Decisions under non-stationarity
Planning and reinforcement learning in large, stochastic, partially observable environments whose dynamics drift, detecting the change and then adapting online.
Robust & Adaptive Intelligence Lab
Decision-making for societal-scale systems.
I am Ayan Mukhopadhyay, Assistant Professor of Computer Science at the College of William & Mary, where I direct the RAIL group. We build robust, adaptive, and interpretable algorithms for decision-making under uncertainty, and deploy them in emergency response, public transit, energy, public health, and conservation.
Featured project
Visiting from the NSF Civic Innovation Challenge PI meeting?
SENTRY
Sensor-enabled enforcement of non-compliant traffic right-of-way closures, our NSF CIVIC project with the city of Nashville. The project site walks through how detection, prioritization, and officer dispatch fit together, and includes a demo you can try.
What we work on
Algorithms that have to work outside the lab
Our problems come from partners who operate real systems: fire departments, transit authorities, public-health non-profits, and utilities. That constraint shapes the methods: the models must stay calibrated as the world shifts, the decisions must be explainable to the people who act on them, and the whole pipeline has to keep running when conditions are nothing like the training data.
Multi-agent planning and decision-making
Many agents acting at once under partial information and shared constraints. In transportation, healthcare, and sustainable energy, a decision made for one of them changes the problem for the rest.
Interpretable planning
Explanations a dispatcher or planner can interrogate, built from formal logic and language models rather than post-hoc rationalization.
News
Recent updates
- Jun 2026 Invited talk on Decision-Making for Societal-Scale Systems at IFORS 2026 in Vienna, and a tutorial on non-stationary decision-making at CPS-IoT Week 2026.
- Apr 2026 Artificial Intelligence and Society, an edited volume with Yevgeniy Vorobeychik, is coming from River Publishers this fall.
- Feb 2026 Six papers accepted to ICCPS 2026, plus papers at ICLR 2026 and AAMAS 2026.
- Sep 2025 Two papers accepted at NeurIPS 2025: ESCORT (belief representation for POMDPs) and NS-Gym.
- 2025 Two new NSF awards, a $1.25 million grant with George Mason and Vanderbilt on designing resilient sheltering for emergency evacuations (role: W&M PI), a $700,000 grant on designing data-driven monitoring of right-of-way permits with Vanderbilt (role: Lead PI).
- 2026 We are setting up the RAIL Laboratory in the College of William & Mary, in the School of Computing, Data Science, and Physics.