A Model-free Learning Algorithm for Infinite-horizon Average-reward MDPs with Near-optimal Regret.
Mehdi Jafarnia-JahromiChen-Yu WeiRahul JainHaipeng LuoPublished in: CoRR (2020)
Keyphrases
- average reward
- model free
- total reward
- reinforcement learning
- infinite horizon
- reinforcement learning algorithms
- learning algorithm
- markov decision processes
- optimal policy
- policy iteration
- rl algorithms
- finite horizon
- semi markov decision processes
- reward function
- policy evaluation
- discount factor
- long run
- function approximation
- state space
- optimal control
- dynamic programming
- stochastic games
- temporal difference
- discounted reward
- state action
- machine learning algorithms
- partially observable
- finite state
- supervised learning
- average cost
- learning problems
- multi agent
- dec pomdps
- policy gradient
- machine learning
- markov decision problems
- action space
- partially observable markov decision processes
- decision problems
- planning under uncertainty
- data mining