An Inversion-based Measure of Memorization for Diffusion Models

2024-05-09

AI Summary

Diffusion models risk memorizing training data, posing threats to copyright compliance and user privacy. To address this, we propose InvMM—the first differentiable, optimization-friendly inversion-based memorization metric framework—which quantifies image-level memorization strength by adaptively searching the latent noise distribution, rigorously distinguishing memorization from membership inference. Our method integrates latent-space noise modeling, gradient clipping, and regularization, enabling evaluation of both unconditional and text-to-image diffusion models. Extensive experiments across diverse diffusion architectures demonstrate that InvMM accurately identifies highly memorized samples; quantitatively uncovers causal relationships between memorization and training steps, data duplication rate, and sampling strategies; and delivers a plug-and-play risk assessment tool. This significantly enhances model trustworthiness and privacy-preserving capabilities.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The past few years have witnessed substantial advances in image generation powered by diffusion models. However, it was shown that diffusion models are vulnerable to training data memorization, raising concerns regarding copyright infringement and privacy invasion. This study delves into a rigorous analysis of memorization in diffusion models. We introduce an inversion-based measure of memorization, InvMM, which searches for a sensitive latent noise distribution accounting for the replication of an image. For accurate estimation of the memorization score, we propose an adaptive algorithm that balances the normality and sensitivity of the inverted distribution. Comprehensive experiments, conducted on both unconditional and text-guided diffusion models, demonstrate that InvMM is capable of detecting heavily memorized images and elucidating the effect of various factors on memorization. Additionally, we discuss how memorization differs from membership. In practice, InvMM serves as a useful tool for model developers to reliably assess the risk of memorization, thereby contributing to the enhancement of trustworthiness and privacy-preserving capabilities of diffusion models.

Problem

• Measure memorization in diffusion models
• Assess copyright and privacy risks
• Develop reliable memorization auditing tool

Innovation

• Inversion-based measure for memorization analysis
• Adaptive algorithm balances noise distribution
• Comprehensive experiments validate memorization quantification