This seminar explores two interconnected problems in sequential decision-making relevant to autonomous and human-in-the-loop systems. The first part discusses models for sequential decision making involving word-of-mouth social learning, and we show that such protocols exhibit a significant slow down in the learning rate. The second part of the talk discusses inverse reinforcement learning (IRL) - how can we infer the objectives of an agent by observing its decisions? We connect IRL to foundational ideas in microeconomics via revealed preferences, and present both classical and Bayesian IRL approaches for detecting the underlying utility. Finally, we briefly discuss adaptive IRL methods based on passive stochastic gradient algorithms.
Vikram Krishamurthy is a professor of Electrical and Computer Engineering at Cornell University. His research interests are in statistical signal processing, stochastic optimization and adaptive sensing.