SPIRE: Conditional Personalization for Federated Diffusion Generative Models

Kaan Ozkara, Ruida Zhou, Suhas Diggavi
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:1666-1674, 2026.

Abstract

Two defining characteristics of federated learning (FL) client data are distributional heterogeneity and small local sample sizes. These properties necessitate data efficient, and client specific adaptation rather than a one-size-fits-all model. Recent advances in diffusion models have revolutionized generative AI. However, their scale is too large for straightforward fine-tuning; making personalization difficult. To enable personalized diffusion generative models, we propose Shared-backbone Personal Identity Representation Embeddings (SPIRE), a framework that casts per-client diffusion based generation as conditional generation in FL. SPIRE factorizes the network into (i) a high-capacity global backbone that learns a population-level score function and (ii) lightweight, learnable client embeddings that encode local data statistics. This separation enables parameter-efficient fine-tuning that touches $<0.01%$ of weights. We provide the first theoretical bridge between conditional diffusion training and maximum-likelihood estimation in Gaussian-mixture models. For a two-component mixture we prove that gradient descent on the DDPM with respect to mixing weights loss recovers the optimal mixing weights and enjoys dimension-free error bounds. Our analysis also hints at how client embeddings act as biases that steer a shared score network toward personalized distributions. Empirically, SPIRE matches or surpasses strong baselines during collaborative pre-training, and vastly outperforms them when adapting to unseen/new clients—reducing Kernel Inception Distance while updating only hundreds of parameters. SPIRE further mitigates catastrophic forgetting and remains robust across fine-tuning learning-rate and epoch choices.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-ozkara26a, title = { SPIRE: Conditional Personalization for Federated Diffusion Generative Models }, author = {Ozkara, Kaan and Zhou, Ruida and Diggavi, Suhas}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {1666--1674}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/ozkara26a/ozkara26a.pdf}, url = {https://proceedings.mlr.press/v300/ozkara26a.html}, abstract = { Two defining characteristics of federated learning (FL) client data are distributional heterogeneity and small local sample sizes. These properties necessitate data efficient, and client specific adaptation rather than a one-size-fits-all model. Recent advances in diffusion models have revolutionized generative AI. However, their scale is too large for straightforward fine-tuning; making personalization difficult. To enable personalized diffusion generative models, we propose Shared-backbone Personal Identity Representation Embeddings (SPIRE), a framework that casts per-client diffusion based generation as conditional generation in FL. SPIRE factorizes the network into (i) a high-capacity global backbone that learns a population-level score function and (ii) lightweight, learnable client embeddings that encode local data statistics. This separation enables parameter-efficient fine-tuning that touches $<0.01%$ of weights. We provide the first theoretical bridge between conditional diffusion training and maximum-likelihood estimation in Gaussian-mixture models. For a two-component mixture we prove that gradient descent on the DDPM with respect to mixing weights loss recovers the optimal mixing weights and enjoys dimension-free error bounds. Our analysis also hints at how client embeddings act as biases that steer a shared score network toward personalized distributions. Empirically, SPIRE matches or surpasses strong baselines during collaborative pre-training, and vastly outperforms them when adapting to unseen/new clients—reducing Kernel Inception Distance while updating only hundreds of parameters. SPIRE further mitigates catastrophic forgetting and remains robust across fine-tuning learning-rate and epoch choices. } }
Endnote
%0 Conference Paper %T SPIRE: Conditional Personalization for Federated Diffusion Generative Models %A Kaan Ozkara %A Ruida Zhou %A Suhas Diggavi %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-ozkara26a %I PMLR %P 1666--1674 %U https://proceedings.mlr.press/v300/ozkara26a.html %V 300 %X Two defining characteristics of federated learning (FL) client data are distributional heterogeneity and small local sample sizes. These properties necessitate data efficient, and client specific adaptation rather than a one-size-fits-all model. Recent advances in diffusion models have revolutionized generative AI. However, their scale is too large for straightforward fine-tuning; making personalization difficult. To enable personalized diffusion generative models, we propose Shared-backbone Personal Identity Representation Embeddings (SPIRE), a framework that casts per-client diffusion based generation as conditional generation in FL. SPIRE factorizes the network into (i) a high-capacity global backbone that learns a population-level score function and (ii) lightweight, learnable client embeddings that encode local data statistics. This separation enables parameter-efficient fine-tuning that touches $<0.01%$ of weights. We provide the first theoretical bridge between conditional diffusion training and maximum-likelihood estimation in Gaussian-mixture models. For a two-component mixture we prove that gradient descent on the DDPM with respect to mixing weights loss recovers the optimal mixing weights and enjoys dimension-free error bounds. Our analysis also hints at how client embeddings act as biases that steer a shared score network toward personalized distributions. Empirically, SPIRE matches or surpasses strong baselines during collaborative pre-training, and vastly outperforms them when adapting to unseen/new clients—reducing Kernel Inception Distance while updating only hundreds of parameters. SPIRE further mitigates catastrophic forgetting and remains robust across fine-tuning learning-rate and epoch choices.
APA
Ozkara, K., Zhou, R. & Diggavi, S.. (2026). SPIRE: Conditional Personalization for Federated Diffusion Generative Models . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:1666-1674 Available from https://proceedings.mlr.press/v300/ozkara26a.html.

Related Material