ABSTRACT
The past years have witnessed the rapid development of deep reinforcement learning (DRL), which is a combination of deep learning and reinforcement learning (RL). However, the adoption of deep neural networks makes the decision-making process of DRL opaque and lacking transparency. Motivated by this, various interpretation methods for DRL have been proposed. However, those interpretation methods make an implicit assumption that they are performed in a reliable and secure environment. In practice, sequential agent-environment interactions expose the DRL algorithms and their corresponding downstream interpretations to extra adversarial risk. In spite of the prevalence of malicious attacks, there is no existing work studying the possibility and feasibility of malicious attacks against DRL interpretations. To bridge this gap, in this paper, we investigate the vulnerability of DRL interpretation methods. Specifically, we introduce the first study of the adversarial attacks against DRL interpretations, and propose an optimization framework based on which the optimal adversarial attack strategy can be derived. In addition, we study the vulnerability of DRL interpretation methods to the model poisoning attacks, and present an algorithmic framework to rigorously formulate the proposed model poisoning attack. Finally, we conduct both theoretical analysis and extensive experiments to validate the effectiveness of the proposed malicious attacks against DRL interpretations.
- Julius Adebayo, Justin Gilmer, Michael Muelly, Ian Goodfellow, Moritz Hardt, and Been Kim. 2018. Sanity checks for saliency maps. In NeurIPS. 9505--9515.Google Scholar
- Akanksha Atrey, Kaleigh Clary, and David Jensen. 2019. Exploratory Not Explanatory: Counterfactual Analysis of Saliency Maps for Deep Reinforcement Learning. arXiv preprint arXiv:1912.05743 (2019).Google Scholar
- Vahid Behzadan and Arslan Munir. 2017. Vulnerability of deep reinforcement learning to policy induction attacks. In International Conference on Machine Learning and Data Mining in Pattern Recognition. Springer, 262--275.Google ScholarCross Ref
- Amirata Ghorbani, Abubakar Abid, and James Zou. 2019. Interpretation of neural networks is fragile. In Proceedings of the AAAI Conference on Artificial Intelligence.Google ScholarDigital Library
- Sam Greydanus, Anurag Koul, Jonathan Dodge, and Alan Fern. 2017. Visualizing and understanding atari agents. arXiv preprint arXiv:1711.00138 (2017).Google Scholar
- Mengdi Huai, Di Wang, Chenglin Miao, and Aidong Zhang. 2020. Towards Interpretation of Pairwise Learning. In Thirty-fourth AAAI Conference on Artificial Intelligence.Google Scholar
- Sandy Huang, Nicolas Papernot, Ian Goodfellow, Yan Duan, and Pieter Abbeel. 2017. Adversarial attacks on neural network policies. arXiv preprint arXiv:1702.02284 (2017).Google Scholar
- Léonard Hussenot, Matthieu Geist, and Olivier Pietquin. 2019. Targeted Attacks on Deep Reinforcement Learning Agents through Adversarial Observations. arXiv preprint arXiv:1905.12282 (2019).Google Scholar
- Rahul Iyer, Yuezhang Li, Huao Li, Michael Lewis, Ramitha Sundar, and Katia Sycara. 2018. Transparency and explanation in deep reinforcement learning neural networks. In Proc. of the AAAI/ACM Conference on AI, Ethics, and Society.Google ScholarDigital Library
- Michael Kearns and Satinder Singh. 2002. Near-Optimal Reinforcement Learning in Polynomial Time. Mach. Learn. (2002).Google Scholar
- Pieter-Jan Kindermans, Sara Hooker, Julius Adebayo, Maximilian Alber, Kristof T Schütt, Sven Dahne, Dumitru Erhan, and Been Kim. 2019. The (un) reliability of saliency methods. In Explainable AI: Interpreting, Explaining and Visualizing Deep Learning. Springer, 267--280.Google Scholar
- Yen-Chen Lin, Zhang-Wei Hong, Yuan-Hong Liao, Meng-Li Shih, Ming-Yu Liu, and Min Sun. 2017. Tactics of adversarial attack on deep reinforcement learning agents. arXiv preprint arXiv:1703.06748 (2017).Google ScholarDigital Library
- Chenglin Miao, Qi Li, Lu Su, Mengdi Huai, Wenjun Jiang, and Jing Gao. 2018a. Attack under Disguise: An Intelligent Data Poisoning Attack Mechanism in Crowdsourcing. In Proc. of the 2018 World Wide Web Conference. 13--22.Google ScholarDigital Library
