SurrogateSHAP: Training-Free Contributor Attribution for Text-to-Image (T2I) Models

TL;DR

SurrogateSHAP introduces a training-free, efficient method for attributing contributions of data sources in text-to-image models, enabling fairer data valuation and model auditing without costly retraining.

Contribution

The paper presents SurrogateSHAP, a novel framework that approximates Shapley values using inference and a gradient-boosted tree, eliminating the need for retraining models.

Findings

Outperforms prior attribution methods in diverse tasks.

Reduces computational overhead significantly.

Effectively localizes data sources for spurious correlations.

Abstract

As Text-to-Image (T2I) diffusion models are increasingly used in real-world creative workflows, a principled framework for valuing contributors who provide a collection of data is essential for fair compensation and sustainable data marketplaces. While the Shapley value offers a theoretically grounded approach to attribution, it faces a dual computational bottleneck: (i) the prohibitive cost of exhaustive model retraining for each sampled subset of players (i.e., data contributors) and (ii) the combinatorial number of subsets needed to estimate marginal contributions due to contributor interactions. To this end, we propose SurrogateSHAP, a retraining-free framework that approximates the expensive retraining game through inference from a pretrained model. To further improve efficiency, we employ a gradient-boosted tree to approximate the utility function and derive Shapley values analytically from the tree-based model. We evaluate SurrogateSHAP across three diverse attribution tasks: (i) image quality for DDPM-CFG on CIFAR-20, (ii) aesthetics for Stable Diffusion on Post-Impressionist artworks, and (iii) product diversity for FLUX.1 on Fashion-Product data. Across settings, SurrogateSHAP outperforms prior methods while substantially reducing computational overhead, consistently identifying influential contributors across multiple utility metrics. Finally, we demonstrate that SurrogateSHAP effectively localizes data sources responsible for spurious correlations in clinical images, providing a scalable path toward auditing safety-critical generative models.