Defending Observation Attacks in Deep Reinforcement Learning via Detection and Denoising.
Zikang XiongJoe EappenHe ZhuSuresh JagannathanPublished in: ECML/PKDD (3) (2022)
Keyphrases
- denoising
- reinforcement learning
- image denoising
- ddos attacks
- noisy images
- attack detection
- detection accuracy
- function approximation
- total variation
- automatic detection
- false positives
- detection method
- detection algorithm
- image processing
- state space
- object detection
- false alarms
- anomaly detection
- natural images
- wavelet domain
- multi agent
- watermarking scheme
- multiresolution
- malware detection
- temporal difference
- model free
- optimal control
- learning process
- detection rate
- markov decision processes
- optimal policy