Model-Based Offline Reinforcement Learning with Pessimism-Modulated Dynamics Belief.
Kaiyang GuoYunfeng ShaoYanhui GengPublished in: NeurIPS (2022)
Keyphrases
- reinforcement learning
- model free
- function approximation
- dynamical systems
- learning algorithm
- reinforcement learning algorithms
- real time
- markov decision processes
- belief revision
- optimal control
- belief space
- dynamic model
- optimal policy
- state space
- radial basis function
- temporal difference
- expected utility
- probability theory
- evolutionary algorithm
- initial conditions
- fully unsupervised
- temporal difference learning
- stochastic approximation
- learning environment
- continuous state
- multi agent reinforcement learning