Minimalistic Attacks: How Little it Takes to Fool a Deep Reinforcement Learning Policy.
Xinghua QuZhu SunYew-Soon OngPengfei WeiAbhishek GuptaPublished in: CoRR (2019)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- action selection
- markov decision process
- markov decision processes
- markov decision problems
- policy iteration
- reinforcement learning problems
- partially observable
- state and action spaces
- actor critic
- function approximators
- reward function
- policy evaluation
- action space
- state action
- function approximation
- approximate dynamic programming
- control policies
- control policy
- state space
- policy gradient
- model free
- partially observable environments
- reinforcement learning methods
- partially observable domains
- countermeasures
- temporal difference
- reinforcement learning algorithms
- watermarking scheme
- learning process
- dynamic programming
- supervised learning
- infinite horizon
- learning algorithm
- inverse reinforcement learning
- state dependent
- average reward
- control problems
- security mechanisms
- security threats
- continuous state spaces
- long run
- ddos attacks