Principled and Scalable Exploration for Reinforcement Learning

Supervisor: Dr Paulo Rauber

Research group(s): Game AI

Reinforcement learning is the study of agents that interact with an environment by observing states and choosing actions that ideally maximize expected cumulative reward signals. If intelligence measures the ability of an agent to achieve goals in a wide range of environments, then reinforcement learning is an excellent framework to study some fundamental challenges that any intelligent agent will face. Despite their recent successes in challenging applications (such as playing Chess or StarCraft II), reinforcement learning methods still require an impractical amount of trial and error in order to produce an agent that solves a task. The goal of this project is to develop principled and scalable exploration methods to address this issue.