Training data attribution in diffusion models via mirrored unlearning and noise-consistent skew
Abstract
Training data attribution (TDA) should enable generative model interpretability and foster a variety of related downstream tasks. Nonetheless, current TDA approaches lack reliability and robustness, preventing their adoption in real-world setups. In this paper, we take a decisive step towards more reliable and robust TDA for diffusion models. We propose to perform TDA with mirrored unlearning and noise-consistent skew (MUCS). The idea is to fine-tune a second model with bounded mirrored gradient ascent, and to measure the normalized skew of this model with respect to the original one using consistent noise samples. We show that, while being conceptually simple and generic, MUCS systematically outperforms existing methods on three different datasets by a large margin. We additionally study the effect that core design choices have on final performance, and analyze novel aspects regarding the overlap of influential instances across generated items and the potential of ensembling TDA approaches. We believe that our findings may have broader implications for more general unlearning setups, as well as for tasks requiring the comparison of diffusion losses.
Related
- A Unified Framework for Training-Dataset Aggregation and Exact Machine Unlearning via Gibbs Probability Measures
- Free Boundary Propagation of a Thin-Film Equation with Nonlinear Diffusion: Scale-Consistent Methods for Optimal Results
- Radial Tail Repair: Post-Training Correction for Diffusion and Flow Models
- VGAS: Variance-Reduced Guidance and Adaptive Selection for Training-Free Reward Alignment in Discrete Diffusion
- Feature Selective Model Collapse in Diffusion Models: Total Replacement versus Fixed-Budget Training
BibTeX
@misc{serra2026training,
title = {{Training data attribution in diffusion models via mirrored unlearning and noise-consistent skew}},
author = {Joan Serrà and Dipam Goswami and Fabio Morreale and Wei-Hsiang Liao and Yuki Mitsufuji},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2605.17938},
eprint = {2605.17938},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.17938},
}
RIS
TY - UNPB
TI - Training data attribution in diffusion models via mirrored unlearning and noise-consistent skew
AU - Joan Serrà
AU - Dipam Goswami
AU - Fabio Morreale
AU - Wei-Hsiang Liao
AU - Yuki Mitsufuji
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2605.17938
UR - https://arxiv.org/abs/2605.17938
AB - Training data attribution (TDA) should enable generative model interpretability and foster a variety of related downstream tasks. Nonetheless, current TDA approaches lack reliability and robustness, preventing their adoption in real-world setups. In this paper, we take a decisive step towards more reliable and robust TDA for diffusion models. We propose to perform TDA with mirrored unlearning and noise-consistent skew (MUCS). The idea is to fine-tune a second model with bounded mirrored gradient ascent, and to measure the normalized skew of this model with respect to the original one using consistent noise samples. We show that, while being conceptually simple and generic, MUCS systematically outperforms existing methods on three different datasets by a large margin. We additionally study the effect that core design choices have on final performance, and analyze novel aspects regarding the overlap of influential instances across generated items and the potential of ensembling TDA approaches. We believe that our findings may have broader implications for more general unlearning setups, as well as for tasks requiring the comparison of diffusion losses.
ER -