling techniques. Our theory applied to both 2D and multidimensional NMR . DTSTART;TZID=America/New_York:20170426T150000 DTEND;TZID=America/New_York:20170426T160000 SEQUENCE:0 SUMMARY:Data Seminar: Wojciech Czaja (University of Maryland\, College Park ) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/data-seminar-wojciech-czaja-univ ersity-maryland-college-park-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Solving Fredholm integrals from incomplete measurements

\nWe p resent an algorithm to solve Fredholm integrals of the first kind with ten sor product structures\, from a limited number of measurements with the go al of using this method to accelerate Nuclear Magnetic Resonance (NMR) acq uisition. This is done by incorporating compressive sampling type argument s to fill in the missing measurements using a priori knowledge of the stru cture of the data. In the first step\, we recover a compressed data matrix from measurements that form a tight frame\, and establish that these meas urements satisfy the restricted isometry property (RIP). In the second ste p\, we solve the zeroth-order regularization minimization problem using th e Venkataramanan-Song-Huerlimann algorithm. We demonstrate the performance of this algorithm on simulated and real data and we compare it with other sampling techniques. Our theory applied to both 2D and multidimensional N MR.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-8416@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Parametrization of discrete optimization problems\, subd eterminants and matrix-decomposition\n \nAbstract:\nThe central goal of th is talk is to identify parameters that explain the complexity of Integer l inear programming defined as follows:\nLet P be a polyhedron. Determine an integral point in P that maximizes a linear function.\n \nIt is obvious t hat the number of integer variables is such a parameter.\nHowever\, in vie w of applications in very high dimensions\, the question emerges whether w e need to treat all variables as integers? In other words\, can we reparam etrize integer programs with significantly less integer variables?\n \nA s econd much less obvious parameter associated with an integer linear progra m is the number Delta defined as the maximum absolute value of any square submatrix of a given integral matrix A with m rows and n columns.\nThis le ads us to the important open question whether we can solve integer linear programming in a polynomial running time in Delta and the instance size?\n \nRegarding the first question\, we exhibit a variety of examples that de monstrate how integer programs can be reformulated using much less integer variables. To this end\, we introduce a generalization of total unimodula rity called the affine TU-dimension of a matrix and study related theory a nd algorithms for determining the affine TU-dimension of a matrix.\n \nReg arding the second question\,\nwe present several new results that illustra te why Delta is an important parameter about the complexity of integer lin ear programs associated with a given matrix A.\nIn particular\, in the non degenerate case integer linear programs with any constant value Delta can be solved in polynomial time.\nThis extends earlier results of Veselov and Chirkov. DTSTART;TZID=America/New_York:20170427T133000 DTEND;TZID=America/New_York:20170427T143000 SEQUENCE:0 SUMMARY:Seminar: Robert Weismantel (ETH Zurich) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-robert-weismantel-eth-zu rich-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Parame trization of discrete optimization problems\, subdeterminants and matrix-d ecomposition

\n\n

Abstract:

\nThe central goal of this talk is to identify parameters that explain the complexity of Integer line ar programming defined as follows:

\nLet P be a polyhedron. Determin e an integral point in P that maximizes a linear function.

\n\n

It is obvious that the number of integer variables is such a parameter.

\nHowever\, in view of applications in very high dimensions\, the q uestion emerges whether we need to treat all variables as integers? In oth er words\, can we reparametrize integer programs with significantly less i nteger variables?

\n\n

A second much less obvious parameter a ssociated with an integer linear program is the number Delta defined as th e maximum absolute value of any square submatrix of a given integral matri x A with m rows and n columns.

\nThis leads us to the important open question whether we can solve integer linear programming in a polynomial running time in Delta and the instance size?

\n\n

Regarding t he first question\, we exhibit a variety of examples that demonstrate how integer programs can be reformulated using much less integer variables. To this end\, we introduce a generalization of total unimodularity called th e affine TU-dimension of a matrix and study related theory and algorithms for determining the affine TU-dimension of a matrix.

\n\n

Reg arding the second question\,

\nwe present several new results that i llustrate why Delta is an important parameter about the complexity of inte ger linear programs associated with a given matrix A.

\nIn particula r\, in the nondegenerate case integer linear programs with any constant va lue Delta can be solved in polynomial time.

\nThis extends earlier r esults of Veselov and Chirkov.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9270@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Vol\, Skew\, and Smile Trading\nAbstract: We use dynamic ally traded portfolios of options to bet on either the quadratic variation of log price\, or on the realized co-variation of log price with log impl ied vol\, or on the quadratic variation of implied vol. Our bets lead to p recise financial meanings for the level\, slope\, and curvature of implied variance in moneyness.\n DTSTART;TZID=America/New_York:20170504T133000 DTEND;TZID=America/New_York:20170504T143000 SEQUENCE:0 SUMMARY:Seminar: Peter Carr (NYU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/seminar-peter-carr-nyu-whitehead -304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Vol\, Skew\, and Smile Tradin g

\nAbstract: We use dynamically traded portfolios of options to bet on either the q uadratic variation of log price\, or on the realized co-variation of log p rice with log implied vol\, or on the quadratic variation of implied vol. Our bets lead to precise financial meanings for the level\, slope\, and curvature of impli ed variance in moneyness.

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-9294@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT:http://www.math.jhu.edu/~data/ DESCRIPTION: DTSTART;TZID=America/New_York:20170510T150000 DTEND;TZID=America/New_York:20170510T160000 SEQUENCE:0 SUMMARY:Data Seminar: Rachel Ward (University of Texas\, Austin) @ Whitehea d 304 URL:https://engineering.jhu.edu/ams/events/data-seminar-rachel-ward-univers ity-texas-austin-whitehead-304/ X-COST-TYPE:free END:VEVENT BEGIN:VEVENT UID:ai1ec-10353@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: \nTitle: Introduction to the Financial Mathematics Semina r\n \nAbstract:\nThe seminar will have two parts:\nPart I) Daniel Naiman w ill introduce the Financial Mathematics seminar.\nPart II) Sonjala William s will speak about networking and job search strategies. DTSTART;TZID=America/New_York:20170905T133000 DTEND;TZID=America/New_York:20170905T150000 LOCATION:Shaffer 101 SEQUENCE:0 SUMMARY:Financial Mathematics Seminar: Daniel Naiman and Sonjala Williams ( JHU) @ Shaffer 101 URL:https://engineering.jhu.edu/ams/events/introduction-to-the-financial-ma thematics-seminar/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

\n

Tit le: Introduction to the Financial Mathematics Seminar

\n\n

Abstract:

\nThe seminar will have two parts:

\nPart I) Dani el Naiman will introduce the Financial Mathematics seminar.

\nPart I I) Sonjala Williams will speak about networking and job search strategies.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10398@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Dr. Dave Schrader will be speaking at Sports Analytics Club’s f irst meeting of the year on Monday\, September 11 at 7pm in Schafer 3. Dr. Schrader is an expert in the field of sports analytics\, and has a wide b readth of experience in industry and in speaking to college students about the field. Individuals with all levels of experience in sports analytics are welcome to attend\, and the talk should give a good overview of how sp orts analytics are currently being used and how students can get involved themselves.\n \nThe Dr. Schrader’s talk is entitled “The Golden Age of Spo rts Analytics\,” and it will cover the following topics:\n\nWhat’s happeni ng around the world to collect and analyze data for recruiting\, player de velopment\, game planning\, and injury prevention?\nHow are analytics bein g used to evaluate and improve business operations – ticket pricing\, sal es\, sponsorships?\nWhat analytics are the leading pro teams and leagues u sing for basketball\, baseball\, football\, hockey\, and soccer?\nHow quic kly are analytics being adopted at the college level? Who is leading? What are they doing?\nHow can other parts of the university\, like the busines s school or computer science departments\, collaborate with sports program s to provide analytics for teams? What are good “Moneyball” projects to l aunch? What have other schools done?\nWhere can you get more information? What to read? What conferences to attend? DTSTART;TZID=America/New_York:20170911T190000 DTEND;TZID=America/New_York:20170911T203000 SEQUENCE:0 SUMMARY:Dr. Dave Schrader at Sports Analytics Club Meeting @ 7pm in Schafer 3 URL:https://engineering.jhu.edu/ams/events/dr-dave-schrader-at-sports-analy tics-club-meeting-7pm-in-schafer-3/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nDr. Dave Schr ader will be speaking at Sports Analytics Club’s first meeting of the year on Monday\, September 11 at 7pm in Schafer 3. Dr. Schrader is an expert i n the field of sports analytics\, and has a wide breadth of experience in industry and in speaking to college students about the field. Individuals with all levels of experience in sports analytics are welcome to attend\, and the talk should give a good overview of how sports analytics are curre ntly being used and how students can get involved themselves.

\n\n

The Dr. Schrader’s talk is entitled “The Golden Age of Sports Analyt ics\,” and it will cover the following topics:

\n- \n
- What’s happe ning around the world to collect and analyze data for recruiting\, player development\, game planning\, and injury prevention? \n
- How are ana lytics being used to evaluate and improve business operations – ticket pr icing\, sales\, sponsorships? \n
- What analytics are the leading pro teams and leagues using for basketball\, baseball\, football\, hockey\, a nd soccer? \n
- How quickly are analytics being adopted at the colleg e level? Who is leading? What are they doing? \n
- How can other part s of the university\, like the business school or computer science departm ents\, collaborate with sports programs to provide analytics for teams? W hat are good “Moneyball” projects to launch? What have other schools done ? \n
- Where can you get more information? What to read? What confere nces to attend? \n

Title: No equ ations\, no variables\, no parameters\, no space\, no time: Data and the c omputational modeling of complex/multiscale systems

\nAbstract: Obta ining predictive dynamical equations from data lies at the heart of scienc e and engineering modeling\, and is the linchpin of our technology. In mat hematical modeling one typically progresses from observations of the world (and some serious thinking!) first to equations for a model\, and then to the analysis of the model to make predictions. Good mathematical models g ive good predictions (and inaccurate ones do not) – but the computational tools for analyzing them are the same: algorithms that are typically based on closed form equations. While the skeleton of the process remains the s ame\, today we witness the development of mathematical techniques that ope rate directly on observations -data-\, and appear to circumvent the seriou s thinking that goes into selecting variables and parameters and deriving accurate equations. The process then may appear to the user a little like making predictions by “looking in a crystal ball”. Yet the “serious thinki ng” is still there and uses the same -and some new- mathematics: it goes i nto building algorithms that “jump directly” from data to the analysis of the model (which is now not available in closed form) so as to make predic tions. Our work here presents a couple of efforts that illustrate this “ne w” path from data to predictions. It really is the same old path\, but it is travelled by new means.

\nRelated papers:

\nParsimonious Re
presentation of Nonlinear Dynamical Systems through Manifold Learning: a C
hemotaxis Case Study

\nAn Equal Space for Complex Data with Unknown I
nternal Order: Observability\, Gauge Invariance and Manifold Learning Kev
rekidis

Symmetry\, Temporal Information\, and Succinct Representatio n of Random Graph Structures

\nI will discuss mathematical aspects o f my recent work on two related problems at the intersection of random gra phs and information theory: (i) node order inference – for a dynamic rando m graph model\, determine the extent to which the order in which nodes arr ived can be inferred from the graph structure\, and (ii) source coding of structures – for a given graph model\, exhibit an efficiently computable a nd invertible mapping from unlabeled graphs to bit strings with minimum po ssible expected code length. Both problems are connected to the study of t he symmetries of the graph model\, as well as another combinatorial quanti ty – the typical number of feasible labeled representatives of a given str ucture. I will focus on the case of the preferential attachment model\, fo r which we are able to give a (nearly) complete characterization of the be havior of the size of the automorphism group\, as well as a provably asymp totically optimal algorithm for (ii)\, and optimal estimators for certain natural formulations of (i).

