AI Paper Atlas

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

Find a Scapegoat: Poisoning Membership Inference Attack and Defense to Federated Learning

Wenjin Mo, Zhiyuan Li, Minghong Fang, Mingwei Fang

ICCV 2025 · 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 greedy mask sample selection instead of random selection (ablation)FedPoisonMIA · Texas100attack accuracy: 0.757 -> 0.894 (DP, Non-IID); improvement of over 20%Improved
add greedy mask sample selection instead of random selection (ablation)FedPoisonMIA · Texas100 (Trimmed-mean, Non-IID)attack accuracy: 0.748 -> 0.890Improved
use FedPoisonMIA instead of best baseline attackFedPoisonMIA · Texas100 (Median, Non-IID)attack accuracy: 0.770 -> 0.930Improved
use FedPoisonMIA instead of best baseline attackFedPoisonMIA · Texas100 (ATM, Non-IID)attack accuracy: 0.773 -> 0.853Improved
defend with ATM instead of FedAvgATM · Texas100 Non-IID (FedPoisonMIA)attack accuracy: 0.897 -> 0.853Improved
defend with ATM instead of MedianATM · Texas100 Non-IID (FedPoisonMIA)attack accuracy: 0.930 -> 0.853Improved
defend with ATM instead of FedAvgATM · CIFAR-10 Non-IID (FedPoisonMIA)attack accuracy: 0.786 -> 0.650Improved
defend with ATM instead of MedianATM · FER2013 IID (FedPoisonMIA)attack accuracy: 0.816 -> 0.786Hurt
use partial-knowledge instead of full-knowledge attackFedPoisonMIA · all datasetsattack accuracy: maximum degradation of 8.0%No clear effect
increase fraction of malicious clients from 10% to 30% (ablation)FedPoisonMIA · Texas100attack accuracy: nearly 100% at 30% malicious versus baselines around 90%Improved
increase the total number of clients (ablation)FedPoisonMIA · Texas100attack accuracy: 8 to 30 clients, accuracy decreases overallHurt
increase the number of attack samples from 100 to 500 (ablation)FedPoisonMIA · Texas100attack accuracy: 100 to 500 samples, accuracy gradually decreasesHurt

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