Certifying Safety in Reinforcement Learning under Adversarial Perturbation Attacks.
Junlin WuHussein SibaiYevgeniy VorobeychikPublished in: SP (Workshops) (2024)
Keyphrases
- reinforcement learning
- multi agent
- function approximation
- countermeasures
- model free
- state space
- traffic analysis
- machine learning
- watermarking scheme
- malicious attacks
- supervised learning
- reinforcement learning algorithms
- dynamic programming
- learning algorithm
- temporal difference
- chosen plaintext
- terrorist attacks
- action selection
- learning process
- optimal policy
- optimal control
- traffic accidents
- data mining
- denial of service attacks
- evaluation function
- learning problems
- attack detection
- robotic control
- detecting malicious
- markov decision processes