Predictable MDP Abstraction for Unsupervised Model-Based RL.
Seohong ParkSergey LevinePublished in: CoRR (2023)
Keyphrases
- markov decision processes
- reinforcement learning
- decision theoretic planning
- markov decision process
- optimal policy
- model free
- state space
- supervised learning
- unsupervised learning
- policy iteration
- fully unsupervised
- data driven
- state abstraction
- action space
- finite state
- state and action spaces
- average cost
- reinforcement learning algorithms
- high level
- function approximation
- factored mdps
- utility function
- semi supervised
- high level abstractions
- reward function
- linear programming
- bayesian reinforcement learning
- action sets
- markov decision problems
- heuristic search
- decision problems
- learning agent
- optimal control
- average reward
- planning under uncertainty
- partially observable markov decision processes
- dynamic programming
- multi agent
- linear program
- machine learning
- partially observable
- infinite horizon