AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

HiddenEcho: Mitigating Noise Amplification in Differentially Private LLMs with Hidden-State Correction

Wenhao Li, Kunhao Li, Lei Yang

ICLR 2026 · Federated learning and privacy › Differential privacy · approach: Differentially private synthetic data

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 residual connections (ablation)HiddenEcho · BBC News, eta=100AUC: 0.659 -> 0.732Improved
add hidden layer filter instead of fixed skip layers (ablation)HiddenEcho · BBC News, eta=100AUC: 0.629 -> 0.732Improved
add information-bottleneck dimension reducer instead of linear layer (ablation)HiddenEcho · Financial Phrasebank, eta=100AUC: 0.789 -> 0.857Improved
add hidden-state denoising correction over LDPLDP · Financial Phrasebank, eta=100AUC: 0.596 -> 0.857Improved
use HiddenEcho instead of SnD pretrained denoiserSnD · BBC News, eta=6000AUC: 0.637 -> 0.951Improved
use layer-filtered HiddenEcho instead of HiddenEcho-Full (ablation)HiddenEcho-Full · BBC News, eta=100AUC: 0.685 -> 0.732Improved
use layer-filtered HiddenEcho instead of HiddenEcho-Full (ablation)HiddenEcho-Full · Financial Phrasebank, eta=100AUC: 0.875 -> 0.857Hurt
add filtered hidden layers to reduce communication (ablation)HiddenEcho-Full · classificationcommunication cost: 2.63 -> 0.38 MiB (85.55% saved)Improved
add filtered hidden layers on generation tasks (ablation)HiddenEcho-Full · IWSLTcommunication cost: 6.00 -> 2.25 MiB (62.50% saved)Improved
use HiddenEcho-Full instead of SnD for training speedSnD · BBC Newstime per epoch: 393 -> 118 s (72.52% reduction)Improved
use HiddenEcho client correction at the final layerLDP · hidden-state MSE, eta=100MSE: 14.69 -> 8.31Improved
train HiddenEcho-Full for more epochsHiddenEcho-Full · BBC Newseval lossHurt

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