Speeding up Reinforcement Learning

1^st July 2015, 12:00 Ashton Lecture Theatre
Prof Ann Nowe
Artificial Intelligence Lab
Vrije Universiteit Brussel
Belgium

Abstract

Reinforcement learning describes how a learning agent can achieve optimal behaviour based on interactions with its environment and reward feedback. A limiting factor in reinforcement learning as employed in artificial intelligence is the need for an often prohibitively large number of environment samples before the agent reaches a desirable level of performance.In this talk, I will focus on two approaches that allow to speed up learning, being reward shaping and demonstrations. I will show how these two approaches can be combined allowing to leverage human input without making the assumption regarding demonstration optimality. We show experimentally that our approach requires significantly fewer demonstrations, is more robust against sub-optimality of demonstrations, and achieves much faster learning than the recently developed HAT algorithm.