CVCLCRLGOct 16, 2024

Meta-Unlearning on Diffusion Models: Preventing Relearning Unlearned Concepts

Tsinghua
arXiv:2410.12777v233 citationsh-index: 41Has Code
Originality Incremental advance
AI Analysis

This addresses a security issue for diffusion model developers and users by preventing misuse through relearning, though it is incremental as it builds on existing unlearning methods.

The paper tackles the problem of malicious finetuning causing diffusion models to relearn unlearned harmful or copyrighted concepts, proposing meta-unlearning to trigger self-destruction of related benign concepts during such finetuning, with validation on Stable Diffusion models.

With the rapid progress of diffusion-based content generation, significant efforts are being made to unlearn harmful or copyrighted concepts from pretrained diffusion models (DMs) to prevent potential model misuse. However, it is observed that even when DMs are properly unlearned before release, malicious finetuning can compromise this process, causing DMs to relearn the unlearned concepts. This occurs partly because certain benign concepts (e.g., "skin") retained in DMs are related to the unlearned ones (e.g., "nudity"), facilitating their relearning via finetuning. To address this, we propose meta-unlearning on DMs. Intuitively, a meta-unlearned DM should behave like an unlearned DM when used as is; moreover, if the meta-unlearned DM undergoes malicious finetuning on unlearned concepts, the related benign concepts retained within it will be triggered to self-destruct, hindering the relearning of unlearned concepts. Our meta-unlearning framework is compatible with most existing unlearning methods, requiring only the addition of an easy-to-implement meta objective. We validate our approach through empirical experiments on meta-unlearning concepts from Stable Diffusion models (SD-v1-4 and SDXL), supported by extensive ablation studies. Our code is available at https://github.com/sail-sg/Meta-Unlearning.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes