Sridhar Mahadevan

 

Professor




School of  Computer Science


AAAI Fellow


Co-Director, Autonomous Learning Laboratory


Editorial Board, Journal of Machine Learning Research



Research Interests




Artificial Intelligence

Machine Learning

Reinforcement Learning

Representation Discovery

Variational Inequalities



Teaching




On sabbatical leave, Fall/Spring 2014-2015.


CONTACT INFO:




mahadeva AT cs.umass.edu

140 Governor’s Drive

Department of Computer Science

University of Massachusetts

Amherst MA 01003

(413)545-3140

Administrative Assistant: Susan Overstreet

 
 
 

My research spans across many areas of artificial intelligence (AI) and machine learning (ML). Most recently, my students and I are investigating a new framework "rethinking" AI and ML based on the concept of equilibration, which unifies a broad class of problems, including (convex) optimization, game theory, complementarity problems, networked equilibrium problems, and nonlinear equations. The approach uses the mathematical framework of variational inequalities.

We are exploring many applications of this framework, including economic models of the next-generation of Internet architectures, new safe, scalable, and reliable reinforcement learning algorithms for solving sequential decision problems, and new methods for learning low-dimensional representations of high-dimensional scientific datasets, such as spectroscopic measurements of rocks on Mars from Curiosity, the rover currently on Mars, and materials from near-Earth asteroids. As an illustration of the equilibration formulation, we have just completed a long paper on Arxiv developing a new approach to sequential decision making and reinforcement learning, which is listed below.


Sridhar Mahadevan, Bo Liu, Philip Thomas, Will Dabney, Steve Giguere, Nicholas Jacek, Ian Gemp and Ji Liu“, Proximal Reinforcement Learning: A New Theory of Sequential Decision Making in Primal-Dual Spaces , Arxiv, May 26, 2014 (126 pages)