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10414@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Semiparametric spectral modeling of the Drosophila conn ectome\nAbstract: We present semiparametric spectral modeling of the comp lete larval Drosophila mushroom body connectome. Motivated by a thorough e xploratory data analysis of the network via Gaussian mixture modeling (GMM ) in the adjacency spectral embedding (ASE) representation space\, we intr oduce the latent structure model (LSM) for network modeling and inference. LSM is a generalization of the stochastic block model (SBM) and a special case of the random dot product graph (RDPG) latent position model\, and i s amenable to semiparametric GMM in the ASE representation space. The resu lting connectome code derived via semiparametric GMM composed with ASE cap tures latent connectome structure and elucidates biologically relevant neu ronal properties. \nRelated papers:\nThe complete connectome of a learning and memory center in an insect brain\nA consistent adjacency spectral emb edding for stochastic blockmodel graphs\nA limit theorem for scaled eigenv ectors of random dot product graphs\nLimit theorems for eigenvectors of th e normalized Laplacian for random graphs DTSTART;TZID=America/New_York:20170920T150000 DTEND;TZID=America/New_York:20170920T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Carey Priebe (JHU) @ Hodson 203 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-carey-prieb e-jhu-hodson-203/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Title: Semip arametric spectral modeling of the Drosophila connectome

\nAbstract: We present semiparametric spectral modeling of the complete larval Droso phila mushroom body connectome. Motivated by a thorough exploratory data a nalysis of the network via Gaussian mixture modeling (GMM) in the adjacenc y spectral embedding (ASE) representation space\, we introduce the latent structure model (LSM) for network modeling and inference. LSM is a general ization of the stochastic block model (SBM) and a special case of the rand om dot product graph (RDPG) latent position model\, and is amenable to sem iparametric GMM in the ASE representation space. The resulting connectome code derived via semiparametric GMM composed with ASE captures latent conn ectome structure and elucidates biologically relevant neuronal properties.

\nRelated papers:

\nThe complete connectome of a learning and
memory center in an insect brain

\nA consistent adjacency spectral e
mbedding for stochastic blockmodel graphs

\nA limit theorem for scale
d eigenvectors of random dot product graphs

\nLimit theorems for eige
nvectors of the normalized Laplacian for random graphs

TITLE – On op timizing a submodular utility function

\nABSTRACT – This talk has tw o related parts. Part one is on the maximization of a particular submodula r utility function\, whereas part two is on its minimization. Both problem s arise naturally in combinatorial optimization with risk aversion\, inclu ding estimation of project duration with stochastic task times\, in reliab ility models\, multinomial logit models\, competitive facility location\, combinatorial auctions\, as well as in portfolio optimization.

\nPar t 1: Given a monotone concave univariate function g\, and two vectors c an d d\, we consider the discrete optimization problem of finding a vertex of a polytope maximizing the utility function c’x + g(d’x). The problem is N P-hard for any strictly concave function g even for simple polytopes\, suc h as the uniform matroid\, assignment and path polytopes. We give a 1/2-ap proximation algorithm for it and improvements for special cases\, where g is the square root\, log utility\, negative exponential utility and multin omial logit probability function. In particular\, for the square root func tion\, the approximation ratio improves to 4/5. Although the worst case bo unds are tight\, computational experiments indicate that the approximation algorithm finds solutions within 1-2% optimality gap for most of the inst ances very quickly and can be considerably faster than the existing altern atives.

\nPart 2: We consider a mixed 0-1 conic quadratic optimizati on problem with indicator variables arising in mean-risk optimization. The indicator variables are often used to model non-convexities such as fixed charges or cardinality constraints. Observing that the problem reduces to a submodular function minimization for its binary restriction\, we derive three classes of strong convex valid inequalities by lifting the polymatr oid inequalities on the binary variables. Computational experiments demons trate the effectiveness of the inequalities in strengthening the convex re laxations and\, thereby\, improving the solution times for mean-risk probl ems with fixed charges and cardinality constraints significantly.

\nTitle: Frames — two case studies: ambiguity and uncertainty

\nAbstract: The theor y of frames is an essential concept for dealing with signal representation in noisy environments. We shall examine the theory in the settings of the narrow band ambiguity function and of quantum information theory. For the ambiguity function\, best possible estimates are derived for applicable c onstant amplitude zero autocorrelation (CAZAC) sequences using Weil’s solu tion of the Riemann hypothesis for finite fields. In extending the theory to the vector-valued case modelling multi-sensor environments\, the defini tion of the ambiguity function is characterized by means of group frames. For the uncertainty principle\, Andrew Gleason’s measure theoretic theorem \, establishing the transition from the lattice interpretation of quantum mechanics to Born’s probabilistic interpretation\, is generalized in terms of frames to deal with uncertainty principle inequalities beyond Heisenbe rg’s. My collaborators are Travis Andrews\, Robert Benedetto\, Jeffrey Don atelli\, Paul Koprowski\, and Joseph Woodworth.

\nRelated papers:**\nSuper-resolution by means of Beurling minimal extrapolation\nG
eneralized Fourier frames in terms of balayage\nUncertainty principl
es and weighted norm inequalities\nA frame reconstruction algorithm
with applications to magnetric resonance imaging\nFrame multiplicati
on theory and a vector-valued DFT and ambiguity functions**

Title: An improved approach to calibrating misspecified mathematical models

\nAbstract:

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10421@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Data-driven discovery of governing equations and physica l laws\nAbstract: The emergence of data methods for the sciences in the la st decade has been enabled by the plummeting costs of sensors\, computatio nal power\, and data storage. Such vast quantities of data afford us new o pportunities for data-driven discovery\, which has been referred to as the 4th paradigm of scientific discovery. We demonstrate that we can use emer ging\, large-scale time-series data from modern sensors to directly constr uct\, in an adaptive manner\, governing equations\, even nonlinear dynamic s\, that best model the system measured using modern regression techniques . Recent innovations also allow for handling multi-scale physics phenomeno n and control protocols in an adaptive and robust way. The overall archite cture is equation-free in that the dynamics and control protocols are disc overed directly from data acquired from sensors. The theory developed is d emonstrated on a number of canonical example problems from physics\, biolo gy and engineering. \nRelated papers:\nDiscovering governing equations fro m data by sparse identification of nonlinear dynamical systems\nData-drive n discovery of partial differential equations\nChaos as an intermittently forced linear system DTSTART;TZID=America/New_York:20171004T150000 DTEND;TZID=America/New_York:20171004T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Nathan Kutz (University of Washington) @ Hods on 203 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-nathan-kutz -university-washington-hodson-203/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Title: Data-d riven discovery of governing equations and physical laws

\nAbstract: The emergence of data methods for the sciences in the last decade has bee n enabled by the plummeting costs of sensors\, computational power\, and d ata storage. Such vast quantities of data afford us new opportunities for data-driven discovery\, which has been referred to as the 4th paradigm of scientific discovery. We demonstrate that we can use emerging\, large-scal e time-series data from modern sensors to directly construct\, in an adapt ive manner\, governing equations\, even nonlinear dynamics\, that best mod el the system measured using modern regression techniques. Recent innovati ons also allow for handling multi-scale physics phenomenon and control pro tocols in an adaptive and robust way. The overall architecture is equation -free in that the dynamics and control protocols are discovered directly f rom data acquired from sensors. The theory developed is demonstrated on a number of canonical example problems from physics\, biology and engineerin g.

\nRelated papers:

\nDiscovering governing equations from da
ta by sparse identification of nonlinear dynamical systems

\nData-dri
ven discovery of partial differential equations

\nChaos as an intermi
ttently forced linear system

Title: Multid imensional wavelet signal denoising via adaptive random partitioning

\n\n

Abstract: Traditional statistical wavelet analysis usually fo cuses on modeling the wavelet coefficients under a given\, predetermined w avelet transform. Such analysis may quickly lose efficiency in multivariat e problems under traditional multivariate wavelet transforms\, which are s ymmetric with respect to the dimensions\, as predetermined wavelet transfo rms cannot adaptively exploit the energy distribution in a problem-specifi c manner. We introduce a principled probabilistic framework for incorporat ing such adaptivity by (i) representing multivariate functions using one-d imensional (1D) wavelet transforms applied to a permuted version of the or iginal function\, and (ii) placing a hyperprior on the corresponding permu tation. Such a representation can achieve substantially better energy conc entration in the wavelet coefficients and highly scalable inference algori thms. In particular\, when combined with the Haar basis\, we obtain the ex act Bayesian inference analytically through a recursive message passing al gorithm with a computational complexity that scales linearly with sample s ize. In addition\, we propose a sequential Monte Carlo (SMC) inference alg orithm for other wavelet bases using the exact Haar solution as the propos al. We demonstrate that with this framework even simple 1D Haar wavelets c an achieve excellent performance in both 2D and 3D image reconstruction vi a numerical experiments\, outperforming state-of-the-art multidimensional wavelet-based methods especially in low signal-to-noise ratio settings\, a t a fraction of the computational cost.

\n\n

This is a joint work with Li Ma.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10566@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Discrete Nonlinear Optimization by State-Space Decomposi tions\n \nAbstract: In this talk we will discuss a decomposition approach for binary optimization problems with nonlinear objectives and linear cons traints. Our methodology relies on the partition of the objective function into separate low-dimensional dynamic programming (DP) models\, each of w hich can be equivalently represented as a shortest-path problem in an unde rlying state transition graph. We show that the associated transition grap hs can be related by a mixed-integer linear program (MILP) so as to produc e exact solutions to the original nonlinear problem. To address DPs with l arge state spaces\, we present a general relaxation mechanism which dynami cally aggregates states during the construction of the transition graphs. The resulting MILP provides both lower and upper bounds to the nonlinear f unction\, and may be embedded in branch-and-bound procedures to find prova bly optimal solutions. We describe how to specialize our technique for str uctured objectives (e.g.\, submodular functions) and consider three proble ms arising in revenue management\, portfolio optimization\, and healthcare . Numerical studies indicate that the proposed technique often outperforms state-of-the-art approaches by orders of magnitude in these applications. DTSTART;TZID=America/New_York:20171006T100000 DTEND;TZID=America/New_York:20171006T110000 SEQUENCE:0 SUMMARY:Additional AMS Seminar: David Bergman (UConn) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/additional-ams-seminar-david-ber gman-uconn-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n* T
itle*: Discrete Nonlinear Optimization by State-Space Decompo
sitions

\n

* Abstract*: In this talk
we will discuss a decomposition approach for binary optimization problems
with nonlinear objectives and linear constraints. Our methodology relies
on the partition of the objective function into separate low-dimensional d
ynamic programming (DP) models\, each of which can be equivalently represe
nted as a shortest-path problem in an underlying state transition graph. W
e show that the associated transition graphs can be related by a mixed-int
eger linear program (MILP) so as to produce exact solutions to the origina
l nonlinear problem. To address DPs with large state spaces\, we present a
general relaxation mechanism which dynamically aggregates states during t
he construction of the transition graphs. The resulting MILP provides both
lower and upper bounds to the nonlinear function\, and may be embedded in
branch-and-bound procedures to find provably optimal solutions. We descri
be how to specialize our technique for structured objectives (e.g.\, submo
dular functions) and consider three problems arising in revenue management
\, portfolio optimization\, and healthcare. Numerical studies indicate tha
t the proposed technique often outperforms state-of-the-art approaches by
orders of magnitude in these applications.

**Title
:** Discussion of quantitative careers in private banking

\n

**Abstract:** TBA

Title: Data a ssimilation with stochastic model reduction

\nAbstract: In weather a nd climate prediction\, data assimilation combines data with dynamical mod els to make prediction\, using ensemble of solutions to represent the unce rtainty. Due to limited computational resources\, reduced models are neede d and coarse-grid models are often used\, and the effects of the subgrid s cales are left to be taken into account. A major challenge is to account f or the memory effects due to coarse graining while capturing the key stati stical-dynamical properties. We propose to use nonlinear autoregression mo ving average (NARMA) type models to account for the memory effects\, and d emonstrate by examples that the resulting NARMA type stochastic reduced mo dels can capture the key statistical and dynamical properties and therefor e can improve the performance of ensemble prediction in data assimilation. The examples include the Lorenz 96 system (which is a simplified model of the atmosphere) and the Kuramoto-Sivashinsky equation of spatiotemporally chaotic dynamics.

