Actor Critic with Differentially Private Critic

Jonathan Lebensold, William Hamilton, Borja Balle +1

Reinforcement learning algorithms are known to be sample inefficient, and often performance on one task can be substantially improved by leveraging information (e.g., via pre-train…