AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

PE-SGD: Differentially Private Deep Learning via Evolution of Gradient Subspace for Text

Tianyuan Zou, Zinan Lin, Sivakanth Gopi, Yang Liu, Ya-Qin Zhang, Robert Sim, Xin Deng, Sergey Yekhanin

ICLR 2026 · Federated learning and privacy › Differential privacy · approach: DP-SGD refinements

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
replace DP-SGD with PE-SGD at eps=1Qwen2.5-3B-Instruct · Congressional Speechnext-token loss: 3.1080 -> 2.7532Improved
replace DP-SGD with PE-SGD at eps=1Qwen2.5-3B-Instruct · PubMednext-token loss: 2.3731 -> 2.1990Improved
evolve synthetic projection samples during training (ablation)PE-SGD · Congressional Speech (eps=1)next-token loss: 2.8024 -> 2.7532Improved
evolve synthetic projection samples during training (ablation)PE-SGD · PubMed (eps=1)next-token loss: 2.2497 -> 2.1990Improved
use PE-SGD without privacy noiseQwen2.5-3B-Instruct · Congressional Speech (eps=inf)next-token loss: 2.6868 -> 2.7468Hurt
add noise to final projection coefficients rather than inner products (ablation)PE-SGD · Qwen2.5 (Fig. 8)loss: STD/Mean 0.533 -> 0.404Improved
add noise to coefficients rather than to private gradient (ablation)PE-SGD · Qwen2.5 (Fig. 8)loss and accuracy: lower loss, slightly higher accuracyImproved
use whole synthetic gradient subspace rather than top-k eigenspacePDP-SGD/GEP · Congressional Speech (eps=1)next-token loss: 2.8025 -> 2.7532Improved
generate synthetic data with GPT-4o-mini instead of Qwen2.5-3BPE-SGD · Congressional Speech (eps=1)next-token loss: 2.7532 -> 2.7219Improved
add synthetic-data SFT before DP training (ablation)PE-SGD · Congressional Speech (eps=1)next-token loss: 2.7740 -> 2.7532Improved
use extreme API fold L=1 or L=infinity (ablation)PE-SGD · Congressional SpeechlossHurt
change prompt templates (ablation)PE-SGD · Congressional Speechnext-token loss: 2.7532 -> 2.7526No clear effect

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