\nRelated papers:

\nDiscrete approach to st
ochastic parametrization and dimension reduction in nonlinear dynamics

\nAccounting for model error from unresolved scales in ensemble Kalman f
ilters by stochastic parametrization

**Title
: **Financial Contagion and Systemic Risk

**Abstract:
**Financial contagion occurs when the distress of one bank jeopard
izes the health of other financial firms\, and can ultimately spread to th
e real economy. The spread of defaults in the financial system can occur d
ue to both local connections\, e.g.\, contractual obligations\, and global
connections\, e.g.\, through the prices of assets due to mark-to-market v
aluation. As evidenced by the 2007-2009 financial crisis\, the cost of a s
ystemic event is tremendous\, thus requiring a detailed look at the contri
buting factors. In this talk\, we will detail the local contagion model of
Eisenberg and Noe (2001). However\, in utilizing this model\, central ban
kers and regulators often must estimate the interbank liabilities because
complete information on bilateral obligations is rarely available. This es
timation can introduce errors to the level of financial contagion and risk
in the system. We will consider a sensitivity analysis of the Eisenberg-N
oe model to determine the size of these potential estimation errors.

Title: “Hyper -Molecules” in Cryo-Electron Microscopy (cryo-EM)

\nAbstract: Cryo-E M is an imaging technology that is revolutionizing structural biology\; th e Nobel Prize in Chemistry 2017 was recently awarded to Jacques Dubochet\, Joachim Frank and Richard Henderson “for developing cryo-electron microsc opy for the high-resolution structure determination of biomolecules in sol ution”. Cryo-electron microscopes produce a large number of very noisy two -dimensional projection images of individual frozen molecules. Unlike rela ted tomography methods\, such as computed tomography (CT)\, the viewing di rection of each image is unknown. The unknown directions\, together with e xtreme levels of noise and additional technical factors\, make the determi nation of the structure of molecules challenging. Unlike other structure d etermination methods\, such as x-ray crystallography and nuclear magnetic resonance (NMR)\, cryo-EM produces measurements of individual molecules an d not ensembles of molecules. Therefore\, cryo-EM could potentially be use d to study mixtures of different conformations of molecules. While current algorithms have been very successful at analyzing homogeneous samples\, a nd can recover some distinct conformations mixed in solutions\, the determ ination of multiple conformations\, and in particular\, continuums of simi lar conformations (continuous heterogeneity)\, remains one of the open pro blems in cryo-EM. I will discuss the “hyper-molecules” approach to continu ous heterogeneity\, and the numerical tools and analysis methods that we a re developing in order to recover such hyper-molecules.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10484@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Optimization with Polyhedral Constraints\n \nAbstract:\n A two-phase strategy is presented for solving an optimization problem whos e feasible set is a polyhedron. Phase one is the gradient projection algor ithm\, while phase two is essentially any algorithm for solving a linearly constrained optimization problem. Using some simple rules for branching b etween the two phases\, it is shown\, under suitable assumptions\, that on ly the linearly constrained optimization algorithm is performed asymptotic ally. Hence\, the asymptotic convergence behavior of the two phase algorit hm coincides with the convergence behavior of the linearly constrained opt imizer. Numerical results are presented using CUTE test problems\, a recen tly developed algorithm for projecting a point onto a polyhedron\, and the conjugate gradient algorithm for the linearly constrained optimizer. DTSTART;TZID=America/New_York:20171019T133000 DTEND;TZID=America/New_York:20171019T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Bill Hager (University of Florida) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-bill-hager-universit y-florida-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Optimi zation with Polyhedral Constraints

\n\n

Abstract:

\nA t wo-phase strategy is presented for solving an optimization problem whose f easible set is a polyhedron. Phase one is the gradient projection algorith m\, while phase two is essentially any algorithm for solving a linearly co nstrained optimization problem. Using some simple rules for branching betw een the two phases\, it is shown\, under suitable assumptions\, that only the linearly constrained optimization algorithm is performed asymptoticall y. Hence\, the asymptotic convergence behavior of the two phase algorithm coincides with the convergence behavior of the linearly constrained optimi zer. Numerical results are presented using CUTE test problems\, a recently developed algorithm for projecting a point onto a polyhedron\, and the co njugate gradient algorithm for the linearly constrained optimizer.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10432@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:TBA DTSTART;TZID=America/New_York:20171025T150000 DTEND;TZID=America/New_York:20171025T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: John Harlim (Penn State University) @ Hodson 203 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-john-harlim -penn-state-university-hodson-203/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTBA

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10487@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Coherence in Statistical Modeling of Networks\nAbstract :\nGeorge Box famously said\, “All models are wrong\, but some are useful. ”\nClassic texts define a statistical model as “a set of distributions on the sample space” (Cox and Hinkley\, 1976\; Lehman\, 1983\; Barndorff-Niel son and Cox\, Bernardo and Smith\, 1994).\nMotivated by some longstanding questions in the analysis of network data\, I will examine both of these s tatements\, first from a general point of view\, and then in the context o f some recent developments in network analysis.\nThe confusion caused by t hese statements is clarified by the realization that the definition of sta tistical model must be refined — it must be more than just a set. With th is\, the ambiguity in Box’s statement — e.g.\, what determines whether a m odel is ‘wrong’ or ‘useful’? — can be clarified by a logical property that I call ‘coherence’. After clarification\, a model is deemed useful as lo ng as it is coherent\, i.e.\, inferences from it ‘make sense’.\nI will the n discuss some implications for the statistical modeling of network data. DTSTART;TZID=America/New_York:20171026T133000 DTEND;TZID=America/New_York:20171026T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Harry Crane ( Rutgers University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-harry-crane-rutgers- university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Coher ence in Statistical Modeling of Networks

\nAbstract:

\nGeorge Box famously said\, “All models are wrong\, but some are useful.”

\nClassic texts define a statistical model as “a set of distributions on the sample space” (Cox and Hinkley\, 1976\; Lehman\, 1983\; Barndorff-Nielson and Cox\, Bernardo and Smith\, 1994).

\nMotivated by some longstand ing questions in the analysis of network data\, I will examine both of the se statements\, first from a general point of view\, and then in the conte xt of some recent developments in network analysis.

\nThe confusion caused by these statements is clarified by the realization that the defini tion of statistical model must be refined — it must be more than just a se t. With this\, the ambiguity in Box’s statement — e.g.\, what determines whether a model is ‘wrong’ or ‘useful’? — can be clarified by a logical pr operty that I call ‘coherence’. After clarification\, a model is deemed u seful as long as it is coherent\, i.e.\, inferences from it ‘make sense’.< /p>\n

I will then discuss some implications for the statistical modeling of network data.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10680@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Functional central limit theorems for rough volatility m odels\nAbstract: We extend Donsker’s approximation of Brownian motion to f ractional Brownian motion with any Hurst exponent (including the ’rough’ c ase H < 1/2)\, and Volterra-like processes. Some of the most relevant cons equences of our ‘rough Donsker (rDonsker) Theorem’ are convergence results (with rates) for discrete approximations of a large class of rough models . This justifies the validity of simple and easy-to-implement Monte-Carlo methods\, for which we provide detailed numerical recipes. We test these a gainst the current benchmark hybrid scheme of Bennedsen\, Lunde\, and Pakk anen and find remarkable agreement (for a large range of values of H). Thi s rDonsker Theorem further provides a weak convergence proof for the hybri d scheme itself\, and allows to construct binomial trees for rough volatil ity models\, the first available scheme (in the rough volatility context) for early exercise options such as American or Bermudan. The talk is based on joint work with B. Horvath and A. Muguruza.\n DTSTART;TZID=America/New_York:20171031T133000 DTEND;TZID=America/New_York:20171031T150000 SEQUENCE:0 SUMMARY:Financial Mathematics Seminar: Dr. Antoine Jacquier (Imperial Colle ge London) URL:https://engineering.jhu.edu/ams/events/financial-mathematics-seminar-dr -antoine-jacquier-imperial-college-london/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Functional central limit theorems for rough volatility models\n

**Abstract: **We extend Donsker’s approximation of Bro
wnian motion to fractional Brownian motion with any Hurst exponent (includ
ing the ’rough’ case H < 1/2)\, and Volterra-like processes. Some of the m
ost relevant consequences of our ‘rough Donsker (rDonsker) Theorem’ are co
nvergence results (with rates) for discrete approximations of a large clas
s of rough models. This justifies the validity of simple and easy-to-imple
ment Monte-Carlo methods\, for which we provide detailed numerical recipes
. We test these against the current benchmark hybrid scheme of Bennedsen\,
Lunde\, and Pakkanen and find remarkable agreement (for a large range of
values of H). This rDonsker Theorem further provides a weak convergence pr
oof for the hybrid scheme itself\, and allows to construct binomial trees
for rough volatility models\, the first available scheme (in the rough vol
atility context) for early exercise options such as American or Bermudan.
The talk is based on joint work with B. Horvath and A. Muguruza.

**<
strong> **

Title: Some m atrix problems in quantum information science

\nAbstract:

\n~~In this talk\, we
present some matrix results and techniques in~~

\nsolving certain opt
imization problems arising in quantum information

\nscience.<
/p>\n

~~No quantum mechanics background is required.~~

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10686@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES;LANGUAGE=en-US:Financial Mathematics Seminar CONTACT: DESCRIPTION:Title: Data scientist at Kensho working focusing on natural lan guage processing DTSTART;TZID=America/New_York:20171107T133000 DTEND;TZID=America/New_York:20171107T150000 SEQUENCE:0 SUMMARY:Financial Mathematics Seminar: Ben Cohen URL:https://engineering.jhu.edu/ams/events/financial-mathematics-seminar-be n-cohen/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
: **Data scientist at Kensho working focusing on natural language p
rocessing

Title: Emerge nt behavior in self-organized dynamics: from consensus to hydrodynamic flo cking

\nAbstract: We discuss several first- and second-order models encountered in opinion and flocking dynamics. The models are driven by dif ferent “rules of engagement”\, which quantify how each member interacts wi th its immediate neighbors. We highlight the role of geometric vs. topolog ical neighborhoods and distinguish between local and global interactions\, while addressing the following two related questions. (i) How local rules of interaction lead\, over time\, to the emergence of consensus\; and (ii ) How the flocking behavior of large crowds captured by their hydrodynamic description.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10494@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Market Efficiency with Micro and Macro Information\nAbs tract:\nWe propose a tractable\, multi-security model in which investors a llocate information processing capacity to acquire micro information about individual stocks and/or macro information about an index fund. Investors solve optimal portfolio selection and information allocation problems. In equilibrium\, all investors are of one of three types: micro informed\, m acro informed\, or uninformed. We investigate the implications for price e fficiency and find an endogenous bias toward micro efficiency: over a rang e of parameter values prices will be more informative about micro than mac ro fundamentals. We explore the model’s implications for the cyclicality o f investor information choices\, for systematic and idiosyncratic return v olatility\, and for excess covariance and volatility. This is joint work w ith Harry Mamaysky.\n\nNo quantum mechanics background is required. DTSTART;TZID=America/New_York:20171109T133000 DTEND;TZID=America/New_York:20171109T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Paul Glasserman (Columbia University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-paul-glasserman-colu mbia-university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Marke t Efficiency with Micro and Macro Information

\nAbstract:

\nWe propose a tractable\, multi-security model in which investors allocate in formation processing capacity to acquire micro information about individua l stocks and/or macro information about an index fund. Investors solve opt imal portfolio selection and information allocation problems. In equilibri um\, all investors are of one of three types: micro informed\, macro infor med\, or uninformed. We investigate the implications for price efficiency and find an endogenous bias toward micro efficiency: over a range of param eter values prices will be more informative about micro than macro fundame ntals. We explore the model’s implications for the cyclicality of investor information choices\, for systematic and idiosyncratic return volatility\ , and for excess covariance and volatility. This is joint work with Harry Mamaysky.

