Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Reinforcement Unlearning (2312.15910v5)

Published 26 Dec 2023 in cs.CR and cs.LG

Abstract: Machine unlearning refers to the process of mitigating the influence of specific training data on machine learning models based on removal requests from data owners. However, one important area that has been largely overlooked in the research of unlearning is reinforcement learning. Reinforcement learning focuses on training an agent to make optimal decisions within an environment to maximize its cumulative rewards. During the training, the agent tends to memorize the features of the environment, which raises a significant concern about privacy. As per data protection regulations, the owner of the environment holds the right to revoke access to the agent's training data, thus necessitating the development of a novel and pressing research field, known as \emph{reinforcement unlearning}. Reinforcement unlearning focuses on revoking entire environments rather than individual data samples. This unique characteristic presents three distinct challenges: 1) how to propose unlearning schemes for environments; 2) how to avoid degrading the agent's performance in remaining environments; and 3) how to evaluate the effectiveness of unlearning. To tackle these challenges, we propose two reinforcement unlearning methods. The first method is based on decremental reinforcement learning, which aims to erase the agent's previously acquired knowledge gradually. The second method leverages environment poisoning attacks, which encourage the agent to learn new, albeit incorrect, knowledge to remove the unlearning environment. Particularly, to tackle the third challenge, we introduce the concept of ``environment inference attack'' to evaluate the unlearning outcomes.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (42)
  1. Artari. 2023. https://paperswithcode.com/task/atari-games.
  2. Machine Unlearning. In Proc. of IEEE S & P. 141–159.
  3. Y. Cao and J. Yang. 2015. Towards Making Systems Forget with Machine Unlearning. In Proc. of IEEE S & P. 463–480.
  4. C. Dwork. 2006. Differential Privacy. In Proc. of International Conference on Automata, Languages and Programming. 1–12.
  5. Experts in a Markov Decision Process. In Proc. of NIPS.
  6. A Theoretical Analysis of Deep Q-Learning. https://arxiv.org/abs/1901.00137 (2020).
  7. Spectral-DP: Differentially Private Deep Learning through Spectral Perturbation and Filtering. In Proc. of IEEE S & P. 1944–1960.
  8. Local Differential Privacy for Regret Minimization in Reinforcement Learning. In Proc. of NIPS.
  9. GDPR. 2016. General Data Protection Regulation. https://gdpr-info.eu (2016).
  10. Making AI Forget You: Data Deletion in Machine Learning. In Proc. of NIPS.
  11. Certified Data Removal from Machine Learning Models. In Proc. of ICML.
  12. Adaptive Machine Unlearning. In Proc. of NIPS.
  13. Gym. 2023. https://gymnasium.farama.org.
  14. MazeExplorer: A Customisable 3D Benchmark for Assessing Generalisation in Reinforcement Learning. In Proc. of IEEE Conference on Games (CoG).
  15. Pinning Down Abuse on Google Maps. In Proc. of WWW. 1471–1479.
  16. Overcoming catastrophic forgetting in neural networks. PNAS 114, 13 (2017), 3521–3526.
  17. Towards Unbounded Machine Unlearning. In Proc. of NeurIPS.
  18. Deep Reinforcement Learning in Continuous Action Spaces: a Case Study in the Game of Simulated Curling. In Proc. of ICML. 2937–2946.
  19. Continuous control with deep reinforcement learning. In Proc. of ICLR.
  20. Learning to Navigate in Cities Without a Map. In Proc. of NeurIPS.
  21. Human-level control through deep reinforcement learning. Nature 518 (2015), 529–533.
  22. Playing Atari with Deep Reinforcement Learning. In Proc. of NIPS Deep Learning Workshop.
  23. How You Act Tells a Lot: Privacy-Leaking Attack on Deep Reinforcement Learning. In Proc. of AAMAS.
  24. Policy Teaching in Reinforcement Learning via Environment Poisoning Attacks. Journal of Machine Learning Research 22 (2021), 1–45.
  25. Private, Efficient, and Accurate: Protecting Models Trained by Multi-party Learning with Differential Privacy. In Proc. of IEEE S & P. 1926–1943.
  26. SoK: Let the Privacy Games Begin! A Unified Treatment of Data Inference Privacy in Machine Learning. In Proc. of IEEE S & P. 327–345.
  27. Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima. In Proc. of NIPS.
  28. Membership Inference Attacks Against Machine Learning Models. In Proc. of IEEE S & P. 3–18.
  29. On the Necessity of Auditable Algorithmic Definitions for Machine Unlearning. In Proc. of USENIX Security.
  30. Boosted Fitted Q-Iteration. In Proc. of ICML.
  31. Private Reinforcement Learning with PAC and Regret Guarantees. In Proc. of ICML.
  32. VirtualHome. 2023. http://virtual-home.org.
  33. Non-Transferable Learning: A New Approach for Model Ownership Verification and Applicability Authorization. In Proc. of ICLR.
  34. Deep Reinforcement Learning: A Survey. IEEE Transactions on Neural Networks and Learning Systems (2022), DOI: 10.1109/TNNLS.2022.3207346.
  35. Z. T. Wang and M. Ueda. 2022. Convergent and Efficient Deep Q Network Algorithm. In Proc. of ICLR.
  36. Machine Unlearning of Features and Labels. In Proc. of NDSS.
  37. Adversarial Policy Training against Deep Reinforcement Learning. In Proc. of USENIX Security. 1883–1900.
  38. Spiking Pitch Black: Poisoning an Unknown Environment to Attack Unknown Reinforcement Learning. In Proc. of AAMAS. 1409–1417.
  39. Machine Unlearning: A Survey. Comput. Surveys (2023). https://doi.org/10.1145/3603620
  40. Enhanced Membership Inference Attacks against Machine Learning Models. In Proc. of CCS.
  41. AIRS: Explanation for Deep Reinforcement Learning based Security Applications. In Proc. of USENIX Security.
  42. Prompt Certified Machine Unlearning with Randomized Gradient Smoothing and Quantization. In Proc. of NIPS.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (9)
  1. Dayong Ye (18 papers)
  2. Tianqing Zhu (85 papers)
  3. Congcong Zhu (7 papers)
  4. Derui Wang (25 papers)
  5. Sheng Shen (68 papers)
  6. Wanlei Zhou (63 papers)
  7. Minhui Xue (72 papers)
  8. Zewei Shi (4 papers)
  9. Kun Gao (32 papers)
Citations (5)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now