Pseudorehearsal in value function approximation

Vladimir Marochko, Leonard Johard, Manuel Mazzara

Catastrophic forgetting is of special importance in reinforcement learning, as the data distribution is generally non-stationary over time. We study and compare several pseudorehea…