Adaptive Reward-Poisoning Attacks against Reinforcement Learning.
Xuezhou ZhangYuzhe MaAdish SinglaXiaojin ZhuPublished in: ICML (2020)
Keyphrases
- reinforcement learning
- adaptive control
- function approximation
- state space
- markov decision processes
- learning capabilities
- countermeasures
- temporal difference
- reinforcement learning algorithms
- multi agent
- learning algorithm
- reward function
- optimal policy
- machine learning
- partially observable
- learning agent
- eligibility traces
- partially observable environments
- model free
- mobile robot
- dynamic programming
- multi agent systems
- reinforcement learning methods
- actor critic
- malicious attacks
- total reward