- Chenglin Miao, Qi Li, Houping Xiao, Wenjun Jiang, Mengdi Huai, and Lu Su. 2018b. Towards data poisoning attacks in crowd sensing systems. In Proc. of the Eighteenth ACM International Symposium on Mobile Ad Hoc Networking and Computing. 111--120.Google ScholarDigital Library
- Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A Rusu, Joel Veness, Marc G Bellemare, Alex Graves, Martin Riedmiller, Andreas K Fidjeland, Georg Ostrovski, et al. 2015. Human-level control through deep reinforcement learning. Nature, Vol. 518, 7540 (2015), 529.Google Scholar
- Anay Pattanaik, Zhenyi Tang, Shuijing Liu, Gautham Bommannan, and Girish Chowdhary. 2018. Robust deep reinforcement learning with adversarial attacks. In Proc. of the 17th International Conference on Autonomous Agents and MultiAgent Systems. 2040--2042.Google Scholar
- Xinghua Qu, Zhu Sun, Pengfei Wei, Yew-Soon Ong, and Abhishek Gupta. 2019. Minimalistic Attacks: How Little it Takes to Fool a Deep Reinforcement Learning Policy. arXiv preprint arXiv:1911.03849 (2019).Google Scholar
- John Schulman, Sergey Levine, Pieter Abbeel, Michael Jordan, and Philipp Moritz. 2015. Trust region policy optimization. In ICML. 1889--1897.Google Scholar
- Jianwen Sun, Tianwei Zhang, Xiaofei Xie, Lei Ma, Yan Zheng, Kangjie Chen, and Yang Liu. 2020. Stealthy and efficient adversarial attacks against deep reinforcement learning. arXiv preprint arXiv:2005.07099 (2020).Google Scholar
- Richard S Sutton and Andrew G Barto. 2018. Reinforcement learning: An introduction. MIT press.Google ScholarDigital Library
- Ziyu Wang, Tom Schaul, Matteo Hessel, Hado Van Hasselt, Marc Lanctot, and Nando De Freitas. 2015. Dueling network architectures for deep reinforcement learning. arXiv preprint arXiv:1511.06581 (2015).Google Scholar
- Laurens Weitkamp, Elise van der Pol, and Zeynep Akata. 2018. Visual rationalizations in deep reinforcement learning for atari games. In Benelux Conference on Artificial Intelligence. Springer, 151--165.Google Scholar
- Kaidi Xu, Sijia Liu, Pu Zhao, Pin-Yu Chen, Huan Zhang, Quanfu Fan, Deniz Erdogmus, Yanzhi Wang, and Xue Lin. 2018. Structured adversarial attack: Towards general implementation and better interpretability. arXiv preprint arXiv:1808.01664 (2018).Google Scholar
- Liu Yuezhang, Ruohan Zhang, and Dana H Ballard. 2018. An Initial Attempt of Combining Visual Selective Attention with Deep Reinforcement Learning. arXiv preprint arXiv:1811.04407 (2018).Google Scholar
- Tom Zahavy, Nir Ben-Zrihem, and Shie Mannor. 2016. Graying the black box: Understanding dqns. In ICML. 1899--1908.Google Scholar
- Xinyang Zhang, Ningfei Wang, Hua Shen, Shouling Ji, Xiapu Luo, and Ting Wang. 2020. Interpretable deep learning under fire. In 29th USENIX Security Symposium (USENIX Security 20).Google Scholar
Index Terms
- Malicious Attacks against Deep Reinforcement Learning Interpretations
Recommendations
Query-based targeted action-space adversarial policies on deep reinforcement learning agents
ICCPS '21: Proceedings of the ACM/IEEE 12th International Conference on Cyber-Physical SystemsAdvances in computing resources have resulted in the increasing complexity of cyber-physical systems (CPS). As the complexity of CPS evolved, the focus has shifted to deep reinforcement learning-based (DRL) methods for control of these systems. This is ...
Reward poisoning attacks in deep reinforcement learning based on exploration strategies
AbstractIn recent years, deep reinforcement learning (DRL) has been widely applied in various fields, and its vulnerability to adversarial samples, which is inherited from deep learning, has also received considerable attention for training robust DRL ...
Highlights- Exploration strategies are applied to attacks against deep reinforcement learning.
- A reward poisoning attack algorithm based on E-values (RPAAE) method is proposed.
- The experimental results prove the effectiveness of the RPAAE ...
Poisoning attacks against knowledge graph-based recommendation systems using deep reinforcement learning
AbstractIn recent years, studies have revealed that introducing knowledge graphs (KGs) into recommendation systems as auxiliary information can improve recommendation accuracy. However, KGs are usually based on third-party data that may be manipulated by ...
Comments