Towards Feature Selection In Actor-Critic Algorithms

Abstract

Choosing features for the critic in actor-critic algorithms with function approximation is known to be a challenge. Too few critic features can lead to degeneracy of the actor gradient, and too many features may lead to slower convergence of the learner. In this paper, we show that a well-studied class of actor policies satisfy the known requirements for… (More)

Topics

1 Figure or Table

Cite this paper

@inproceedings{Rohanimanesh2007TowardsFS, title={Towards Feature Selection In Actor-Critic Algorithms}, author={Khashayar Rohanimanesh and Nicholas Roy and Russ Tedrake}, year={2007} }