Analysis of an Alternate Policy Gradient Estimator for Softmax Policies

Reinforcement learning, policy gradient, softmax policy estimator, policy saturation, non-stationary environments, incremental learning