Reinforcement learning seeks for a policy that receives the highest total reward from its environement. Functional representation is a powerful tool to approximate complex functions, which can help learn complex policies for fitting practical environments.
Papers:
- Boosting nonparametric policies: In the AAMAS'16 (PDF) paper, we propose PolicyBoost to learn functional policies with less overfitting.
- Improving speed by napping: By functional representation, a policy can be a sum of many basis functions, which causes a high time cost especially for reinforcement learning. In the AAMAS'14 (PDF) paper, we tackle this time cost barrier by the napping mechanism, which results a significant improvement in time as well as a potential improvement in total reward.
Codes: