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