Policy Smoothing for Provably Robust Reinforcement Learning.
Aounon KumarAlexander LevineSoheil FeiziPublished in: CoRR (2021)
Keyphrases
- reinforcement learning
- optimal policy
- markov decision processes
- policy search
- robust statistical
- action selection
- robust statistics
- action space
- partially observable
- function approximation
- state space
- dynamic programming
- worst case
- model free
- reinforcement learning algorithms
- reinforcement learning problems
- markov decision process
- continuous state spaces
- neural network
- partially observable environments
- markov decision problems
- theoretical guarantees
- learning classifier systems
- computationally efficient
- learning process
- search algorithm
- learning algorithm