Professor Mahadevan’s research interests span several subfields of artificial intelligence and computer science, including machine learning, multi-agent systems, planning, perception, and robotics. His research in machine learning has been eclectic, ranging from pioneering work in explanation-based learning where his thesis introduced the model of learning apprentices for knowledge acquisition from experts, to the first rigorous study of concept learning with prior determination knowledge using the framework of Probably Approximately Correct (PAC) learning. Over the past decade, his research has centered around a general framework for autonomous learning and sequential decision-making, which studies how agents embedded in real-world environments can acquire knowledge on how to act from a stream of noisy percepts. The framework is rigorously validated using temporal statistical process models, principally Markov decision processes. His recent research has focused on hierarchical probabilistic models, including hierarchical hidden Markov processes, semi-Markov decision processes, and hierarchical partially observable Markov decision processes. Professor Mahadevan has also developed state-of-the-art applications, including mobile robot navigation in indoor office environments, an active vision system for finding objects in cluttered rooms, and coordination among teams of factory agents optimizing production control.
Other
Manifold Alignment using Procrustes Analysis (with Chang Wang), Computer Science Department Faculty Publication Series (2008)
In this paper we introduce a novel approach to manifold alignment, based on Procrustes analysis....
Fast Direct Policy Evaluation using Multiscale Analysis of Markov Diffusion Processes (with Mauro Maggioni), Computer Science Department Faculty Publication Series (2006)
Policy evaluation is a critical step in the approximate solution of large Markov decision processes...
Proto-transfer Learning in Markov Decision Processes Using Spectral Methods (with Kimberly Ferguson), Computer Science Department Faculty Publication Series (2006)
In this paper we introduce proto-transfer leaning, a new framework for transfer learning. We explore...
Proto-Value Functions: Developmental Reinforcement Learning, Computer Science Department Faculty Publication Series (2005)
This paper presents a novel framework called proto-reinforcement learning (PRL), based on a mathematical model...
Learning to Communicate and Act using Hierarchical Reinforcement Learning (with Mohammad Ghavamzadeh), Computer Science Department Faculty Publication Series (2004)
In this paper, we address the issue of rational communication behavior among autonomous agents. The...