\nNo quantum mechanics background is required.\n

\n\n
END:VEVENT
BEGIN:VEVENT
UID:ai1ec-10663@engineering.jhu.edu/ams
DTSTAMP:20180715T165111Z
CATEGORIES:
CONTACT:
DESCRIPTION:What geometries can we learn from data?\nIn the field of manifo
ld learning\, the foundational theoretical results of Coifman and Lafon (D
iffusion Maps\, 2006) showed that for data sampled near an embedded manifo
ld\, certain graph Laplacian constructions are consistent estimators of th
e Laplace-Beltrami operator on the underlying manifold. Since these operat
ors determine the Riemannian metric\, they completely describe the geometr
y of the manifold (as inherited from the embedding). It was later shown th
at different kernel functions could be used to recover any desired geometr
y\, at least in terms of pointwise estimation of the associated Laplace-Be
ltrami operator. In this talk I will first briefly review the above result
s and then introduce new results on the spectral convergence of these grap
h Laplacians. These results reveal that not all geometries are accessible
in the stronger spectral sense. However\, when the data set is sampled fro
m a smooth density\, there is a natural conformally invariant geometry whi
ch is accessible on all compact manifolds\, and even on a large class of n
on-compact manifolds. Moreover\, the kernel which estimates this geometry
has a very natural construction which we call Continuous k-Nearest Neighbo
rs (CkNN).
DTSTART;TZID=America/New_York:20171115T150000
DTEND;TZID=America/New_York:20171115T160000
SEQUENCE:0
SUMMARY:Data Science Seminar: Tyrus Berry (George Mason University) @ Hodso
n 203
URL:https://engineering.jhu.edu/ams/events/data-science-seminar-tyrus-berry
-george-mason-university-hodson-203/
X-COST-TYPE:free
X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nWhat geometri es can we learn from data?

\nIn the field of manifold learning\, the foundational theoretical results of Coifman and Lafon (Diffusion Maps\, 2 006) showed that for data sampled near an embedded manifold\, certain grap h Laplacian constructions are consistent estimators of the Laplace-Beltram i operator on the underlying manifold. Since these operators determine the Riemannian metric\, they completely describe the geometry of the manifold (as inherited from the embedding). It was later shown that different kern el functions could be used to recover any desired geometry\, at least in t erms of pointwise estimation of the associated Laplace-Beltrami operator. In this talk I will first briefly review the above results and then introd uce new results on the spectral convergence of these graph Laplacians. The se results reveal that not all geometries are accessible in the stronger s pectral sense. However\, when the data set is sampled from a smooth densit y\, there is a natural conformally invariant geometry which is accessible on all compact manifolds\, and even on a large class of non-compact manifo lds. Moreover\, the kernel which estimates this geometry has a very natura l construction which we call Continuous k-Nearest Neighbors (CkNN).

\n< /BODY> END:VEVENT BEGIN:VEVENT UID:ai1ec-10496@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Transaction Clock\, Stochastic Time Changes and Stochas tic Volatility\nAbstract: The first part of the talk will establish that\, by No Arbitrage\, the log – price process of a stock has to be a time-cha nged Brownian motion under the physical probability measure. Aggregate vol ume and number of trades are empirically tested as possible drivers of the stochastic clock allowing one to recover normality of stock returns.\nThe second part of the talk will show how stochastic volatility can be repres ented through a stochastic time change\, outside the stochastic differenti al equations classically used for volatility in a number of founding model s in Finance. This representation is particularly useful if one wishes to choose a Levy process (outside Brownian motion) for the stock log- price\, as independent increments are contradicted by volatility clustering obser ved in financial markets. The CGMY process with stochastic volatility will be provided as an example. DTSTART;TZID=America/New_York:20171116T133000 DTEND;TZID=America/New_York:20171116T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Helyette Geman (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-eric-xing-carnegie-m ellon-university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Transaction Clock\, Stochastic Time Changes and Stochastic Vol
atility

**Abstract**: The first part of the talk will
establish that\, by No Arbitrage\, the log – price process of a stock has
to be a time-changed Brownian motion under the physical probability measu
re. Aggregate volume and number of trades are empirically tested as possib
le drivers of the stochastic clock allowing one to recover normality of st
ock returns.

The second part of the talk will show how stochastic volatility can be represented through a stochastic time change\, outside t he stochastic differential equations classically used for volatility in a number of founding models in Finance. This representation is particularly useful if one wishes to choose a Levy process (outside Brownian motion) fo r the stock log- price\, as independent increments are contradicted by vol atility clustering observed in financial markets. The CGMY process with st ochastic volatility will be provided as an example.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10603@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Join alumni\, faculty\, and students for a Financial Mathematic s alumni reunion\, including speakers\, networking event\, and happy hour. Food and drink will be served. (Informal Happy hour to follow)\n \nRegis ter @ https://jhu.us6.list-manage.com/track/click?u=40512314224886c4ca8b85 6c2&id=836eea02d4&e=1907de3a2c\n \nIf you have any questions\, please cont act Sonjala Williams @ sonjala@jhu.edu\n \n DTSTART;TZID=America/New_York:20171118T120000 DTEND;TZID=America/New_York:20171118T160000 SEQUENCE:0 SUMMARY:JHU Whiting School of Engineering AMS- Financial Math Alumni Reunio n @ Glass Pavilion URL:https://engineering.jhu.edu/ams/events/jhu-whiting-school-engineering-a ms-financial-math-alumni-reunion-glass-pavilion/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nJoin alumni\, faculty\, and students for a Financial Mathematics alumni reunion\, inclu ding speakers\, networking event\, and happy hour. Food and drink will be served. (Informal Happy hour to follow)

\n\n

Register @ https://jhu.us6.list-manage.com/track/click ?u=40512314224886c4ca8b856c2&id=836eea02d4&e=1907de3a2c

\n\n

If you have any questions\, please contact Sonjala Williams @ sonjala@jhu.edu

\n\n

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10664@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:TBA DTSTART;TZID=America/New_York:20171129T150000 DTEND;TZID=America/New_York:20171129T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Yingzhou Li (Duke University) @ Hodson 203 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-yingzhou-li -duke-university-hodson-203/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

TBA

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10552@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Distributed Synchronization in Engineering Networks\nAbs tract:\nThis talk presents a systematic study of synchronization on distri buted (networked) systems that spans from theoretical modeling and stabili ty analysis to distributed controller design\, implementation and verifica tion. We first focus on developing a theoretical foundation for synchroniz ation of networked oscillators. We study how the interaction type (couplin g) and network configuration (topology) affect the behavior of a populatio n of heterogeneous coupled oscillators. Unlike existing literature that re stricts to specific scenarios\, we show that phase consensus (common phase value) can be achieved for arbitrary network topologies under very genera l conditions on the oscillators’ model.\nWe then focus on more practical a spects of synchronization on computer networks. Unlike existing solutions that tend to rely on expensive hardware to improve accuracy\, we provide a novel algorithm that reduces jitter by synchronizing networked computers without estimating the frequency difference between clocks (skew) or intro ducing offset corrections. We show that a necessary and sufficient conditi on on the network topology for synchronization (in the presence of noise) is the existence of a unique leader in the communication graph. A Linux-ba sed implementation on a cluster of IBM BladeCenter servers experimentally verifies that the proposed algorithm outperforms well-established solution s and that loops can help reduce jitter.\n DTSTART;TZID=America/New_York:20171130T133000 DTEND;TZID=America/New_York:20171130T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Enrique Mallada Garcia (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-enrique-mallada-garc ia-jhu-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Distributed Synchronization in Engineering Networks

This talk presents a systematic study of s ynchronization on distributed (networked) systems that spans from theoreti cal modeling and stability analysis to distributed controller design\, imp lementation and verification. We first focus on developing a theoretical f oundation for synchronization of networked oscillators. We study how the i nteraction type (coupling) and network configuration (topology) affect the behavior of a population of heterogeneous coupled oscillators. Unlike exi sting literature that restricts to specific scenarios\, we show that phase consensus (common phase value) can be achieved for arbitrary network topo logies under very general conditions on the oscillators’ model.

\nWe then focus on more practical aspects of synchronization on computer netwo rks. Unlike existing solutions that tend to rely on expensive hardware to improve accuracy\, we provide a novel algorithm that reduces jitter by syn chronizing networked computers without estimating the frequency difference between clocks (skew) or introducing offset corrections. We show that a n ecessary and sufficient condition on the network topology for synchronizat ion (in the presence of noise) is the existence of a unique leader in the communication graph. A Linux-based implementation on a cluster of IBM Blad eCenter servers experimentally verifies that the proposed algorithm outper forms well-established solutions and that loops can help reduce jitter.

\n\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10832@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:The Graduate Career Advisor for Financial Math and Applied Math & Statistics will teach some strategies to make the most of the winter af ter fall classes end.\nLearn how to best kick off or revamp your job or in ternship search!\n \n*Food will be served\, grab it 15 minutes before the event!\nRSVP on Handshake @ https://app.joinhandshake.com/events/108603\n \nWalk-ins welcome DTSTART;TZID=America/New_York:20171205T180000 DTEND;TZID=America/New_York:20171205T193000 SEQUENCE:0 SUMMARY:Winter Wonderland: Strategies for Your Winter Job or Internship Sea rch! URL:https://engineering.jhu.edu/ams/events/winter-wonderland-strategies-win ter-job-internship-search/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

The Graduate Career Advisor for Financial Math and Applied Math & Statistics will teach some strategies to make the most of the winter after fall classes end.

\nLearn how to best kick off or revamp your job or internship search!< /p>\n

\n

***Food will be served\, grab it 15 minutes before
the event!**

RSVP on Handshake @ https://app.joinhandshake.com/events/108603

\n\n

Walk-ins welcome

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10444@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:TBA DTSTART;TZID=America/New_York:20171206T150000 DTEND;TZID=America/New_York:20171206T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Hau-Tieng Wu (Duke University) @ Hodson 203 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-hau-tieng-w u-duke-university-hodson-203/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTBA

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10556@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: The Growing Importance of Satellite Data for Health and Air Quality Applications\n \nAbstract:\nSatellite data are growing in impo rtance for health and air quality end users in the U.S. and around the wor ld. From their “Gods-eye” view\, satellites provide a level of spatial co verage unobtainable by surface monitoring networks. Satellite observation s of various pollutants\, such as nitrogen dioxide and sulfur dioxide\, vi vidly demonstrate the steady improvement of air quality in the U.S. over t he last several decades thanks to environmental regulations\, such as the Clean Air Act. However\, while better\, U.S. air quality is still not at h ealthy levels and there are occasionally extreme events (e.g.\, wildfires\ , toxic spills in Houston after Hurricane Harvey) that expose Americans to high levels of pollution. Satellite data also show that air quality in m any parts of the world is rapidly degrading\, and is likely to continue to do so as the global population is expected to increase by 2 billion by 20 50. In this presentation\, I will discuss the strengths and limitations of current satellite data for health and air quality applications as well as the potential upcoming satellites offer. I will present examples of succe ssful uses of satellite data\, discuss potential uses\, and highlight ongo ing challenges (e.g.\, data processing and visualization) for satellite da ta end users.\n \nBiographical Sketch\nDr. Bryan Duncan is an Earth scient ist at NASA’s Goddard Space Flight Center and has a keen interest in using NASA satellite data for societal benefit\, including for health and air q uality applications. He frequently speaks to representatives of various U. S. and international agencies (e.g.\, World Bank\, UNICEF) about how satel lite data may benefit their objectives and is a member of the NASA Health and Air Quality Applied Sciences Team (HAQAST). He is also the Project Sci entist of the NASA Aura satellite mission\, which has observing air qualit y from space as one of its objectives. DTSTART;TZID=America/New_York:20171207T133000 DTEND;TZID=America/New_York:20171207T143000 SEQUENCE:0 SUMMARY:The John C. & Susan S.G. Wierman Lecture Series: Bryan Duncan (NASA Goddard Space Flight Center) @ Olin 305 URL:https://engineering.jhu.edu/ams/events/john-c-susan-s-g-wierman-lecture -series-bryan-duncan-nasa-goddard-space-flight-center-tba/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
: **The Growing Importance of Satellite Data for Health and Air Qua
lity Applications

\n

**Abstract:**

Sat ellite data are growing in importance for health and air quality end users in the U.S. and around the world. From their “Gods-eye” view\, satellite s provide a level of spatial coverage unobtainable by surface monitoring n etworks. Satellite observations of various pollutants\, such as nitrogen dioxide and sulfur dioxide\, vividly demonstrate the steady improvement of air quality in the U.S. over the last several decades thanks to environme ntal regulations\, such as the Clean Air Act. However\, while better\, U.S . air quality is still not at healthy levels and there are occasionally ex treme events (e.g.\, wildfires\, toxic spills in Houston after Hurricane H arvey) that expose Americans to high levels of pollution. Satellite data also show that air quality in many parts of the world is rapidly degrading \, and is likely to continue to do so as the global population is expected to increase by 2 billion by 2050. In this presentation\, I will discuss t he strengths and limitations of current satellite data for health and air quality applications as well as the potential upcoming satellites offer. I will present examples of successful uses of satellite data\, discuss pote ntial uses\, and highlight ongoing challenges (e.g.\, data processing and visualization) for satellite data end users.

