MICo: Improved representations via sampling-based state similarity for Markov decision processes.
Pablo Samuel CastroTyler KastnerPrakash PanangadenMark RowlandPublished in: NeurIPS (2021)
Keyphrases
- markov decision processes
- state space
- action space
- optimal policy
- reinforcement learning
- transition matrices
- finite state
- policy iteration
- state abstraction
- dynamic programming
- real time dynamic programming
- markov decision process
- partially observable
- reachability analysis
- average cost
- decision processes
- planning under uncertainty
- decision theoretic planning
- model based reinforcement learning
- action sets
- infinite horizon
- state variables
- reinforcement learning algorithms
- reward function
- markov chain
- discounted reward
- interval estimation
- state and action spaces
- factored mdps
- risk sensitive
- multi agent