Session 10D: Machine Unlearning
Authors, authors and presenters: Dayong Ye (University of Technology Sydney), Tianqing Zhu (City University of Macau), Congcong Zhu (City University of Macau), Derui Wang (CSIRO’s Data61), Kun Gao (Sydney University of Technology), Zewei Shi (CSIRO’s Data61), Sheng Shen (Torrens University, Australia), Wanlei Zhou (Macau City University), Minhui Xue (CSIRO’s Data61)
paper
Reinforcement unlearning
Machine unlearning refers to the process of reducing the impact of certain training data on a machine learning model based on a deletion request from the data owner. However, one important area that has been largely overlooked in non-learning research is reinforcement learning. Reinforcement learning focuses on training agents to make optimal decisions within an environment and maximize cumulative rewards. During training, agents tend to memorize features of the environment, which raises major privacy concerns. According to data protection regulations, the owner of the environment retains the right to revoke access to the agent’s training data, thus necessitating the development of a new and urgent research area called reinforcement non-learning. Reinforcement unlearning focuses on undoing the entire environment rather than individual data samples. This unique characteristic presents three distinct challenges: 1) How to propose a non-learning scheme for the environment. 2) How to avoid agent performance degradation in the rest of the environment. 3) How to evaluate the effectiveness of unlearning. To address these challenges, we propose two reinforcement non-learning techniques. The first method is based on progressive reinforcement learning, which aims to gradually erase previously acquired knowledge by the agent. The second method leverages environment poisoning attacks to force the agent to learn new knowledge, even if it is inaccurate, and remove the unlearned environment. In particular, to address the third challenge, we introduce the concept of “environmental inference” to evaluate non-learning outcomes.
About NDSS
The Network and Distributed Systems Security Symposium (NDSS) facilitates the exchange of information between network and distributed systems security researchers and practitioners. The intended audience includes anyone interested in the practical aspects of network and distributed systems security, with an emphasis on practical system design and implementation. A primary goal is to encourage and enable the Internet community to adapt, adopt, and advance available security technologies.
We would like to thank the Network and Distributed Systems Security (NDSS) Symposium creators, authors, and presenters for sharing great content from the NDSS Symposium 2025 conference on their organization’s YouTube channel.
Permalink
*** This is a syndicated blog from the Security Bloggers Network brought to you by Infosecurity.US and written by Marc Handelman. Read the original post: https://www.youtube-nocookie.com/embed/K0RqQoPOA40?si=YDTMEoN4iO9VyywG
Source link
