Vulnerability of Deep Reinforcement Learning to Policy Induction Attacks.
Vahid BehzadanArslan MunirPublished in: MLDM (2017)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- security risks
- action selection
- markov decision process
- buffer overflow
- partially observable environments
- state space
- policy gradient
- reinforcement learning problems
- denial of service
- partially observable
- reinforcement learning algorithms
- function approximation
- approximate dynamic programming
- markov decision processes
- markov decision problems
- control policy
- action space
- actor critic
- control policies
- policy evaluation
- security vulnerabilities
- countermeasures
- state and action spaces
- dos attacks
- function approximators
- state action
- partially observable markov decision processes
- reward function
- dynamic programming
- machine learning
- average reward
- infinite horizon
- multi agent
- policy iteration
- rl algorithms
- agent learns
- continuous state spaces
- learning algorithm
- transition model
- long run
- inductive learning
- security threats
- attack detection
- policy gradient methods
- denial of service attacks
- continuous state
- state dependent
- temporal difference
- watermarking scheme
- learning problems
- decision trees
- agent receives