General RL Pages: Myths of RL, Successes of RL, Algorithms of RL and Demos of RL

In the field of reinforcement learning, we have made tremendous progress by adopting the formalisms of Markov Decision Processes (MDPs) and Partially Observable Markov Decision Processes (POMDPs) from operations research and optimal control. I detail some of my contributions to this progress below.

Currently, I am focused on rethinking all of the three basic aspects of RL: state, action, and reward.

1. Rethinking state. This effort has led to the projects on Predictive State Representations.
2. Rethinking action. My own effort on this started with my early work on temporally abstract actions in RL that led to later work on options.
3. Rethinking reward. This effort has led to the projects on Intrinsically Motivated RL and on Mechanism Design and RL.

Some of my past contributions include papers on:

1. Exploration versus exploitation.
2. Dealing with hidden state.
3. Theory of RL.
4. New learning algorithms.
5. Applications.