AI Paper Atlas

Research map / Federated learning and privacy / Personalized and heterogeneous FL

FedHPro: Federated Hyper-Prototype Learning via Gradient Matching

Huan Wang, Jun Shen, Haoran Li, Zhenyu Yang, Jun Yan, Ousman Manjang, Yanlong Zhai, Di Wu, Guansong Pang

ICML 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated learning with pretrained models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add HPCL (hyper-prototype contrastive learning) to FedAvg (ablation)FedAvg · Digitsavg accuracy: 78.82 -> 83.58Improved
add HPAL (hyper-prototype alignment learning) to FedAvg (ablation)FedAvg · Digitsavg accuracy: 78.82 -> 83.94Improved
combine HPCL and HPAL (ablation)FedHPro · Digitsavg accuracy: 83.94 -> 84.80Improved
combine HPCL and HPAL (ablation)FedHPro · Office-Caltechavg accuracy: 61.34 -> 64.52Improved
replace hyper-prototypes with FedAvg-style global prototypes P in both HPCL and HPAL (ablation)FedHPro · Digitsavg accuracy: 84.80 -> 81.35Hurt
replace hyper-prototypes with global prototypes P in both modules (ablation)FedHPro · Office-Caltechavg accuracy: 64.52 -> 59.69Hurt
replace hyper-prototypes with global prototypes P in both modules (ablation)FedHPro · CIFAR10-LTaccuracy: 74.69 -> 72.56Hurt
replace global prototypes with hyper-prototypes in FedProtoFedProto · Digitsavg accuracy: 80.41 -> 81.75Improved
plug hyper-prototypes into FedTGP, FedGMKD and FedSAFedTGP, FedGMKD, FedSA · Digitsavg accuracy: FedTGP 82.61, FedGMKD 82.79, FedSA 83.25 (+0.94, +0.96, +0.71)Improved
add client-specific margin d_k in HPCL (ablation)FedHPro · CIFAR10 NID10.5, CIFAR10-LT, Digits, Office-Caltechaccuracy: Digits 84.22 -> 84.80; Office-Caltech 63.90 -> 64.52; CIFAR10 88.83 -> 89.56Improved
increase hyper-prototype rounds M up to 30 (ablation)FedHPro · Digits and CIFAR10-LTaccuracy: gain marginal beyond M=30Improved
excessively increase hyper-prototype length |I| (ablation)FedHPro · Digits and CIFAR10-LTaccuracyHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy