AI Paper Atlas

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

One Trap to Block Them All: Defending Encoder Stealing via Isotropic Uniformity

YITONG SHI, Kang Wei, Fushuo Huo, Shuchi Wu, Chuan Ma

ECCV 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Membership inference and auditing · 0 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
apply UniTrap defense (eps 0.3) against RDASSL encoder · CIFAR-10RDA stolen accuracy (%): 77.90 -> 35.26Improved
apply UniTrap defense against RDASSL encoder · SVHNRDA stolen accuracy (%): 73.85 -> 25.96Improved
apply UniTrap with ImageNet target encoderSSL encoder · SVHN (ImageNet target)RDA stolen accuracy (%): 60.03 -> 19.81Improved
apply UniTrap defenseSSL encoder · CIFAR-10target accuracy (%): 91.20 -> 90.18No clear effect
replace cosine distance-oriented perturbation with Gaussian-potential uniformity (ablation)UniTrap · CIFAR-10StolenEncoder stolen accuracy (%): 44.10 -> 36.27Improved
replace cosine distance-oriented perturbation with uniformity (ablation)UniTrap · STL-10StolenEncoder stolen accuracy (%): 47.67 -> 37.79Improved
replace cosine distance-oriented perturbation with uniformity (ablation)UniTrap · CIFAR-10RDA stolen accuracy (%): 40.35 -> 35.26Improved
replace cosine distance-oriented perturbation with uniformity (ablation)UniTrap · F-MNISTRDA stolen accuracy (%): 79.79 -> 81.20No clear effect
replace cosine distance-oriented perturbation with uniformity (ablation)UniTrap · CIFAR-10target accuracy (%): 90.19 -> 90.18No clear effect
replace cosine with uniformity against ContSteal (ablation)UniTrap · CIFAR-10ContSteal stolen accuracy (%): 37.38 -> 37.77No clear effect
UniTrap vs Top-k defense at similar stolen accuracyUniTrap · CIFAR-10target accuracy (%): 73.44 -> 90.18Improved
add Gaussian noise 0.5 as defense instead of UniTrapGaussian noise · CIFAR-10target accuracy (%): 90.18 -> 71.85Hurt

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