Automating Evaluation of Diffusion Model Unlearning with (Vision-) Language Model World Knowledge
This work addresses the challenge of verifying unlearning in diffusion models for researchers and practitioners, but it is incremental as it builds on existing unlearning methods with a new evaluation tool.
The authors tackled the problem of evaluating machine unlearning in diffusion models by introducing autoeval-dmun, an automated tool that uses vision-language models to assess unlearning effectiveness and identify damage to nearby concepts, revealing that language models impose semantic orderings correlating with damage and can circumvent unlearning with adversarial prompts.
Machine unlearning (MU) is a promising cost-effective method to cleanse undesired information (generated concepts, biases, or patterns) from foundational diffusion models. While MU is orders of magnitude less costly than retraining a diffusion model without the undesired information, it can be challenging and labor-intensive to prove that the information has been fully removed from the model. Moreover, MU can damage diffusion model performance on surrounding concepts that one would like to retain, making it unclear if the diffusion model is still fit for deployment. We introduce autoeval-dmun, an automated tool which leverages (vision-) language models to thoroughly assess unlearning in diffusion models. Given a target concept, autoeval-dmun extracts structured, relevant world knowledge from the language model to identify nearby concepts which are likely damaged by unlearning and to circumvent unlearning with adversarial prompts. We use our automated tool to evaluate popular diffusion model unlearning methods, revealing that language models (1) impose semantic orderings of nearby concepts which correlate well with unlearning damage and (2) effectively circumvent unlearning with synthetic adversarial prompts.