Differentially Private Policy Gradient

Alexandre Rio, Merwan Barlier, Igor Colin

Motivated by the increasing deployment of reinforcement learning in the real world, involving a large consumption of personal data, we introduce a differentially private (DP) polic…