Actor-critic is implicitly biased towards high entropy optimal policies.
Yuzheng HuZiwei JiMatus TelgarskyPublished in: CoRR (2021)
Keyphrases
- optimal policy
- average reward
- actor critic
- policy iteration
- reinforcement learning
- markov decision processes
- decision problems
- reinforcement learning algorithms
- state space
- multistage
- partially observable markov decision processes
- long run
- dynamic programming
- policy gradient
- finite state
- finite horizon
- infinite horizon
- average cost
- optimal control
- function approximation
- model free
- temporal difference
- sufficient conditions
- approximate dynamic programming
- fixed point
- linear programming
- initial state
- control system
- reinforcement learning methods
- learning algorithm