Diffusion Soup: Model Merging for Text-to-Image Diffusion Models

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Biggs, Benjamin, Seshadri, Arjun, Zou, Yang, Jain, Achin, Golatkar, Aditya, Xie, Yusheng, Achille, Alessandro, Swaminathan, Ashwin, Soatto, Stefano
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866909222520750080
author Biggs, Benjamin
Seshadri, Arjun
Zou, Yang
Jain, Achin
Golatkar, Aditya
Xie, Yusheng
Achille, Alessandro
Swaminathan, Ashwin
Soatto, Stefano
author_facet Biggs, Benjamin
Seshadri, Arjun
Zou, Yang
Jain, Achin
Golatkar, Aditya
Xie, Yusheng
Achille, Alessandro
Swaminathan, Ashwin
Soatto, Stefano
contents We present Diffusion Soup, a compartmentalization method for Text-to-Image Generation that averages the weights of diffusion models trained on sharded data. By construction, our approach enables training-free continual learning and unlearning with no additional memory or inference costs, since models corresponding to data shards can be added or removed by re-averaging. We show that Diffusion Soup samples from a point in weight space that approximates the geometric mean of the distributions of constituent datasets, which offers anti-memorization guarantees and enables zero-shot style mixing. Empirically, Diffusion Soup outperforms a paragon model trained on the union of all data shards and achieves a 30% improvement in Image Reward (.34 $\to$ .44) on domain sharded data, and a 59% improvement in IR (.37 $\to$ .59) on aesthetic data. In both cases, souping also prevails in TIFA score (respectively, 85.5 $\to$ 86.5 and 85.6 $\to$ 86.8). We demonstrate robust unlearning -- removing any individual domain shard only lowers performance by 1% in IR (.45 $\to$ .44) -- and validate our theoretical insights on anti-memorization using real data. Finally, we showcase Diffusion Soup's ability to blend the distinct styles of models finetuned on different shards, resulting in the zero-shot generation of hybrid styles.
format Preprint
id arxiv_https___arxiv_org_abs_2406_08431
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Diffusion Soup: Model Merging for Text-to-Image Diffusion Models
Biggs, Benjamin
Seshadri, Arjun
Zou, Yang
Jain, Achin
Golatkar, Aditya
Xie, Yusheng
Achille, Alessandro
Swaminathan, Ashwin
Soatto, Stefano
Computer Vision and Pattern Recognition
Artificial Intelligence
Cryptography and Security
Machine Learning
We present Diffusion Soup, a compartmentalization method for Text-to-Image Generation that averages the weights of diffusion models trained on sharded data. By construction, our approach enables training-free continual learning and unlearning with no additional memory or inference costs, since models corresponding to data shards can be added or removed by re-averaging. We show that Diffusion Soup samples from a point in weight space that approximates the geometric mean of the distributions of constituent datasets, which offers anti-memorization guarantees and enables zero-shot style mixing. Empirically, Diffusion Soup outperforms a paragon model trained on the union of all data shards and achieves a 30% improvement in Image Reward (.34 $\to$ .44) on domain sharded data, and a 59% improvement in IR (.37 $\to$ .59) on aesthetic data. In both cases, souping also prevails in TIFA score (respectively, 85.5 $\to$ 86.5 and 85.6 $\to$ 86.8). We demonstrate robust unlearning -- removing any individual domain shard only lowers performance by 1% in IR (.45 $\to$ .44) -- and validate our theoretical insights on anti-memorization using real data. Finally, we showcase Diffusion Soup's ability to blend the distinct styles of models finetuned on different shards, resulting in the zero-shot generation of hybrid styles.
title Diffusion Soup: Model Merging for Text-to-Image Diffusion Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2406.08431