Fast Data Attribution for Text-to-Image Models

2025-11-13

AI Summary

Attribution of training data for text-to-image models suffers from prohibitive computational costs, hindering practical deployment. This paper proposes an efficient attribution framework that distills expensive unlearning-based attribution into the feature embedding space via knowledge distillation, learnable embedding modeling, and approximate nearest-neighbor search—enabling millisecond-scale retrieval of high-impact training samples. Evaluated on large-scale generative models (e.g., Stable Diffusion) using MSCOCO and LAION, our method achieves attribution in seconds per query—2,500× to 400,000× faster than state-of-the-art approaches—while matching or exceeding their accuracy. Our key contribution is the first effective transfer of unlearning-based attribution to the embedding space, reconciling high fidelity with real-world efficiency. This enables scalable, practical data provenance and compliance auditing for generative AI systems.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

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.

Problem

• Identifying influential training images for text-to-image model outputs
• Reducing computational costs of existing attribution methods significantly
• Enabling scalable data attribution for real-world applications efficiently

Innovation

• Distills unlearning-based attribution to embedding space
• Enables efficient retrieval via indexing and search methods
• Achieves 2,500x-400,000x speedup over existing attribution methods