Spring 2021

Inferring Specifications From Demonstrations; A Maximum (Causal) Entropy Approach

Monday, Mar. 29, 2021 10:00 am10:30 am

Add to Calendar



In many settings, episodic demonstrations provide a natural and robust mechanism to partially specify a task, even in the presence of unlabeled demonstration errors. This problem, inferring intent from demonstrations, has received a fair amount of attention over the past two decades particularly within the robotics and AI communities; but until recently has remained unexplored for learning concept classes such as temporal logic and automata. In this talk, I will review a promising thread of research that adapts maximum (causal) entropy inverse reinforcement learning to estimate the posteriori probability of a bounded trace property given a multi-set of demonstrations.

PDF icon Slides5.45 MB