Forcement Learning Agents
نویسندگان
چکیده
We introduce two novel tactics for adversarial attack on deep reinforcement learning (RL) agents: strategically-timed and enchanting attack. For strategicallytimed attack, our method selectively forces the deep RL agent to take the least likely action. For enchanting attack, our method lures the agent to a target state by staging a sequence of adversarial attacks. We show that DQN and A3C agents are vulnerable to both tactics. Future work on defending is discussed in App. C.
منابع مشابه
Adaptive neuro-fuzzy modeling applied to policy gradient reinforcement learning
-Function approximation has been used extensively with rein forcement learning, even though theoretical support was based mainly on tabular representations. This paper proposes an actor-critic structure following the existing convergence proofs as much as possible. The actor and critic modules employ an adaptive neuro-fuzzy architecture based on fuzzy ARTMAP concepts and gradient descent. Resul...
متن کاملAn on-line algorithm for dynamic reinforcement learning and planning in reactive environments
An on line learning algorithm for reinforcement learning with continually running recur rent networks in non stationary reactive environments is described Various kinds of rein forcement are considered as special types of input to an agent living in the environment The agent s only goal is to maximize the amount of reinforcement received over time Supervised learning techniques for recurrent ne...
متن کاملSupplement to : Kim , Minjae , and Ezra W . Zuckerman Sivan . 2017 . “ Faking It Is Hard to Do : Entrepreneurial Norm En - forcement and Suspicions of Deviance ” Sociological
متن کامل
Hierarchical Functional Concepts for Knowledge Transfer among Reinforcement Learning Agents
This article introduces the notions of functional space and concept as a way of knowledge representation and abstraction for Reinforcement Learning agents. These definitions are used as a tool of knowledge transfer among agents. The agents are assumed to be heterogeneous; they have different state spaces but share a same dynamic, reward and action space. In other words, the agents are assumed t...
متن کاملA Sensory Control System for Adjusting Group Emotion Using Bayesian Networks and Reinforcement Learning
The relationship between sensory stimuli and emotion has been actively investigated, but it is relatively undisclosed to determine appropriate stimuli for inducing the target emotion for a group of people in the same space like school, hospital, store, etc. In this paper, we propose a stimuli control system to adjust group emotion in a closed space, especially kindergarten. The proposed system ...
متن کامل