AI Paper Atlas

Research map / Federated learning and privacy / Communication-efficient FL

Adaptive Guidance for Local Training in Heterogeneous Federated Learning

Jianqing Zhang, Yang Liu, Hua Yang, Zhenzhe Zheng, Han Yu, Yuyin Zhou, Jian Cao, Qiang Yang

TPAMI 2026 · Federated learning and privacy › Communication-efficient FL · approach: Client participation analysis · 3 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 global guiding vectors (FedL2G-l) over local training only (ablation)local training · Cifar100 (Dirichlet)accuracy (%): 35.6 -> 42.3Improved
add global guiding vectors (FedL2G-f) over local training only (ablation)local training · Cifar10 (Dirichlet)accuracy (%): 83.2 -> 87.6Improved
use feature-space sharing (FedL2G-f) instead of logit spaceFedL2G-l · Cifar100 (pathological)accuracy (%): 59.2 -> 64.2Improved
use feature-space sharing (FedL2G-f) instead of logit spaceFedL2G-l · TinyImageNet (Dirichlet)accuracy (%): 49.5 -> 50.3Improved
add warm-up period T'=50 to FedL2G-f (ablation)FedL2G-f · Cifar100 (Dirichlet)accuracy (%): 40.9 -> 43.8Improved
add warm-up period T'=50 to FedL2G-l (ablation)FedL2G-l · Cifar100 (Dirichlet)accuracy (%): 41.7 -> 42.3No clear effect
raise server learning rate eta_s of FedL2G-f from 1 to 100 (ablation)FedL2G-f · Cifar100 (Dirichlet)accuracy (%): 41.1 -> 43.8Improved
shrink quiz set size from 10 to 2 (ablation)FedL2G-f · Cifar100 (Dirichlet)accuracy (%): 43.8 -> 44.2No clear effect
add Gaussian noise to guiding-vector gradients (s=0.05, p=0.2) (ablation)FedL2G-f · Cifar100 (Dirichlet)accuracy (%): 43.8 -> 42.6Hurt
increase heterogeneity to HtFE9 with FedL2G-f vs FedProtoFedProto · Cifar100 (Dirichlet, HtFE9)accuracy (%): 32.7 -> 45.7Improved
increase clients to 100 with participation 0.1FedMRL · Cifar100 (Dirichlet, 100 clients, rho=0.1)accuracy (%): 41.7 -> 42.3Improved
FedL2G-f vs best baseline FedDistill at beta=0.01FedDistill · Cifar100 (Dirichlet, beta=0.01)accuracy (%): 67.0 -> 70.6Improved

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