Saria and Van Durme of Computer Science Honored With Google Faculty Research Awards

March 7, 2014

Suchi Saria and Benjamin Van Durme, both faculty members in the Department of ComputerScience, have received Google Faculty Research Awards. Saria is an assistant professor and Van Durme is an assistant research professor.

Google’s Faculty Research Award program provides world-class, full-time faculty members at top universities around the world with support for cutting-edge research in computer science, engineering, and related fields in areas of interest to Google.

Suchi Saria of the Department of Computer Science

Suchi Saria

Saria’s Google Faculty Research Award provides one-year unrestricted funding to support her work in “Machine Learning for Knowledge Extraction from Electronic Medical Records,” and Ben’s award will support his research in “Integrating Structured and Unstructured Evidence for Question Answering.” Her research addresses the problem of improving patient care using data-driven tools derived from large scale electronic medical record (EMR) databases. A number of different clinical measurements are made and collected within EMRs as part of routine care. Tools that can detect downward progression due to a disease earlier can allow a clinician to intervene sooner. With this funding, Suchi will develop probabilistic models that integrate heterogeneous measurements present in electronic medical records to discover canonical patterns of development. She will apply these towards early detection of acute conditions in the inpatient setting.

Benjamin Van Durme of the Department of Computer Science

Benjamin Van Durme

Van Durme’s Google funding will enable him to pursue research in Question Answering (QA). Over the past decade, QA has pursued two paths: text-based QA driven by years of Text REtrieval Conference (TREC) evaluation; and knowledge-based QA driven by the rise of semantic parsing and resources such as Freebase. Van Durme proposes an alignment framework that finds answers over combined evidence from both structured and unstructured sources, with the aim of solving more complex queries at both higher precision and recall values.

Back to top