FedUP: Uncertainty-Aware Personalized Federated Learning via Probabilistic Prototypes

Qi, Furui; Zhang, Weishan; Meng, Lingzhao; Liu, Yuru; Feng, Zijun; Liu, Yuange; Zhang, Baoyu; Luo, Daobin; Chen, Tao

doi:10.24963/ijcai.2026/171

FedUP: Uncertainty-Aware Personalized Federated Learning via Probabilistic Prototypes

Furui Qi, Weishan Zhang, Lingzhao Meng, Yuru Liu, Zijun Feng, Yuange Liu, Baoyu Zhang, Daobin Luo, Tao Chen

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. Main Track. Pages 1532-1540.

https://doi.org/10.24963/ijcai.2026/171

Keywords

Machine Learning: Federated learning
Computer Vision: Biomedical image analysis
Computer Vision: Recognition (object detection, categorization)

Abstract

Prototype-based federated learning enables efficient knowledge sharing by exchanging class prototypes rather than full model parameters. However, heterogeneous client data and limited local samples increase prototype estimation variance, making many client prototypes unreliable. Existing methods usually treat prototypes as deterministic point estimates and cannot quantify their reliability, which may contaminate global prototypes and cause negative transfer. To address these challenges, we propose FedUP, an uncertainty-aware personalized federated learning framework that models prototypes as probability distributions. FedUP captures both aleatoric and epistemic uncertainty on a probabilistic simplex to guide local training and global aggregation. It further uses global probabilistic prototypes as class-conditional priors for feature augmentation, alleviating client-side data sparsity. Finally, FedUP aggregates prototype distributions through reliability-informed barycenters on the statistical manifold, suppressing unreliable contributions while preserving geometric structure. Experiments on natural and medical benchmarks demonstrate that FedUP consistently outperforms state-of-the-art methods.

Furui Qi, Weishan Zhang, Lingzhao Meng, Yuru Liu, Zijun Feng, Yuange Liu, Baoyu Zhang, Daobin Luo, and Tao Chen. 2026. FedUP: Uncertainty-Aware Personalized Federated Learning via Probabilistic Prototypes. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 1532–1540. https://doi.org/10.24963/ijcai.2026/171

@inproceedings{ijcai2026p171,
  title     = {FedUP: Uncertainty-Aware Personalized Federated Learning via Probabilistic Prototypes},
  author    = {Qi, Furui and Zhang, Weishan and Meng, Lingzhao and Liu, Yuru and Feng, Zijun and Liu, Yuange and Zhang, Baoyu and Luo, Daobin and Chen, Tao},
  booktitle = {Proceedings of the Thirty-Fifth International Joint Conference on
               Artificial Intelligence, {IJCAI-26}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Diego Calvanese},
  pages     = {1532--1540},
  year      = {2026},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2026/171},
  url       = {https://doi.org/10.24963/ijcai.2026/171},
}

TY  - CPAPER
TI  - FedUP: Uncertainty-Aware Personalized Federated Learning via Probabilistic Prototypes
AU  - Qi, Furui
AU  - Zhang, Weishan
AU  - Meng, Lingzhao
AU  - Liu, Yuru
AU  - Feng, Zijun
AU  - Liu, Yuange
AU  - Zhang, Baoyu
AU  - Luo, Daobin
AU  - Chen, Tao
T2  - Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2026
SP  - 1532
EP  - 1540
DO  - 10.24963/ijcai.2026/171
UR  - https://doi.org/10.24963/ijcai.2026/171
ER  -