Optimal Bayesian Reinforcement Learning on Trees

Add to your list(s) Download to your calendar using vCal

• Philipp Hennig (University of Cambridge)
• Monday 18 May 2009, 15:00-16:00
• TCM Seminar Room, Cavendish Laboratory, Department of Physics.

If you have a question about this talk, please contact Philipp Hennig.

The “Q-Learning” algorithm is the classical solution to the so-called “Optimal” Reinforcement Learning Problem. Q-Learning uses samples of future rewards generated by a non-optimal policy to derive point estimates of the future rewards from the (unknown) optimal policy.

In the first part of this talk, I will show that a Bayesian treatment, in forcing us to explicitly define our assumptions, reveals some interesting aspects of this problem that seem to have been overlooked so far.

In the second part, I will introduce an algorithm that uses Expectation Propagation to generate beliefs over possible future rewards from the optimal policy if the Markov Environment forms a tree (i.e. “Bayesian Q-Learning on trees”) and will show some preliminary results for its application to Game Trees.

This talk is part of the Machine Learning Journal Club series.

This talk is included in these lists:

• Cambridge talks
• Guy Emerson's list
• Hanchen DaDaDash
• Inference Group Journal Clubs
• Inference Group Summary
• Interested Talks
• ML
• Machine Learning Journal Club
• Machine Learning Summary
• Quantum Matter Journal Club
• TCM Seminar Room, Cavendish Laboratory, Department of Physics
• TQS Journal Clubs
• rp587
• yk373's list

Note that ex-directory lists are not shown.