Unconsciously Forget: Mitigating Memorization; Without Knowing What is being Memorized

TL;DR

This paper introduces UniForget, a model pruning technique that reduces memorization of copyrighted content in generative models without needing to identify specific concepts, thereby enhancing privacy and legal compliance.

Contribution

UniForget provides a novel, scalable approach to mitigate memorization by pruning responsible model parts, complementing existing unlearning methods without high computational costs.

Findings

Pruning specific model parts reduces copyrighted content generation.

The method preserves overall generative quality.

It is orthogonal and complementary to existing unlearning techniques.

Abstract

Recent advances in generative models have demonstrated an exceptional ability to produce highly realistic images. However, previous studies show that generated images often resemble the training data, and this problem becomes more severe as the model size increases. Memorizing training data can lead to legal challenges, including copyright infringement, violations of portrait rights, and trademark violations. Existing approaches to mitigating memorization mainly focus on manipulating the denoising sampling process to steer image embeddings away from the memorized embedding space or employ unlearning methods that require training on datasets containing specific sets of memorized concepts. However, existing methods often incur substantial computational overhead during sampling, or focus narrowly on removing one or more groups of target concepts, imposing a significant limitation on their scalability. To understand and mitigate these problems, our work, UniForget, offers a new perspective on understanding the root cause of memorization. Our work demonstrates that specific parts of the model are responsible for copyrighted content generation. By applying model pruning, we can effectively suppress the probability of generating copyrighted content without targeting specific concepts while preserving the general generative capabilities of the model. Additionally, we show that our approach is both orthogonal and complementary to existing unlearning methods, thereby highlighting its potential to improve current unlearning and de-memorization techniques.