\n\n

*Biographical Sketch*

Dr. Bryan Duncan is an Earth sc ientist at NASA’s Goddard Space Flight Center and has a keen interest in u sing NASA satellite data for societal benefit\, including for health and a ir quality applications. He frequently speaks to representatives of variou s U.S. and international agencies (e.g.\, World Bank\, UNICEF) about how s atellite data may benefit their objectives and is a member of the NASA Hea lth and Air Quality Applied Sciences Team (HAQAST). He is also the Project Scientist of the NASA Aura satellite mission\, which has observing air qu ality from space as one of its objectives.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10963@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Principled non-convex optimization for deep learning and phase retrieval\nAbstract: This talk looks at two classes of non-convex p roblems. First\, we discuss phase retrieval problems\, and present a new f ormulation\, called PhaseMax\, that reduces this class of non-convex probl ems into a convex linear program. Then\, we turn our attention to more com plex non-convex problems that arise in deep learning. We’ll explore the no n-convex structure of deep networks using a range of visualization methods . Finally\, we discuss a class of principled algorithms for training “bina rized” neural networks\, and show that these algorithms have theoretical p roperties that enable them to overcome the non-convexities present in neur al loss functions. DTSTART;TZID=America/New_York:20180131T150000 DTEND;TZID=America/New_York:20180131T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Tom Goldsten (University of Maryland College Park) @ Gilman 219 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-tom-goldste n-university-of-maryland-college-park-gilman-219/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTitle: Princi pled non-convex optimization for deep learning and phase retrieval

\nAbstract: This talk looks at two classes of non-convex problems. First\, we discuss phase retrieval problems\, and present a new formulation\, call ed PhaseMax\, that reduces this class of non-convex problems into a convex linear program. Then\, we turn our attention to more complex non-convex p roblems that arise in deep learning. We’ll explore the non-convex structur e of deep networks using a range of visualization methods. Finally\, we di scuss a class of principled algorithms for training “binarized” neural net works\, and show that these algorithms have theoretical properties that en able them to overcome the non-convexities present in neural loss functions .

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10861@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Approximating Minimal Cut-Generating Functions by Extrem e Functions\nAbstract:\nWith applications in scheduling\, networks\, and g eneralized assignment problems\, integer programs are ubiquitous in a vari ety of engineering disciplines. Often\, integer programming algorithms ma ke use of strategically chosen cutting planes in order to trim the region bounded by the linear constraints without removing any feasible points. Re cently\, there has been a resurgence of interest in the theory of (minimal ) cut generating functions\, as such functions can be used to produce qual ity cuts. Moreover\, the family of minimal functions forms a convex set\; in order to better understand this class of functions\, we wish to study the extreme functions of this set. In this talk\, we shall see that the s et of continuous minimal cut generating functions contains a dense subset of extreme function. DTSTART;TZID=America/New_York:20180201T133000 DTEND;TZID=America/New_York:20180201T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Teresa Lebair (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-tba-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
**: Approximating Minimal Cut-Generating Functions by Extreme Funct
ions

**Abstract**:

\nWith applications in schedu
ling\, networks\, and generalized assignment problems\, integer programs a
re ubiquitous in a variety of engineering disciplines. Often\, integer pr
ogramming algorithms make use of strategically chosen cutting planes in or
der to trim the region bounded by the linear constraints without removing
any feasible points. Recently\, there has been a resurgence of interest in
the theory of (minimal) cut generating functions\, as such functions can
be used to produce quality cuts. Moreover\, the family of minimal functio
ns forms a convex set\; in order to better understand this class of functi
ons\, we wish to study the extreme functions of this set. In this talk\,
we shall see that the set of continuous minimal cut generating functions c
ontains a dense subset of extreme function.

**Title
: ** Limit theorems for eigenvectors of the normalized Laplacian fo
r random graphs

\n

**Abstract:**

“We p rove a central limit theorem for the components of the eigenvectors corres ponding to the d largest eigenvalues of the normalized Laplacian matrix of a finite dimensional random dot product graph. As a corollary\, we show t hat for stochastic blockmodel graphs\, the rows of the spectral embedding of the normalized Laplacian converge to multivariate normals and furthermo re the mean and the covariance matrix of each row are functions of the ass ociated vertex’s block membership. Together with prior results for the eig envectors of the adjacency matrix\, we then compare\, via the Chernoff inf ormation between multivariate normal distributions\, how the choice of emb edding method impacts subsequent inference. We demonstrate that neither em bedding method dominates with respect to the inference task of recovering the latent block assignments.”

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10997@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Monotonicity of optimal contracts without the first-ord er approach\n \nAbstract:\nWe develop a simple sufficient condition for an optimal contract of a moral\nhazard problem to be monotone in the output signal. Existing results on monotonicity\nrequire conditions on the output distribution (namely\, the monotone likelihood ratio\nproperty (MLRP)) an d additional conditions to guarantee that agent’s decision is\napproachabl e via the first-order approach of replacing that\nproblem with its first-o rder conditions. We know of no positive monotonicity\nresults in the setti ng where the first-order approach does not apply. Indeed\, it is\nwell-doc umented that when there are finitely-many possible outputs\, and the\nfirs t-order approach does not apply\, the MLRP alone is insufficient to guaran tee monotonicity.\nHowever\, we show that when there is an interval of pos sible output signals\,\nthe MLRP does suffice to establish monotonicity un der additional technical assumptions\nthat do not guarantee the validity o f the first-order approach.\n \nThis is joint work with Rongzhu Ke (Hong K ong Baptist University). DTSTART;TZID=America/New_York:20180209T090000 DTEND;TZID=America/New_York:20180209T100000 SEQUENCE:0 SUMMARY:Optimization Seminar: Christopher Ryan (University of Chicago) @ Wh itehead 304 URL:https://engineering.jhu.edu/ams/events/optimization-seminar-christopher -ryan-university-chicago-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
: **Monotonicity of optimal contracts without the first-order appr
oach

\n

**Abstract:**

We develop a sim ple sufficient condition for an optimal contract of a moral

\nhazard problem to be monotone in the output signal. Existing results on monotoni city

\nrequire conditions on the output distribution (namely\, the m onotone likelihood ratio

\nproperty (MLRP)) and additional condition s to guarantee that agent’s decision is

\napproachable via the first -order approach of replacing that

\nproblem with its first-order con ditions. We know of no positive monotonicity

\nresults in the settin g where the first-order approach does not apply. Indeed\, it is

\nwe ll-documented that when there are finitely-many possible outputs\, and the

\nfirst-order approach does not apply\, the MLRP alone is insuffici ent to guarantee monotonicity.

\nHowever\, we show that when there i s an interval of possible output signals\,

\nthe MLRP does suffice t o establish monotonicity under additional technical assumptions

\nth at do not guarantee the validity of the first-order approach.

\n\n

This is joint work with Rongzhu Ke (Hong Kong Baptist University).\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10863@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Maximum Likelihood Density Estimation under Total Positi vity\n \nAbstract: Nonparametric density estimation is a challenging probl em in theoretical statistics—in general the maximum likelihood estimate (M LE) does not even exist! Introducing shape constraints allows a path forwa rd. This talk offers an invitation to non-parametric density estimation un der total positivity (i.e. log-supermodularity) and log-concavity. Totally positive random variables are ubiquitous in real world data and possess a ppealing mathematical properties. Given i.i.d. samples from such a distrib ution\, we prove that the maximum likelihood estimator under these shape c onstraints exists with probability one. We characterize the domain of the MLE and show that it is in general larger than the convex hull of the obse rvations. If the observations are 2-dimensional or binary\, we show that t he logarithm of the MLE is a tent function (i.e. a piecewise linear functi on) with “poles” at the observations\, and we show that a certain convex p rogram can find it. In the general case the MLE is more complicated. We gi ve necessary and sufficient conditions for a tent function to be concave a nd supermodular\, which characterizes all the possible candidates for the MLE in the general case.\n DTSTART;TZID=America/New_York:20180215T133000 DTEND;TZID=America/New_York:20180215T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Dr. Elina Robeva (MIT) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-tba-whitehead-304-3/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
:** Maximum Likelihood Density Estimation under Total Positivity

\n** **

\n**Abstract:** Nonparametric de
nsity estimation is a challenging problem in theoretical statistics—in gen
eral the maximum likelihood estimate (MLE) does not even exist! Introducin
g shape constraints allows a path forward. This talk offers an invitation
to non-parametric density estimation under total positivity (i.e. log-supe
rmodularity) and log-concavity. Totally positive random variables are ubiq
uitous in real world data and possess appealing mathematical properties. G
iven i.i.d. samples from such a distribution\, we prove that the maximum l
ikelihood estimator under these shape constraints exists with probability
one. We characterize the domain of the MLE and show that it is in general
larger than the convex hull of the observations. If the observations are 2
-dimensional or binary\, we show that the logarithm of the MLE is a tent f
unction (i.e. a piecewise linear function) with “poles” at the observation
s\, and we show that a certain convex program can find it. In the general
case the MLE is more complicated. We give necessary and sufficient conditi
ons for a tent function to be concave and supermodular\, which characteriz
es all the possible candidates for the MLE in the general case.

