Single-Timescale Actor-Critic Provably Finds Globally Optimal Policy.
Zuyue FuZhuoran YangZhaoran WangPublished in: ICLR (2021)
Keyphrases
- optimal policy
- average reward
- actor critic
- reinforcement learning
- policy iteration
- markov decision processes
- long run
- finite horizon
- infinite horizon
- decision problems
- dynamic programming
- state space
- partially observable markov decision processes
- finite state
- multistage
- optimal control
- sufficient conditions
- reinforcement learning algorithms
- markov decision problems
- markov decision process
- approximate dynamic programming
- policy gradient
- gradient method
- markov chain
- decision making
- initial state
- average cost
- partially observable
- temporal difference
- data mining
- learning problems