AI Paper Atlas

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

Taming Fat-Tailed (“Heavier-Tailed” with Potentially Infinite Variance) Noise in Federated Learning

Haibo Yang, Peiwen Qiu, Jia Liu

NeurIPS 2022 · Federated learning and privacy › Communication-efficient FL · approach: Local drift correction · 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 clipping of client updates once per communication roundFedAvg-style federated learning · Synthetic strongly convex objective with Cauchy-tailed noiseconvergenceImproved
Clip gradients before each local update instead of clipping the accumulated update once per round (ablation)FAT-Clipping-PR · Synthetic strongly convex objective with Cauchy-tailed noiseconvergence speedImproved
Clip gradients before each local updateFedAvg-style federated learning · CIFAR-10 non-i.i.d. (p=2)successful training rate: fails in all 5 trials -> 100%Improved
Clip gradients before each local updateFedAvg-style federated learning · CIFAR-10 i.i.d. (p=10)successful training rate: fails in all 5 trials -> 100%Improved
Clip gradients before each local update rather than clipping accumulated updates once per round (ablation)FAT-Clipping-PR · CIFAR-10 non-i.i.d. (p=2)successful training rate: 60% -> 100%Improved
Clip gradients before each local update rather than clipping accumulated updates once per round (ablation)FAT-Clipping-PR · CIFAR-10 i.i.d. (p=10)successful training rate: 20% -> 100%Improved
Apply clipping with per-round updates in the non-fat-tailed case (α=2) (ablation)FAT-Clipping-PR · Strongly convex and nonconvex objectives with non-fat-tailed noiseconvergence rate relative to FedAvgNo clear effect
Apply clipping with per-iteration updates in the non-fat-tailed case (α=2) (ablation)FAT-Clipping-PI · Strongly convex and nonconvex objectives with non-fat-tailed noiseconvergence rate relative to FedAvgNo clear effect

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