Efficient Wasserstein Natural Gradients for Reinforcement Learning
This work addresses optimization efficiency for reinforcement learning practitioners, but it is incremental as it builds on existing divergence penalty methods.
The paper tackles the problem of slow optimization in reinforcement learning by proposing a computationally efficient Wasserstein natural gradient descent method, which improves computational cost and performance on challenging tasks compared to advanced baselines.
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 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 a divergence penalty in the objective to establish a trust region. Experiments on challenging tasks demonstrate improvements in both computational cost and performance over advanced baselines.