AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

Label Smoothing Improves Machine Unlearning

Zonglin Di, Zhaowei Zhu, Jinghan Jia, Jiancheng Liu, Zafar Takhirov, Bo Jiang, Yuanshun Yao, Sijia Liu, Yang Liu

ICLR 2026 · Federated learning and privacy › Differential privacy · approach: Private training with public 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 negative label smoothing to gradient ascent (UGradSL)gradient ascent · CIFAR-10 class-wiseUA: 25.19 -> 94.99Improved
add gradient-mixed plus NLS to fine-tuning (UGradSL+)fine-tuning · CIFAR-100 random forgettingUA: 2.55 -> 63.89Improved
add NLS to gradient-mixed UGradSL (ablation)UGradSL · CIFAR-10 random forgettingUA: 0.58 -> 20.77Improved
add gradient-mixed to GA with NLS (ablation)UGradSL · CIFAR-10 random forgettingRA / TA: RA 76.41 -> 79.83, TA 70.15 -> 73.94Improved
use negative instead of positive label smoothing (ablation)UGradSL · CIFAR-10 random forgettingUA: 2.02 -> 20.77Improved
use negative instead of positive label smoothing (ablation)UGradSL+ · CIFAR-10 random forgettingUA: 10.81 -> 25.13Improved
add NLS to UGradSL+ gradient-mixed (ablation)UGradSL+ · CIFAR-10 random forgettingUA: 14.12 -> 25.13Improved
increase forgetting strength in random forgettingUGradSL+ · CIFAR-100 random forgettingTA: 70.51 (retrain) vs 61.09Hurt
tune UGradSL+ toward retrain gap objective (R)UGradSL+ · CIFAR-100 random forgettingAvg. Gap: 17.40 -> 3.75Improved
use UGradSL+ vs GA at scaleUGradSL+ · ImageNet class-wiseUA: 81.23 -> 100.00Improved
use UGradSL (cheap) vs FT-based UGradSL+UGradSL · CIFAR-10 class-wiseRTE (min): 3.07 -> 0.22Improved
apply UGradSL+ to 20 Newsgroup textUGradSL+ · 20 Newsgroup random forgettingUA: 2.26 -> 35.53Improved

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