AI Paper Atlas

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

MAS: Towards Resource-Efficient Federated Multiple-Task Learning

Weiming Zhuang, Yonggang Wen, Lingjuan Lyu, Shuai Zhang

ICCV 2023 · Federated learning and privacy › Communication-efficient FL · approach: Efficient local training in FL · 30 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
merge FL tasks into one multi-task all-in-one model, then split by affinitymultiple simultaneous FL tasks · Taskonomy sdnkttraining time and energy: ~2x training time reduction, ~40% less energyImproved
initialize each split from all-in-one model instead of from scratch (ablation)MAS splits · Taskonomy sdnkt/erckttotal test loss: optimal split sdnkt-2: 0.622 -> 0.595Improved
use MAS affinity-based splitting instead of optimal split found by exhaustive search (ablation)MAS · Taskonomy sdnkt (2 splits)total test loss: 0.595 -> 0.578Improved
apply GradNorm or FedProx to all-in-one trainingall-in-one FL · Taskonomy sdnkttotal test lossNo clear effect
increase number of splits (MAS-2 to MAS-3 to MAS-4) (ablation)MAS · Taskonomy sdnkttotal test loss: 0.578 -> 0.555 (2 to 3 splits)Improved
increase number of splits on nine-task set sdnkterca (ablation)MAS · Taskonomy sdnktercatotal test lossNo clear effect
train all-in-one for R0 of 30 to 40 rounds before splitting (of 100) (ablation)MAS · Taskonomy task setstotal test lossImproved
increase local epochs E from 5 to 10 (ablation)all-in-one FL · Taskonomy sdnkttotal test lossNo clear effect
increase selected clients K from 4 to 8 (ablation)MAS-2 · Taskonomy sdnkttotal test loss: All-in-one 0.618 -> MAS-2 0.512 at K=8Improved
use HOA task grouping instead of MAStask grouping · Taskonomy sdnkttraining time and energy: around 4-6x of oursHurt

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