Policy Gradient with Second Order Momentum

Tianyu Sun

We develop Policy Gradient with Second-Order Momentum (PG-SOM), a lightweight second-order optimisation scheme for reinforcement-learning policies. PG-SOM augments the classical RE…