Multi-Environment Training Against Reward Poisoning Attacks on Deep Reinforcement Learning.
Myria BouhaddiKamel AdiPublished in: SECRYPT (2023)
Keyphrases
- reinforcement learning
- function approximation
- supervised learning
- learning agent
- robocup soccer
- real time
- model free
- virtual training
- state space
- training examples
- partially observable environments
- exploration strategy
- eligibility traces
- initially unknown
- reinforcement learning algorithms
- agent learns
- agent receives
- e learning
- optimal control
- markov decision processes
- dynamic environments
- reward function
- action selection
- long run
- multi agent environments
- training samples
- dynamic programming
- learning process
- training set
- multi agent