Recent work has shown that diffusion models memorize and reproduce training data examples. At the same time, large copyright lawsuits and legislation such as GDPR have highlighted the need for erasing datapoints from diffusion models. However, retraining from scratch is often too expensive. This motivates the setting of data unlearning, i.e., the study of efficient techniques for unlearning specific datapoints from the training set. Existing concept unlearning techniques require an anchor prompt/class/distribution to guide unlearning, which is not available in the data unlearning setting. General-purpose machine unlearning techniques were found to be either unstable or failed to unlearn data. We therefore propose a family of new loss functions called Subtracted Importance Sampled Scores (SISS) that utilize importance sampling and are the first method to unlearn data with theoretical guarantees. SISS is constructed as a weighted combination between simpler objectives that are responsible for preserving model quality and unlearning the targeted datapoints. When evaluated on CelebA-HQ and MNIST, SISS achieved Pareto optimality along the quality and unlearning strength dimensions. On Stable Diffusion, SISS successfully mitigated memorization on nearly 90% of the prompts we tested.
Towards Irreversible Machine Unlearning for Diffusion Models
Department of Electrical and Computer Engineering, College of Design and Engineering, National University of Singapore（新加坡国立大学电子与计算机工程系，设计与工程学院）；Betterdata, Singapore（新加坡Betterdata公司）；Department of Computer Science, University of Sheffield（谢菲尔德大学计算机科学系）
Roots Beneath the Cut: Uncovering the Risk of Concept Revival in Pruning-Based Unlearning for Diffusion Models
University of Georgia（佐治亚大学）；Carnegie Mellon University（卡内基梅隆大学）；Northeastern University（东北大学）；Stevens Institute of Technology（史蒂文斯理工学院）；University of Arizona（亚利桑那大学）
Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient
Southeast University（东南大学）；Tsinghua University（清华大学）；The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））；University of California, Berkeley（加州大学伯克利分校）；Nanyang Technological University（南洋理工大学）
GUDA: Counterfactual Group-wise Training Data Attribution for Diffusion Models via Unlearning
University of Tokyo（东京大学）；Toyota Central Research Laboratory（丰田中央研究所）；University of California, Berkeley（加州大学伯克利分校）；Massachusetts Institute of Technology（麻省理工学院）；National Institute of Advanced Industrial Science and Technology（国家工业科学与技术研究院）
Reference-Specific Unlearning Metrics Can Hide the Truth: A Reality Check
University of Wisconsin-Madison（威斯康星大学麦迪逊分校）；LG AI Research（LG人工智能研究）；Seoul National University of Science and Technology（首尔科学技术大学）；New York University（纽约大学）；Genentech（基因泰克）
Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuning
Nanyang Technological University, Singapore（南洋理工大学，新加坡）；Central South University, China（中南大学，中国）；Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, China（航空航天信息安全部门，教育部，武汉大学，中国）；State Key Laboratory of Blockchain and Data Security, Zhejiang University, China（区块链与数据安全国家重点实验室，浙江大学，中国）
Image Can Bring Your Memory Back: A Novel Multi-Modal Guided Attack against Image Generation Model Unlearning
Institute of Data Science, National University of Singapore（数据科学研究所，新加坡国立大学）；S-Lab, Nanyang Technological University（南洋理工大学S实验室）；College of Computing and Data Science, Nanyang Technological University（计算与数据科学学院，南洋理工大学）
National University of Singapore（新加坡国立大学）；MIT（麻省理工学院）
This paper studies the problem of approximately unlearning a Bayesian model from a small subset of the training data to be erased. We frame this problem as one of minimizing the Kullback-Leibler divergence between the approximate posterior belief of model parameters after directly unlearning from erased data vs. the exact posterior belief from retraining with remaining data. Using the variational inference (VI) framework, we show that it is equivalent to minimizing an evidence upper bound which trades off between fully unlearning from erased data vs. not entirely forgetting the posterior belie…
The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））；Zhejiang University（浙江大学）；University of Sydney（悉尼大学）；Northwestern University（西北大学）；Analogy AI, Inc.（Analogy AI公司）；Mohamed Bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）
Pohang University of Science and Technology（浦项科技大学）
We consider a practical scenario of machine unlearning to erase a target dataset, which causes unexpected behavior from the trained model. The target dataset is often assumed to be fully identifiable in a standard unlearning scenario. Such a flawless identification, however, is almost impossible if the training dataset is inaccessible at the time of unlearning. Unlike previous approaches requiring a complete set of targets, we consider few-shot unlearning scenario when only a few samples of target data are available. To this end, we formulate the few-shot unlearning problem specifying intentio…
Unlearning in- vs. out-of-distribution data in LLMs under gradient-based method
School of Cybersecurity and Privacy, Georgia Institute of Technology（佐治亚理工学院网络安全与隐私学院）；Google Brain（谷歌大脑）；National University of Singapore（新加坡国立大学）；Google DeepMind（谷歌DeepMind）
