03/05/2021

Efficient Wasserstein Natural Gradients for Reinforcement Learning

Ted Moskovitz, Michael Arbel, Ferenc Huszar, Arthur Gretton

Keywords: reinforcement learning, optimization

Abstract Paper Similar Papers

Abstract: A novel optimization approach is proposed for application to policy gradient methods and evolution strategies for reinforcement learning (RL). The procedure uses a computationally efficient \emph{Wasserstein natural gradient} (WNG) descent that takes advantage of the geometry induced by a Wasserstein penalty to speed optimization. This method follows the recent theme in RL of including divergence penalties in the objective to establish trust regions. Experiments on challenging tasks demonstrate improvements in both computational cost and performance over advanced baselines.

 0
 0
 0
 0
  Share    
This is an embedded video. Talk and the respective paper are published at ICLR 2021 virtual conference. If you are one of the authors of the paper and want to manage your upload, see the question "My papertalk has been externally embedded..." in the FAQ section.

Comments

Post Comment
no comments yet
code of conduct: tbd Characters remaining: 140

Similar Papers

 0
 0
 0
 0
 4:56 
 0
 0
 0
 0
 4:49 
 0
 0
 0
 0
 3:16 
 0
 0
 0
 0
 11:43 
 0
 0
 0
 0
 13:37 
 0
 0
 0
 0
 5:05 
 0
 0
 0
 0
 2:55 
 0
 0
 0
 0
 15:07 
 0
 0
 0
 0
 2:44 
 0
 0
 0
 0
 5:02 
 0
 0
 0
 0
 16:03 
 0
 0
 0
 0
 13:12 
 0
 0
 0
 0
 2:57 
 0
 0
 0
 0
 2:59 
 0
 0
 0
 0
 14:10 
 0
 0
 0
 0
 14:31 
 0
 0
 0
 0
 18:23 
 0
 0
 0
 0
 14:10 
 0
 0
 0
 0
 15:31 
 0
 0
 0
 0
 3:21 
 0
 0
 0
 0
 3:13 
 0
 0
 0
 0
 13:45 
 0
 0
 0
 0
 12:27 
 0
 0
 0
 0
 5:16 
 0
 0
 0
 0
 13:38 
 0
 0
 0
 0
 0:50 
 0
 0
 0
 0
 19:46 
 0
 0
 0
 0
 4:47 
 0
 0
 0
 0
 8:30 
 0
 0
 0
 0
 3:15 
 0
 0
 0
 0
 15:10 
 0
 0
 0
 0
 13:21 
 0
 0
 0
 0
 12:59 
 0
 0
 0
 0
 3:20 
 0
 0
 0
 0
 14:17 
 0
 0
 0
 0
 5:25 
 0
 0
 0
 0
 13:44 
 0
 0
 0
 0
 3:21 
 0
 0
 0
 0
 3:25 
 0
 0
 0
 0
 11:42