Reward is enough for convex MDPs.
Tom ZahavyBrendan O'DonoghueGuillaume DesjardinsSatinder SinghPublished in: NeurIPS (2021)
Keyphrases
- reinforcement learning
- markov decision processes
- reward function
- average reward
- convex optimization
- optimal policy
- state space
- discounted reward
- total reward
- expected reward
- factored mdps
- long run
- finite horizon
- piecewise linear
- semi markov decision processes
- convex relaxation
- function approximation
- model free
- dynamic programming
- reinforcement learning algorithms
- planning under uncertainty
- inverse reinforcement learning
- policy iteration
- markov decision process
- initial state
- machine learning
- finite state
- state and action spaces
- model based reinforcement learning
- average cost
- decision theoretic planning
- policy gradient
- probabilistic planning
- infinite horizon
- transition probabilities
- markov chain
- learning algorithm