FedNeuro: Multi-Site fMRI Analysis Using Hypernetwork Personalized and Privacy Enhanced Federated Learning

Sunny Gupta, Shambhavi Shanker, Amit Sethi
Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare, PMLR 317:135-140, 2026.

Abstract

Multi-site functional MRI (fMRI) studies enable comprehensive understanding of brain disorders by integrating data across institutions, yet centralized model training remains limited by privacy regulations and domain heterogeneity. We propose FedNeuro, a hypernetwork-personalized and privacy-enhanced federated learning framework for collaborative fMRI analysis. Unlike conventional federated averaging, FedNeuro employs a global hypernetwork that generates site-specific model parameters from private client embeddings, allowing each site to learn personalized representations while maintaining global consistency. This bi-level meta-optimization decouples data-dependent gradients from shared parameters, providing structural privacy protection and improving cross-site generalization. Evaluated on the multi-site ABIDE dataset, FedNeuro achieves robust gains over federated baselines, marking a step toward scalable, privacy-preserving, and domain-fair neuroimaging for precision neuroscience. Code: https://github.com/sunnyinAI/FedNeuro

Cite this Paper

BibTeX
@InProceedings{pmlr-v317-gupta26b, title = {FedNeuro: Multi-Site fMRI Analysis Using Hypernetwork Personalized and Privacy Enhanced Federated Learning}, author = {Gupta, Sunny and Shanker, Shambhavi and Sethi, Amit}, booktitle = {Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare}, pages = {135--140}, year = {2026}, editor = {Wu, Junde and Pan, Jiazhen and Zhu, Jiayuan and Luo, Luyang and Li, Yitong and Xu, Min and Jin, Yueming and Rueckert, Daniel}, volume = {317}, series = {Proceedings of Machine Learning Research}, month = {20--21 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v317/main/assets/gupta26b/gupta26b.pdf}, url = {https://proceedings.mlr.press/v317/gupta26b.html}, abstract = {Multi-site functional MRI (fMRI) studies enable comprehensive understanding of brain disorders by integrating data across institutions, yet centralized model training remains limited by privacy regulations and domain heterogeneity. We propose FedNeuro, a hypernetwork-personalized and privacy-enhanced federated learning framework for collaborative fMRI analysis. Unlike conventional federated averaging, FedNeuro employs a global hypernetwork that generates site-specific model parameters from private client embeddings, allowing each site to learn personalized representations while maintaining global consistency. This bi-level meta-optimization decouples data-dependent gradients from shared parameters, providing structural privacy protection and improving cross-site generalization. Evaluated on the multi-site ABIDE dataset, FedNeuro achieves robust gains over federated baselines, marking a step toward scalable, privacy-preserving, and domain-fair neuroimaging for precision neuroscience. Code: https://github.com/sunnyinAI/FedNeuro} }
Endnote
%0 Conference Paper %T FedNeuro: Multi-Site fMRI Analysis Using Hypernetwork Personalized and Privacy Enhanced Federated Learning %A Sunny Gupta %A Shambhavi Shanker %A Amit Sethi %B Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare %C Proceedings of Machine Learning Research %D 2026 %E Junde Wu %E Jiazhen Pan %E Jiayuan Zhu %E Luyang Luo %E Yitong Li %E Min Xu %E Yueming Jin %E Daniel Rueckert %F pmlr-v317-gupta26b %I PMLR %P 135--140 %U https://proceedings.mlr.press/v317/gupta26b.html %V 317 %X Multi-site functional MRI (fMRI) studies enable comprehensive understanding of brain disorders by integrating data across institutions, yet centralized model training remains limited by privacy regulations and domain heterogeneity. We propose FedNeuro, a hypernetwork-personalized and privacy-enhanced federated learning framework for collaborative fMRI analysis. Unlike conventional federated averaging, FedNeuro employs a global hypernetwork that generates site-specific model parameters from private client embeddings, allowing each site to learn personalized representations while maintaining global consistency. This bi-level meta-optimization decouples data-dependent gradients from shared parameters, providing structural privacy protection and improving cross-site generalization. Evaluated on the multi-site ABIDE dataset, FedNeuro achieves robust gains over federated baselines, marking a step toward scalable, privacy-preserving, and domain-fair neuroimaging for precision neuroscience. Code: https://github.com/sunnyinAI/FedNeuro
APA
Gupta, S., Shanker, S. & Sethi, A.. (2026). FedNeuro: Multi-Site fMRI Analysis Using Hypernetwork Personalized and Privacy Enhanced Federated Learning. Proceedings of The Second AAAI Bridge Program on AI for Medicine and Healthcare, in Proceedings of Machine Learning Research 317:135-140 Available from https://proceedings.mlr.press/v317/gupta26b.html.

Related Material