Optimistic Agents are Asymptotically Optimal

2012/09/29 by Peter Sunehag, Marcus Hutter · 1 voice
Computer Science · #cs.AI #cs.LG

paper · pdf

published as Proc. 25th Australasian Joint Conference on Artificial Intelligence (AusAI 2012) 15-26 · 13 LaTeX pages

Abstract

We use optimism to introduce generic asymptotically optimal reinforcement learning agents. They achieve, with an arbitrary finite or compact class of environments, asymptotically optimal behavior. Furthermore, in the finite deterministic case we provide finite error bounds.

Discussions

• Optimistic Agents Are Asymptotically Optimal [hn, 1 points, 0 comments]

Related

• Towards Tractable Optimism in Model-Based Reinforcement Learning
• Curiosity Killed or Incapacitated the Cat and the Asymptotically Optimal Agent
• Minimax Optimal Reinforcement Learning with Quasi-Optimism
• The End of Optimism? An Asymptotic Analysis of Finite-Armed Linear Bandits
• Simple Agent, Complex Environment: Efficient Reinforcement Learning with Agent States