AI Paper Atlas

Research map / Federated learning and privacy / Privacy attacks and secure learning

FedCut: A Spectral Analysis Framework for Reliable Detection of Byzantine Colluders

Hanlin Gu, Lixin Fan, Xingxing Tang, Qiang Yang

TPAMI 2024 · Federated learning and privacy › Privacy attacks and secure learning · approach: Defending federated learning against poisoning · 5 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 spatial-temporal consistency across iterations (FedCut vs per-iteration NCut) (ablation)FedCut · MNIST/Fashion-MNIST Non-IID, 30 attackers, Fang-v1model performance: 31.7% (NCut) vs FedCut stableImproved
use FedCut instead of robust-statistics defenses under Fang-v1 attackMedian · MNIST IID, 30 attackersmodel performance: 43.4 -> 92.3Improved
use FedCut instead of Bulyan under Fang-v2 attackBulyan · MNIST IID, 30 attackersmodel performance: 18.8 -> 92.3Improved
use FedCut instead of DnC under Gaussian attackDnC · MNIST IID, 30 attackersmodel performance: 74.3 -> 92.3Improved
use FedCut instead of Kmeans under multi-group collusionKmeans · Fashion-MNIST/MNIST collusionmodel performance: 22.4 -> 91.9 worst-case MNISTImproved
increase number of attackers from 10 to 30Median · Fashion-MNISTaveraged model performance: 88.6% -> 74.6%Hurt
increase number of attackersFedCut · Fashion-MNISTaveraged model performanceNo clear effect
make client data more heterogeneous (IID to beta=0.1)Trimmed Mean · Fashion-MNISTaveraged model performance: 74.9% -> 55.4%Hurt
make client data more heterogeneousFedCut · Fashion-MNISTmodel performance: drops no more than 3%No clear effect
increase number of colluder groups from 2 to 3DnC · MNIST Non-IID beta=0.5, 30 attackersmodel performance: 91.9 -> 87.2Hurt
vary number of colluder groups from 2 to 5FedCut · MNIST Non-IID beta=0.5model performance: 91.7, 91.6, 92.1, 91.7No clear effect
use FedCut spectral heuristics on toy mimic collusion scenario S3FedCut vs Median · toy example 1000 runsByzantine tolerant rate: 37.7% -> 98.7%Improved

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