< /p>\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11000@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: The Learning Premium\nAbstract: We find equilibrium sto ck prices and interest rates in a\nrepresentative-agent model with uncerta in dividends’ growth\, gradually\nrevealed by dividends themselves\, where asset prices are rational –\nreflect current information and anticipate t he impact of future\nknowledge on future prices. In addition to the usual premium for risk\,\nstock returns include a learning premium\, which refle cts the expected\nchange in prices from new information. In the long run\, the learning\npremium vanishes\, as prices and interest rates converge to their\ncounterparts in the standard setting with known growth. The model \nexplains the increase in price-dividend ratios of the past century if\nb oth relative risk aversion and elasticity of intertemporal\nsubstitution a re above one. This is a joint work with Paolo Guasoni. DTSTART;TZID=America/New_York:20180220T133000 DTEND;TZID=America/New_York:20180220T143000 SEQUENCE:0 SUMMARY:Financial Math Seminar: Maxim Bichuch (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/financial-math-seminar-maxim-bic huch-jhu-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
:** The Learning Premium

**Abstract**: We fin
d equilibrium stock prices and interest rates in a

representative- agent model with uncertain dividends’ growth\, gradually

\nrevealed by dividends themselves\, where asset prices are rational –

\nreflec t current information and anticipate the impact of future

\nknowledg e on future prices. In addition to the usual premium for risk\,

\nst ock returns include a learning premium\, which reflects the expected

\nchange in prices from new information. In the long run\, the learning\n

premium vanishes\, as prices and interest rates converge to their\n

counterparts in the standard setting with known growth. The model\n

explains the increase in price-dividend ratios of the past century if

\nboth relative risk aversion and elasticity of intertemporal

\nsubstitution are above one. This is a joint work with Paolo Guasoni.< /p>\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11096@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Data-driven modeling of vector fields and differential f orms by spectral exterior calculus\nAbstract: We discuss a data-driven fra mework for exterior calculus on manifolds. This framework is based on a re presentations of vector fields\, differential forms\, and operators acting on these objects in frames (overcomplete bases) for L^2 and higher-order Sobolev spaces built entirely from the eigenvalues and eigenfunctions of t he Laplacian of functions. Using this approach\, we represent vector field s either as linear combinations of frame elements\, or as operators on fun ctions via matrices. In addition\, we construct a Galerkin approximation s cheme for the eigenvalue problem for the Laplace-de-Rham operator on 1-for ms\, and establish its spectral convergence. We present applications of th is scheme to a variety of examples involving data sampled on smooth manifo lds and the Lorenz 63 fractal attractor. This work is in collaboration wit h Tyrus Berry. DTSTART;TZID=America/New_York:20180221T150000 DTEND;TZID=America/New_York:20180221T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Dimitris Giannakis (NYU) @ Shaffer 304 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-dimitris-gi annakis-shaffer-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

Title: Data-d riven modeling of vector fields and differential forms by spectral exterio r calculus

\nAbstract: We discuss a data-driven framework for exteri or calculus on manifolds. This framework is based on a representations of vector fields\, differential forms\, and operators acting on these objects in frames (overcomplete bases) for L^2 and higher-order Sobolev spaces bu ilt entirely from the eigenvalues and eigenfunctions of the Laplacian of f unctions. Using this approach\, we represent vector fields either as linea r combinations of frame elements\, or as operators on functions via matric es. In addition\, we construct a Galerkin approximation scheme for the eig envalue problem for the Laplace-de-Rham operator on 1-forms\, and establis h its spectral convergence. We present applications of this scheme to a va riety of examples involving data sampled on smooth manifolds and the Loren z 63 fractal attractor. This work is in collaboration with Tyrus Berry.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10630@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Information\, Computation\, Optimization: Connecting the dots in the Traveling Salesman Problem\nAbstract: Few math models scream impossible as loudly as the traveling salesman problem.\nGiven n cities\, the TSP asks for the shortest route to take you to all of them.\nEasy to s tate\, but if P ≠ NP then no solution method can have good asymptotic\nper formance as n goes off to infinity. The popular interpretation is that we simply\ncannot solve realistic examples. But this skips over nearly 70 yea rs of intense\nmathematical study. Indeed\, in 1949 Julia Robinson describ ed the TSP challenge in\npractical terms: “Since there are only a finite n umber of paths to consider\, the\nproblem consists in finding a method for picking out the optimal path when n is\nmoderately large\, say n = 50.” S he went on to propose a linear programming attack\nthat was adopted by her RAND colleagues Dantzig\, Fulkerson\, and Johnson several\nyears later.\n Following in the footsteps of these giants\, we show that a certain tour o f 49\,603\nhistoric sites in the US is shortest possible\, measuring dista nce with point-to-point\nwalking routes obtained from Google Maps. Along t he way\, we discuss the history\,\napplications\, and computation of this fascinating problem.\n \nBiographical Sketch\nWilliam Cook is a University Professor in Combinatorics and Optimization at the University of Waterloo \, where he received his Ph.D. in 1983. Bill\nwas elected a SIAM Fellow i n 2009\, an INFORMS Fellow in 2010\, a member of the National Academy of E ngineering in 2011\, and an American\nMathematics Society Fellow in 2012. He is the author of the popular book In Pursuit of the Traveling Salesman : Mathematics at the Limits of Computation.\nBill is a former Editor-in-Ch ief of the journals Mathematical Programming (Series A and B) and Mathemat ical Programming Computation. He is the past chair and current vice-chair of the Mathematical Optimization Society and a past chair of the INFORMS C omputing Society. DTSTART;TZID=America/New_York:20180222T133000 DTEND;TZID=America/New_York:20180222T143000 SEQUENCE:0 SUMMARY:The Goldman Distinguished Lecture Series: William Cook (University of Waterloo) @ Shaffer 100 URL:https://engineering.jhu.edu/ams/events/ams-seminar-william-cook-univers ity-waterloo-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Information\, Computation\, Optimization: Connecting the dots i
n the Traveling Salesman Problem

**Abstract:** Few ma
th models scream impossible as loudly as the traveling salesman problem.**\nGiven n cities\, the TSP asks for the shortest route to take you to
all of them.\nEasy to state\, but if P ≠ NP then no solution method
can have good asymptotic\nperformance as n goes off to infinity. The
popular interpretation is that we simply\ncannot solve realistic ex
amples. But this skips over nearly 70 years of intense\nmathematical
study. Indeed\, in 1949 Julia Robinson described the TSP challenge in\npractical terms: “Since there are only a finite number of paths to con
sider\, the\nproblem consists in finding a method for picking out th
e optimal path when n is\nmoderately large\, say n = 50.” She went o
n to propose a linear programming attack\nthat was adopted by her RA
ND colleagues Dantzig\, Fulkerson\, and Johnson several\nyears later
.**

Following in the footsteps of these giants\, we show that a cert
ain tour of 49\,603

\nhistoric sites in the US is shortest possible\,
measuring distance with point-to-point

\nwalking routes obtained fro
m Google Maps. Along the way\, we discuss the history\,

\napplication
s\, and computation of this fascinating problem.

\n

*Biographical Sketch*

William Cook is a University Professor in Combinatorics a nd Optimization at the University of Waterloo\, where he received his Ph.D . in 1983. Bill

\nwas e lected a SIAM Fellow in 2009\, an INFORMS Fellow in 2010\, a member of the National Academy of Engineering in 2011\, and an American

\nMathematics Society Fellow in 2012. He is the author of the popular book In Pursuit of the Traveling Salesman: Mathematics at the Limits of Computation.

\nBill is a former Editor-in-Chief of the journals Math ematical Programming (Series A and B) and Mathematical Programming Computa tion. He is the past chair and current vice-chair of the Mathematical Opti mization Society and a past chair of the INFORMS Computing Society.

\n< /BODY> END:VEVENT BEGIN:VEVENT UID:ai1ec-10858@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION: Title: PetuumMed: algorithm and system for EHR-based medical decision-making\n \nAbstract:\nWith the rapid growth of electronic health records (EHRs) and the advancement of machine learning technologies\, need s for AI-enabled clinical decision-making support is emerging. In this tal k\, I will present some recent work toward these needs at Petuum Inc. wher e an integrative system that distills insights from large-scale and hetero geneous patient data\, as well as learns and integrates medical knowledge from broader sources such as the literatures and domain experts\, and empo wers medical professionals to make accurate and efficient decisions within the clinical flow\, is being built. I will discuss several aspects of pra ctical clinical decision-support\, such as real-time information extractio n from clinical notes and images\, diagnosis and treatment recommendation\ , automatic report generation and ICD code filling\; and the algorithmic a nd computational challenges behind production-quality solution to these pr oblems. DTSTART;TZID=America/New_York:20180301T133000 DTEND;TZID=America/New_York:20180301T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Eric Xing (Carnegie Mellon) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-eric-xing-carnegie-m ellon-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n** Titl
e: **PetuumMed: algorithm and system for EHR-based medical decisio
n-making

\n

**Abstract:**

With the rap id growth of electronic health records (EHRs) and the advancement of machi ne learning technologies\, needs for AI-enabled clinical decision-making s upport is emerging. In this talk\, I will present some recent work toward these needs at Petuum Inc. where an integrative system that distills insig hts from large-scale and heterogeneous patient data\, as well as learns an d integrates medical knowledge from broader sources such as the literature s and domain experts\, and empowers medical professionals to make accurate and efficient decisions within the clinical flow\, is being built. I will discuss several aspects of practical clinical decision-support\, such as real-time information extraction from clinical notes and images\, diagnosi s and treatment recommendation\, automatic report generation and ICD code filling\; and the algorithmic and computational challenges behind producti on-quality solution to these problems.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11131@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Variance swap DTSTART;TZID=America/New_York:20180306T133000 DTEND;TZID=America/New_York:20180306T143000 SEQUENCE:0 SUMMARY:Financial Math Seminar: John Miller (JHU) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/financial-math-seminar-john-mill er-jhu-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Variance swap

Title: Compar ing relaxations via volume for nonconvex optimization

\nAbstract: Pr actical exact methods for global optimization of mixed-integer nonlinear o ptimization formulations rely on convex relaxation. Then\, one way or anot her (via refinement and/or disjunction)\, global optimality is sought. Suc cess of this paradigm depends on balancing tightness and lightness of rela xations. We will investigate this from a mathematical viewpoint\, comparin g polyhedral relaxations via their volumes. Specifically\, I will present some results concerning: fixed charge problems\, vertex packing in graphs\ , boolean quadratic formulations\, and convexification of monomials in the context of spatial branch-and-bound” for factorable formulations. Our res ults can be employed by users (at the modeling level) and by algorithm des igners/implementers alike.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11106@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:TBA DTSTART;TZID=America/New_York:20180314T150000 DTEND;TZID=America/New_York:20180314T160000 SEQUENCE:0 SUMMARY:Data Science Seminar: Edriss Titi (Texas A&M University) @ Shaffer 304 URL:https://engineering.jhu.edu/ams/events/data-science-seminar-edriss-titi -texas-am-university/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\nTBA

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10867@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Bayesian monotone regression: Rates\, coverage and tests \nAbstract:\nShape restrictions such as monotonicity often naturally arise . In this talk we consider a Bayesian approach to monotone nonparametric r egression with normal error. We assign a prior through piecewise constant functions and impose a conjugate normal prior on the coefficient. Since th e resulting functions need not be monotone\, we project samples from the p osterior on the allowed parameter space to construct a “projection posteri or”. We first obtain contraction rates of the projection posterior distrib utions under various settings. We next obtain the limit posterior distribu tion of a suitably centered and scaled posterior distribution for the func tion value at a point. The limit distribution has some interesting similar ity and difference with the corresponding limit distribution for the maxim um likelihood estimator. By comparing the quantiles of these two distribut ions\, we observe an interesting new phenomenon that coverage of a credibl e interval may be more than the credibility level\, an exact opposite of a phenomenon observed by Cox for smooth regression. We describe a recalibra tion strategy to modify the credible interval to meet the correct level of coverage. Finally we discuss asymptotic properties of Bayes tests for mon otonicity.\nThis talk is based on joint work with Moumita Chakraborty\, a doctoral student at North Carolina State University. DTSTART;TZID=America/New_York:20180315T133000 DTEND;TZID=America/New_York:20180315T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Subhashis Ghoshal (North Carolina State University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-tba-whitehead-304-5/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
: **Bayesian monotone regression: Rates\, coverage and tests

Shape restrictions such as monotonici ty often naturally arise. In this talk we consider a Bayesian approach to monotone nonparametric regression with normal error. We assign a prior thr ough piecewise constant functions and impose a conjugate normal prior on t he coefficient. Since the resulting functions need not be monotone\, we pr oject samples from the posterior on the allowed parameter space to constru ct a “projection posterior”. We first obtain contraction rates of the proj ection posterior distributions under various settings. We next obtain the limit posterior distribution of a suitably centered and scaled posterior d istribution for the function value at a point. The limit distribution has some interesting similarity and difference with the corresponding limit di stribution for the maximum likelihood estimator. By comparing the quantile s of these two distributions\, we observe an interesting new phenomenon th at coverage of a credible interval may be more than the credibility level\ , an exact opposite of a phenomenon observed by Cox for smooth regression. We describe a recalibration strategy to modify the credible interval to m eet the correct level of coverage. Finally we discuss asymptotic propertie s of Bayes tests for monotonicity.

