DSI Colloquium: Professor Michael Littman

Friday, September 21, 2018

4:00pm - 5:00pm

The Data Science Initiative

Watson Center for Information Technology (CIT)

241

MICHAEL LITTMAN
PROFESSOR OF COMPUTER SCIENCE, BROWN UNIVERSITY

ASSESSING AND IMPROVING GENERALIZATION IN DEEP REINFORCEMENT LEARNING

Deep reinforcement-learning approaches have been shown to produce a remarkable performance on a range of challenging control tasks. Observations of the resulting behavior give the impression that agents construct rich task representations that support insightful action decisions. Looking closely at the generalization capacity of these deep Q-networks, however, we find that the learned value computations often reduce to brittle memorization and that the network does not know how to handle even small non-adversarial modifications to the states it encounters during execution. We examine training methods that improve generalization capability. Our results provide strong evidence that not all deep networks learn robust behaviors, and that careful consideration must be made to achieve results to the contrary. 

 

Coffee and pastry to be served.