AI Paper Atlas

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

FedMOP: Achieving Enhanced Privacy and Performance in Federated Learning via Momentum Orthogonal Projection

Yunlong Zhao, Xiaoheng Deng, Hongyan Xu, Zhuohua Qiu, Xiaowen Hu, Shan You, Yi Chen, Chang Xu, Xiu Su

CVPR 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Gradient inversion attacks

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 gradient orthogonal projection offset to FedAvg (ablation)FedAvg · CIFAR-10 (ablation)accuracy (%): 68.3 -> 70.4Improved
add random scaling beta without momentum (ablation)orthogonal offset · CIFAR-10 (ablation)attack MSE: 0.08 -> 0.31Improved
add momentum-based trajectory hiding (ablation)orthogonal offset with random beta · CIFAR-10 (ablation)attack MSE: 0.31 -> 0.85Improved
add momentum-based trajectory hiding (ablation)orthogonal offset with random beta · CIFAR-10 (ablation)accuracy (%): 70.2 -> 72.7Improved
increase sigma_gamma from 0.02 to 0.05 (ablation)FedMOP · CIFAR-10 (ablation)accuracy and attack MSE: accuracy 72.7 -> 71.5; MSE 0.85 -> 0.93Mixed
increase sigma_beta from 0.10 to 0.15 (ablation)FedMOP · CIFAR-10 (ablation)accuracy and attack MSE: accuracy 72.7 -> 71.3; MSE 0.85 -> 0.96Mixed
vary initial offset variance sigma_0 by 10x (ablation)FedMOP · CIFAR-10 (ablation)accuracy: 72.6 / 72.7 / 72.2No clear effect
vary mean momentum coefficient gamma from 0.85 to 0.95 (ablation)FedMOP · CIFAR-10 (ablation)accuracy: 72.3 / 72.7 / 72.1No clear effect
use FedMOP instead of FedAvgFedAvg · CIFAR-100 (full participation, D1)top-1 accuracy (%): 31.61 -> 42.07Improved
use FedMOP instead of HierFed under 20% participation, D1convergence speed · CIFAR-100 rounds to 30% accuracycommunication rounds: 48 -> 64Hurt
use FedMOP instead of Soteria/DP-style defenses against csDLGgradient leakage defense · csDLG, batch size 1attack MSE: 0.46 -> 0.85 (vs OUTPOST)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