\nThis talk is based on joint wor k with Moumita Chakraborty\, a doctoral student at North Carolina State Un iversity.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11155@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Presentation – Career Opportunities at the NSA Targeting AMS\, CS\, ECE\, JHUISI \n \nWhen: Wednesday\, March 28\, 2018 5:30 pm – 8:00 pm EDT\nWhere: Hodson\, 110\, Baltimore\, MD 21218\, United States\nThe Nat ional Security Agency (NSA) currently has opportunities for highly motivat ed researchers to provide expertise\, guidance and support to the developm ent and implementation of mission capabilities that align with mission dri ven challenges.\nThe Advanced Computing Systems Research Program (ACS) at NSA is looking to hire talented researchers for a variety of positions. Th e ACS mission is to collaborate with industry\, academia\, and the governm ent to drive innovative research that will improve advanced computing syst ems for a range of mission applications including cybersecurity\, cryptana lysis\, and complex data analytics. The ACS has significant research proje cts in neuromorphic and probabilistic computing\, novel computer architect ures and technologies\, advanced modeling and simulation\, energy efficien cy\, productivity\, and resilience.\nDr. David J. Mountain will describe o pportunities at NSA\, using his 36 year career as an example. He will also provide an overview of the ACS program and highlight specific research po sitions currently available. This will be followed by an open Q&A.\nDr. Mo untain is the Senior Technical Director at the Laboratory for Physical Sci ences at Research Park\, a Department of Defense research lab in Catonsvil le\, MD. He received a BS in Electrical Engineering from the University of Notre Dame in 1982\, an MS in Electrical Engineering from the University of Maryland\, College Park\, in 1986\, and a PhD in Computer Engineering f rom the University of Maryland\, Baltimore County\, in 2017. His personal research projects have included radiation effects studies\, hot carrier re liability characterization\, and chip-on-flex process development utilizin g ultra-thin circuits. He has been actively involved with 3D electronics r esearch for 25 years and is presently focused on specialized architectures to support advanced neural networks and tensor analysis. Dr. Mountain is the author of more than two dozen papers\, has been awarded eight patents\ , and is a Senior Member of the IEEE.\nStudents RSVP @ https://app.joinhan dshake.com/events/135261\nFaculty and Staff RSVP @ https://www.eventbrite. com/e/career-opportunities-at-the-nsa-targeting-cs-ams-ece-jhuisi-informat ion-session-tickets-43894265931?aff=affiliate1\n\nNote:\n\n*Food will be served at this even during the first 30 minutes\, Please arrive early. The program will begin promptly at 6pm!*\nFor additional information about th is event\, please contact Dr. Antwan D. Clark at aclark66@jhu.edu.\n\n DTSTART;TZID=America/New_York:20180328T173000 DTEND;TZID=America/New_York:20180328T200000 SEQUENCE:0 SUMMARY:Dr. David J. Mountain will present on Career Opportunities at the N SA in Hodson 110 URL:https://engineering.jhu.edu/ams/events/dr-david-j-mountain-will-present -on-career-opportunities-at-the-nsa-in-hodson-110/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Prese
ntation – Career Opportunities at the NSA Targeting AMS\, CS\, ECE\, JHUIS
I **

\n

When: Wednesday\, March 28\, 2018 5:30 pm – 8:00 pm EDT

\nWhere: Hodson\, 110\, Baltimore\, MD 21218\, United St ates

\nThe National Security Agency (NSA) currently has opportuniti es for highly motivated researchers to provide expertise\, guidance and su pport to the development and implementation of mission capabilities that a lign with mission driven challenges.

\nThe Advanced Computing System s Research Program (ACS) at NSA is looking to hire talented researchers fo r a variety of positions. The ACS mission is to collaborate with industry\ , academia\, and the government to drive innovative research that will imp rove advanced computing systems for a range of mission applications includ ing cybersecurity\, cryptanalysis\, and complex data analytics. The ACS ha s significant research projects in neuromorphic and probabilistic computin g\, novel computer architectures and technologies\, advanced modeling and simulation\, energy efficiency\, productivity\, and resilience.

\nDr . David J. Mountain will describe opportunities at NSA\, using his 36 year career as an example. He will also provide an overview of the ACS program and highlight specific research positions currently available. This will be followed by an open Q&A.

\nDr. Mountain is the Senior Technical D irector at the Laboratory for Physical Sciences at Research Park\, a Depar tment of Defense research lab in Catonsville\, MD. He received a BS in Ele ctrical Engineering from the University of Notre Dame in 1982\, an MS in E lectrical Engineering from the University of Maryland\, College Park\, in 1986\, and a PhD in Computer Engineering from the University of Maryland\, Baltimore County\, in 2017. His personal research projects have included radiation effects studies\, hot carrier reliability characterization\, and chip-on-flex process development utilizing ultra-thin circuits. He has be en actively involved with 3D electronics research for 25 years and is pres ently focused on specialized architectures to support advanced neural netw orks and tensor analysis. Dr. Mountain is the author of more than two doze n papers\, has been awarded eight patents\, and is a Senior Member of the IEEE.

\nStudents RSVP @ https://app.joinhandshake.com/events/135261

\nFaculty and Staff RSVP @ https://www.eventbrite.com/e/career-opportunities-a t-the-nsa-targeting-cs-ams-ece-jhuisi-information-session-tickets-43894265 931?aff=affiliate1

\n

\nNote:

\n- \n
- *
**Foo d will be served at this even during the first 30 minutes\, Please arrive early. The program will begin promptly at 6pm!*** \n - For add itional information about this event\, please contact Dr. Antwan D. Clark at aclark66@jhu.edu. \n

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10871@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Real and Artificial Neural Networks\nAbstract: Lately\, just about everybody has been thinking about deep neural networks (DNNs). Do they work? If so\, how? Do they overfit? If not\, why? I will discuss t hese questions and suggest some uncomplicated answers\, at least as a firs t approximation. Turning to biological learning\, I will argue that the st ubborn gap between human and machine performance\, when it comes to interp retation (as opposed to classification)\, can not be substantially closed without architectures that support stronger representations. In particular \, how are we to accommodate the rich collection of spatial and abstract r elationships (‘on’ or ‘inside’\, ‘talking’ or ‘holding hands’\, ‘same’ or ‘different’) that bind parts and objects and define context? I will propos e that the nonlinearities of dendritic integration in real neurons is the missing ingredient in artificial neurons. I will suggest a mechanism for e mbedding relationships in a generative network. DTSTART;TZID=America/New_York:20180329T133000 DTEND;TZID=America/New_York:20180329T143000 SEQUENCE:0 SUMMARY:Duncan Lecture Series- AMS Seminar: Stuart Geman (Brown University) @ Shaffer 100 URL:https://engineering.jhu.edu/ams/events/ams-seminar-stuart-geman-brown-u niversity-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
: **Real and Artificial Neural Networks

**Abstract: <
/strong>Lately\, just about everybody has been thinking about deep neural
networks (DNNs). Do they work? If so\, how? Do they overfit? If not\, why?
I will discuss these questions and suggest some uncomplicated answers\, a
t least as a first approximation. Turning to biological learning\, I will
argue that the stubborn gap between human and machine performance\, when i
t comes to interpretation (as opposed to classification)\, can not be subs
tantially closed without architectures that support stronger representatio
ns. In particular\, how are we to accommodate the rich collection of spati
al and abstract relationships (‘on’ or ‘inside’\, ‘talking’ or ‘holding ha
nds’\, ‘same’ or ‘different’) that bind parts and objects and define conte
xt? I will propose that the nonlinearities of dendritic integration in rea
l neurons is the missing ingredient in artificial neurons. I will suggest
a mechanism for embedding relationships in a generative network.**

**Title
: **Random Walks on Secondary Structure and the Folding of RNA

**Abstract:**

What is the correct characterizati on of the native structure of a protein or a non-coding RNA molecule? Is i t the minimum energy state\, a metastable state\, or a sample from thermal equilibrium? The problem is unsettled and a topic of enduring debate. The “agnostic” approach is through molecular dynamics—make a proper accountin g of the intramolecular forces and interactions with the surrounding liqui d (mostly water)\, write down the corresponding kinetic equation (e.g. a L angevin equation)\, and simulate folding. But this usually isn’t practical \, and approximations need to be made. I will explore an approximation in which the bulk of the folding process is replaced by a random walk\, with discrete moves from one secondary structure to another. An analytic result identifies conditions for guaranteed accuracy. Simulation results\, runni ng the approximation against an optimized integrator of the Langevin equat ion\, achieve the expected accuracy\, but 100 times faster.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11203@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Machine Learning For Health Care\nAbstract:\nWe will pre sent multiple ways in which healthcare data is acquired and machine learni ng methods are currently being introduced into clinical settings.\nThis wi ll include:\n1.)Modeling the prediction of disease\, including Sepsis\, an d ways in which the best treatment decisions for Sepsis patients can be ma de\, from electronic health record (EHR) data using Gaussian processes and deep learning methods.\n2) Predicting surgical complications and transfer learning methods for combining databases\n3) Using mobile apps and integr ated sensors for improving the granularity of recorded health data for chr onic conditions. Current work in these areas will be presented and the fut ure of machine learning contributions to the field will be discussed. DTSTART;TZID=America/New_York:20180404T120000 DTEND;TZID=America/New_York:20180404T130000 SEQUENCE:0 SUMMARY:AMS & BME Presents Speaker: Katherine Heller @ Shaffer 101 URL:https://engineering.jhu.edu/ams/events/ams-bme-presents-speaker-katheri ne-heller-shaffer-101/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Machine Learning For Health Care

**Abstract:**

We will present multiple ways in which healthcare data is acq uired and machine learning methods are currently being introduced into cli nical settings.

\nThis will include:

\n1.)Modeling the predicti
on of disease\, including Sepsis\, and ways in which the best treatment de
cisions for Sepsis patients can be made\, from electronic health record (E
HR) data using Gaussian processes and deep learning methods.

2) Pr edicting surgical complications and transfer learning methods for combinin g databases

\n3) Using mobile apps and integrated sensors for improv ing the granularity of recorded health data for chronic conditions. Curren t work in these areas will be presented and the future of machine learning contributions to the field will be discussed.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10878@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Adaptive Robust Control Under Model Uncertainty\nAbstrac t: We propose a new methodology\, called adaptive robust control\, for so lving a discrete-time Markovian control problem subject to Knightian uncer tainty. We apply the general framework to a financial hedging problem wher e the uncertainty comes from the fact that the true law of the underlying model is only known to belong to a certain family of probability laws. We provide a learning algorithm that reduces the model uncertainty through pr ogressive learning about the unknow system. One of the pillars in the prop osed methodology is the recursive construction of the confidence sets for the unknown parameter. This allows\, in particular\, to establish the corr esponding Bellman system of equations. DTSTART;TZID=America/New_York:20180405T133000 DTEND;TZID=America/New_York:20180405T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Igor Cialenco (Illinois Institute of Technology) @ Whi tehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-igor-cialenco-illino is-institute-technology-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
: **Adaptive Robust Control Under Model Uncertainty

**Abstract: **We propose a new methodology\, called adaptive robust
control\, for solving a discrete-time Markovian control problem subject t
o Knightian uncertainty. We apply the general framework to a financial hed
ging problem where the uncertainty comes from the fact that the true law o
f the underlying model is only known to belong to a certain family of prob
ability laws. We provide a learning algorithm that reduces the model uncer
tainty through progressive learning about the unknow system. One of the pi
llars in the proposed methodology is the recursive construction of the con
fidence sets for the unknown parameter. This allows\, in particular\, to e
stablish the corresponding Bellman system of equations.

