Frame-Correlation Transfers Trigger Economical Attacks on Deep Reinforcement Learning Policies.
Xinghua QuYew-Soon OngAbhishek GuptaPublished in: IEEE Trans. Cybern. (2022)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- markov decision process
- state space
- countermeasures
- function approximation
- control policies
- markov decision processes
- fitted q iteration
- reward function
- partially observable markov decision processes
- dynamic programming
- hierarchical reinforcement learning
- reinforcement learning agents
- learning algorithm
- malicious attacks
- reference frame
- transfer learning
- multiagent reinforcement learning
- chosen plaintext
- markov decision problems
- machine learning
- control policy
- reinforcement learning algorithms
- frame rate
- correlation coefficient
- decision problems
- security threats
- infinite horizon
- continuous state
- optimal control
- video frames
- policy gradient methods
- learning process
- multi agent