AMS Weekly Seminar | Chinmay Maheshwari

Location: Krieger 205

When: September 18th at 1:30 p.m.

Title: Design of Dynamic Multi-agent Autonomous Systems for Societal Transformation

Abstract: Data science and AI are driving new services in robotics, mobility, energy, and online marketplaces. While autonomous agents have traditionally been designed to operate in isolation, deploying them in multi-agent environments—often subject to dynamic evolution and resource constraints—introduces new theoretical, algorithmic, and societal challenges. Without careful design, these interactions can result in inefficiencies, inequities, and safety risks at the societal scale. In this talk, I will present new frameworks for the design and analysis of Dynamic Multi-agent Autonomous Systems (DMAS), highlighting several research vignettes that address these challenges.

First, I will begin by introducing Markov Near-Potential Functions (MNPFs), a novel theoretical tool to design and analyze multi-agent learning in dynamic environments. I will highlight the role of MNPFs in obtaining the first characterization of the convergent set of decentralized (multi-agent) reinforcement learning algorithms in general-sum Markov games. Additionally, I will demonstrate the utility of MNPFs in the design of competitive strategies in multi-agent systems.  A key challenge, however, is that computing MNPFs requires solving a convex–non-concave min–max optimization problem, a class of problems for which no exact solution methods previously existed. To address this, I will present a novel algorithmic framework which not only enables the computation of MNPFs but also advances the broader theory of min–max optimization.

Next, I will present a new algorithmic framework for decentralized multi-agent learning in resource-constrained environments, focusing on two-sided matching markets. Here, I will describe a novel combination of stochastic and adversarial bandit methods that disentangles the challenges of learning from noisy feedback from the challenges of competing for scarce resources.

Finally, through some motivating case studies in societal systems, I will discuss the role of data-driven economic incentives and markets in aligning multi-agent systems for efficient, fair, and safe societal outcomes.

Bio: Chinmay Maheshwari is an Assistant Professor in the Department of Electrical and Computer Engineering at Johns Hopkins University. He is also a member of the Data Science and AI Institute. He obtained his PhD in Electrical Engineering and Computer Sciences from the University of California, Berkeley in 2025. He completed his undergraduate studies (B.Tech and M.Tech) at the Indian Institute of Technology (IIT) Bombay in 2019, where he received the Institute Academic Medal. He was selected as an NSF CPS Rising Star in 2025 and received the Lotfi A. Zadeh Prize in 2025.

Zoom link: https://wse.zoom.us/j/93600407710?pwd=JBL8VsObRxX6MkhdjAUxCadqJDoZrZ.1