AI Paper Atlas

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

Stealing Split Learning Bottom Models by Recovering Embedding Geometry

Qinbo Zhang, Yanhang Shi, Ziyi Zhang, Hao Wang, Sai Qian Zhang, Jian Li

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 neighbor matching in the contrastive space (ablation)VENOM · CIFAR-10, Model RakeS-ACC (%): 24.72 -> 52.58Improved
use contrastive space instead of raw embedding space for neighbor mining (ablation)VENOM · CIFAR-10, Model RakeS-ACC (%): 32.03 -> 52.58Improved
add the KL feature-shape term (ablation)VENOM · CIFAR-10, Model RakeS-ACC (%): 46.43 -> 52.58Improved
add the KL feature-shape term (ablation)VENOM · CIFAR-10, Vanilla (no defense)S-ACC (%): 68.24 -> 69.97Improved
increase neighbor set size k from 1% to 10% of auxiliary set (ablation)VENOM · CIFAR-10S-ACC and AGRImproved
increase neighbor size from 10% to 15% (ablation)VENOM · CIFAR-10S-ACC and AGRNo clear effect
set attraction-repulsion weight alpha to 0.5 instead of skewed values (ablation)VENOM · CIFAR-10S-ACC and AGRImproved
increase auxiliary data from 1% to 10% of dataset (ablation)VENOM · CIFAR-10S-ACC and AGRImproved
use near-OOD auxiliary data (CIFAR-100, Tiny-ImageNet) instead of in-distribution (ablation)VENOM · CIFAR-10S-ACC and AGRHurt
use far-OOD auxiliary data (MNIST) (ablation)VENOM · CIFAR-10S-ACC and AGRHurt
deepen surrogate beyond the target capacity (3 blocks) (ablation)VENOM · CIFAR-10S-ACC and AGRNo clear effect
use approximate subset-based neighbor selection instead of exact search (ablation)VENOM · SUSY, Model RakeS-ACC (%): 68.81 -> 63.16Hurt

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