Fast Data Attribution for Text-to-Image Models

TL;DR

This paper introduces a scalable, efficient data attribution method for text-to-image models that significantly reduces computation time by using a distilled embedding space, enabling rapid identification of influential training images.

Contribution

We propose a novel distillation-based approach that enables fast, scalable data attribution for large text-to-image models, outperforming existing methods in speed and efficiency.

Findings

Achieves 2,500x to 400,000x faster attribution than prior methods.

Effective on both medium-scale and large-scale models trained on MSCOCO and LAION.

Demonstrates practical applicability for real-world models like Stable Diffusion.

Abstract

Data attribution for text-to-image models aims to identify the training images that most significantly influenced a generated output. Existing attribution methods involve considerable computational resources for each query, making them impractical for real-world applications. We propose a novel approach for scalable and efficient data attribution. Our key idea is to distill a slow, unlearning-based attribution method to a feature embedding space for efficient retrieval of highly influential training images. During deployment, combined with efficient indexing and search methods, our method successfully finds highly influential images without running expensive attribution algorithms. We show extensive results on both medium-scale models trained on MSCOCO and large-scale Stable Diffusion models trained on LAION, demonstrating that our method can achieve better or competitive performance in a few seconds, faster than existing methods by 2,500x - 400,000x. Our work represents a meaningful step towards the large-scale application of data attribution methods on real-world models such as Stable Diffusion.