Attacking Deep Reinforcement Learning With Decoupled Adversarial Policy.
Kanghua MoWeixuan TangJin LiXu YuanPublished in: IEEE Trans. Dependable Secur. Comput. (2023)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- markov decision process
- action selection
- markov decision processes
- control policies
- reinforcement learning problems
- state and action spaces
- policy evaluation
- multi agent
- policy iteration
- function approximation
- partially observable
- state space
- control policy
- function approximators
- action space
- approximate dynamic programming
- markov decision problems
- reward function
- partially observable environments
- reinforcement learning algorithms
- actor critic
- dynamic programming
- learning algorithm
- policy gradient
- control problems
- rl algorithms
- partially observable markov decision processes
- temporal difference
- infinite horizon
- least squares
- state action
- sufficient conditions
- agent receives
- average reward
- partially observable domains
- continuous state
- decision problems
- long run
- approximate policy iteration
- model free reinforcement learning
- temporal difference learning
- deep learning
- model free
- finite state
- learning process
- learning classifier systems
- optimal control
- natural actor critic
- supervised learning