Mark your calendars fo r the 5th USA Science & Engineering Festival Expo on April 7-8\, 2018! Exp lore 3\,000 hands-on exhibits from the world’s leading scientific and engi neering societies\, universities\, government agencies\, high-tech corpora tions and STEM organizations. The two-day Expo is perfect for children\, t eens\, and families who want to inspire their curious minds.

\n

\n**W
hen: Saturday 10 am- 6 pm and Sunday 10 am- 4 pm **

Join 35 0K+ attendees to celebrate science at the Expo and engage in activities wi th some of the biggest names in STEM. Hear stories of inspiration and cour age\, participate in mind-boggling experiments and rock out to science dur ing our incredible stage shows.

\n**Date<
/strong>: Monday\, April 9th\, 2018\n Time: 7:00 pm:
Enjoy refreshments and snacks with students and faculty.\n7:30 pm:
The Mathemagics performance begins!\nLocation: John
s Hopkins University\, Homewood Campus\; Hodson Hall\, Room 110**

\n

\nOr E-mail us at husam.jhu@gmail.com

\n\n

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10983@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: TBA\n \nAbstract: TBA DTSTART;TZID=America/New_York:20180410T133000 DTEND;TZID=America/New_York:20180410T143000 SEQUENCE:0 SUMMARY:Financial Math Seminar: Sebastien Bossu (NYU Courant & JHU Carey Sc hool) @ Whitehead Hall 304 URL:https://engineering.jhu.edu/ams/events/financial-math-seminar-sebastien -bossu-nyu-courant-jhu-carey-school-whitehead-hall-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
:** TBA

\n

**Abstract:** TBA

**Title
: **Optimal Portfolio under Fractional Stochastic Environment

**Abstract:**

Rough stochastic volatility models h ave attracted a lot of attention recently\, in particular for the linear o ption pricing problem. In this talk\, starting with power utilities\, we p ropose to use a martingale distortion representation of the optimal value function for the nonlinear asset allocation problem in a (non-Markovian) f ractional stochastic environment (for all Hurst index $H \\in (0\, 1)$). W e rigorously establish a first order approximation of the optimal value\, when the return and volatility of the underlying asset are functions of a stationary slowly varying fractional Ornstein-Uhlenbeck process. We prove that this approximation can be also generated by the zeroth order trading strategy providing an explicit strategy which is asymptotically optimal in all admissible controls. Furthermore\, we extend the discussion to genera l utility functions\, and obtain the asymptotic optimality of this strateg y in a specific family of admissible strategies. If time permits\, we will also discuss the problem under fast mean-reverting fractional stochastic environment.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-11005@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: On the joint calibration of SPX and VIX options\nAbstrac t: Since VIX options started trading in 2006\, many researchers have attem pted to build a model for the SPX that jointly calibrates to SPX and VIX o ptions. In 2008\, Jim Gatheral showed that a diffusive model could approxi mately\, but not exactly\, fit both markets. Later\, others have argued th at jumps in the SPX were needed to jointly calibrate both markets. We revi sit this problem\, asking the following questions: Does there exist a cont inuous model on the SPX that jointly calibrates to SPX options\, VIX futur es\, and VIX options? If so\, how to build one such model? If not\, why? W e present a novel approach based on the SPX smile calibration condition. I n the limiting case of instantaneous VIX\, the answers are clear and invol ve the timewise convex ordering of two distributions (local variance and instantaneous variance) and a novel application of martingale transport to finance. The real case of a 30-day VIX is more involved\, as time-averagi ng and projection onto a filtration can undo convex ordering. We observe t hat in usual market conditions the distribution of VIX^2 in the local vola tility model and the market-implied distribution of VIX^2 are not in conve x order\, and we show that fast mean-reverting volatility models and rough volatility models are able to reproduce this surprising behavior.\n DTSTART;TZID=America/New_York:20180417T133000 DTEND;TZID=America/New_York:20180417T143000 SEQUENCE:0 SUMMARY:Financial Math Seminar: Dr. Julien Guyon (Columbia & NYU ) @ Whiteh ead 304 URL:https://engineering.jhu.edu/ams/events/financial-math-seminar-dr-julien -guyon-columbia-nyu-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
: **On the joint calibration of SPX and VIX options

**Abstract: **Since VIX options started trading in 2006\, many resea
rchers have attempted to build a model for the SPX that jointly calibrates
to SPX and VIX options. In 2008\, Jim Gatheral showed that a diffusive mo
del could approximately\, but not exactly\, fit both markets. Later\, othe
rs have argued that jumps in the SPX were needed to jointly calibrate both
markets. We revisit this problem\, asking the following questions: Does t
here exist a continuous model on the SPX that jointly calibrates to SPX op
tions\, VIX futures\, and VIX options? If so\, how to build one such model
? If not\, why? We present a novel approach based on the SPX smile calibra
tion condition. In the limiting case of instantaneous VIX\, the answers ar
e clear and involve the timewise convex ordering of two distributions (loc
al variance and instantaneous variance) and a novel application of martin
gale transport to finance. The real case of a 30-day VIX is more involved\
, as time-averaging and projection onto a filtration can undo convex order
ing. We observe that in usual market conditions the distribution of VIX^2
in the local volatility model and the market-implied distribution of VIX^2
are not in convex order\, and we show that fast mean-reverting volatility
models and rough volatility models are able to reproduce this surprising
behavior.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10880@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Characterizing the Worst-Case Performance of Algorithms for Nonconvex Optimization\nAbstract: Motivated by various applications\, e.g.\, in data science\, there has been increasing interest in numerical m ethods for minimizing nonconvex functions. Users of such methods often cho ose one algorithm versus another due to worst-case complexity guarantees\, which in contemporary analyses bound the number of iterations required un til a first- or second-order stationarity condition is approximately satis fied. In this talk\, we question whether this is indeed the best manner in which to compare algorithms\, especially since the worst-case behavior of an algorithm is often only seen when it is employed to minimize pedagogic al examples which are quite distinct from functions seen in normal practic e. We propose a new strategy for characterizing algorithms that attempts t o better represent algorithmic behavior in real-world settings. DTSTART;TZID=America/New_York:20180419T133000 DTEND;TZID=America/New_York:20180419T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Frank Curtis (Lehigh University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-frank-curtis-lehigh- university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
: **Characterizing the Worst-Case Performance of Algorithms for Non
convex Optimization

**Abstract: **Motivated by variou
s applications\, e.g.\, in data science\, there has been increasing intere
st in numerical methods for minimizing nonconvex functions. Users of such
methods often choose one algorithm versus another due to worst-case comple
xity guarantees\, which in contemporary analyses bound the number of itera
tions required until a first- or second-order stationarity condition is ap
proximately satisfied. In this talk\, we question whether this is indeed t
he best manner in which to compare algorithms\, especially since the worst
-case behavior of an algorithm is often only seen when it is employed to m
inimize pedagogical examples which are quite distinct from functions seen
in normal practice. We propose a new strategy for characterizing algorithm
s that attempts to better represent algorithmic behavior in real-world set
tings.

**Title
**: Merchant Options of Energy Trading Network

\n

< /p>\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10884@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: “Statistical network modeling via exchangeable interacti on processes”\n \nAbstract:\nMany modern network datasets arise from proce sses of interactions in a population\, such as phone calls\, e-mail exchan ges\, co-authorships\, and professional collaborations. In such interactio n networks\, the interactions comprise the fundamental statistical units\, making a framework for interaction-labeled networks more appropriate for statistical analysis. In this talk\, we present exchangeable interaction n etwork models and explore their basic statistical properties. These models allow for sparsity and power law degree distributions\, both of which are widely observed empirical network properties. I will start by presenting the Hollywood model\, which is computationally tractable\, admits a clear interpretation\, exhibits good theoretical properties\, and performs reaso nably well in estimation and prediction.\nIn many settings\, the series of interactions are structured. E-mail exchanges\, for example\, have a sing le sender and potentially multiple receivers. I will introduce hierarchica l exchangeable interaction models for the study of structured interaction networks. In particular\, I will introduce the Enron model as a canonical example\, which partially pools information via a latent\, shared populati on-level distribution. A detailed simulation study and supporting theoreti cal analysis provide clear model interpretation\, and establish global pow er-law degree distributions. A computationally tractable Gibbs sampling al gorithm is derived. Inference will be shown on the Enron e-mail dataset. I will end with a discussion of how to perform posterior predictive checks on interaction data. Using these proposed checks\, I will show that the m odel fits the data well. DTSTART;TZID=America/New_York:20180426T133000 DTEND;TZID=America/New_York:20180426T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Walter Dempsey (Harvard University) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-walter-dempsey-harva rd-university-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n

**Title
: **“Statistical network modeling via exchangeable interaction proc
esses”

\n

**Abstract:**

Many modern ne twork datasets arise from processes of interactions in a population\, such as phone calls\, e-mail exchanges\, co-authorships\, and professional col laborations. In such interaction networks\, the interactions comprise the fundamental statistical units\, making a framework for interaction-labeled networks more appropriate for statistical analysis. In this talk\, we pre sent exchangeable interaction network models and explore their basic stati stical properties. These models allow for sparsity and power law degree di stributions\, both of which are widely observed empirical network properti es. I will start by presenting the Hollywood model\, which is computationa lly tractable\, admits a clear interpretation\, exhibits good theoretical properties\, and performs reasonably well in estimation and prediction.

\nIn many settings\, the series of interactions are structured. E-mail exchanges\, for example\, have a single sender and potentially multiple r eceivers. I will introduce hierarchical exchangeable interaction models fo r the study of structured interaction networks. In particular\, I will int roduce the Enron model as a canonical example\, which partially pools info rmation via a latent\, shared population-level distribution. A detailed si mulation study and supporting theoretical analysis provide clear model int erpretation\, and establish global power-law degree distributions. A compu tationally tractable Gibbs sampling algorithm is derived. Inference will b e shown on the Enron e-mail dataset. I will end with a discussion of how to perform posterior predictive checks on interaction data. Using these pr oposed checks\, I will show that the model fits the data well.

\n END:VEVENT BEGIN:VEVENT UID:ai1ec-10888@engineering.jhu.edu/ams DTSTAMP:20180715T165111Z CATEGORIES: CONTACT: DESCRIPTION:Title: Computational Anatomy: Structuring and Searching Shape S paces.\nAbstract: 100 years after the celebrated D’Arcy Thompson’s master piece “Growth and Forms”\, the modeling and the understanding of both vari ability and dynamics of related biological shapes are still particularly c hallenging from both modeling and computational point of view. The luminou s idea of his “Theory of Transformations” has been turned within the digit al era into a versatile mathematical and computational framework coined a s diffeomorphometry and living in the vicinity of riemannian geometry\, fl uid dynamics\, optimal control and statistics. We will discuss about the m athematical side of this framework as well as some of challenges that stil l need to be faced. DTSTART;TZID=America/New_York:20180503T133000 DTEND;TZID=America/New_York:20180503T143000 SEQUENCE:0 SUMMARY:AMS Seminar: Alain Trouve’ (Ecole Normale Supe’rieure) @ Whitehead 304 URL:https://engineering.jhu.edu/ams/events/ams-seminar-alain-trouve-ecole-n ormale-superieure-whitehead-304/ X-COST-TYPE:free X-ALT-DESC;FMTTYPE=text/html:\\n\\n\\n**Title
:** Computational Anatomy: Structuring and Searching Shape Spaces.<
/p>\n

**Abstract: **100 years after the celebrated D’Arcy
Thompson’s masterpiece “Growth and Forms”\, the modeling and the understan
ding of both variability and dynamics of related biological shapes are sti
ll particularly challenging from both modeling and computational point of
view. The luminous idea of his “Theory of Transformations” has been turned
within the digital era into a versatile mathematical and computational fr
amework coined as diffeomorphometry and living in the vicinity of riemann
ian geometry\, fluid dynamics\, optimal control and statistics. We will di
scuss about the mathematical side of this framework as well as some of cha
llenges that still need to be faced.