AI Paper Atlas

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

Towards Rule-Based Knowledge Sharing in Federated Learning

Zixuan Qin, Qi Shen, Liu Yang, Qilong Wang, Qinghua Hu

ICML 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Non-IID federated learning

Read the paper

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 rule distillation (KD) loss to rule regularization (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 56.06 -> 58.25Improved
add rule regularization loss to KD (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 53.59 -> 58.25Improved
raise KD weight alpha_kd from 0.1 to 0.6 (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 55.72 -> 58.17Improved
raise KD weight alpha_kd from 0.6 to 0.9 (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 58.17 -> 57.19Hurt
raise rule weight alpha_ru from 0.5 to 1.5 (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 56.05 -> 58.23Improved
raise rule weight alpha_ru from 1.5 to 2.0 (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 58.23 -> 56.72Hurt
raise margin m from 0.2 to 0.7 (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 55.20 -> 58.21Improved
raise margin m from 0.7 to 0.9 (ablation)RFL · CIFAR-100 (Dir 0.1)Accuracy: 58.21 -> 57.05Hurt
share rules instead of class prototypesRFL vs FedTGP · CIFAR-100 (Dir 0.3)Accuracy: 39.34 -> 43.80Improved
share rules instead of class prototypesRFL vs FedProto · CIFAR-100 (Pathological)Accuracy: 58.92 -> 63.02Improved
use rule carrier instead of proxy-model carrierRFL vs FedSCE · CIFAR-100Communication cost (MB/round): 265.36 -> 41.98Improved
use rule carrier instead of proxy-model carrierRFL vs FedSCE · CIFAR-100 (Dir 0.3)Accuracy: 44.53 vs 43.80No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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