Intriguing Properties of Data Attribution on Diffusion Models

TL;DR

This paper investigates data attribution methods for diffusion models, revealing surprising empirical results that challenge theoretical assumptions and proposing a more efficient attribution approach based on extensive experiments.

Contribution

It introduces a novel, efficient data attribution method for diffusion models and uncovers counter-intuitive empirical findings that question existing theoretical guidance.

Findings

Counter-intuitive design choices outperform baselines

The new method is more computationally efficient

Theoretical assumptions may hinder attribution performance

Abstract

Data attribution seeks to trace model outputs back to training data. With the recent development of diffusion models, data attribution has become a desired module to properly assign valuations for high-quality or copyrighted training samples, ensuring that data contributors are fairly compensated or credited. Several theoretically motivated methods have been proposed to implement data attribution, in an effort to improve the trade-off between computational scalability and effectiveness. In this work, we conduct extensive experiments and ablation studies on attributing diffusion models, specifically focusing on DDPMs trained on CIFAR-10 and CelebA, as well as a Stable Diffusion model LoRA-finetuned on ArtBench. Intriguingly, we report counter-intuitive observations that theoretically unjustified design choices for attribution empirically outperform previous baselines by a large margin, in terms of both linear datamodeling score and counterfactual evaluation. Our work presents a significantly more efficient approach for attributing diffusion models, while the unexpected findings suggest that at least in non-convex settings, constructions guided by theoretical assumptions may lead to inferior attribution performance. The code is available at https://github.com/sail-sg/D-TRAK.