AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

Towards Natural Machine Unlearning

Zhengbao He, Tao Li, Xinwen Cheng, Zhehao Huang, Xiaolin Huang

TPAMI 2025 · Continual learning › Continual fine-tuning of LLMs · approach: Synthetic data and merging for CLIP · 13 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
use NatMU (inject remaining-data information into forgetting samples) instead of best baselinerelabeling-based unlearning · Full-class Cattle, ideal, CIFAR-100/20 ResNet-18Avg.Gap: 0.42 -> 0.11Improved
use NatMU instead of best baselinerelabeling-based unlearning · Full-class Sea, practical hyperparametersAvg.Gap: 1.02 -> 0.35Improved
use NatMU instead of SCRUBSCRUB · Sub-class Rocket, ideal hyperparametersAvg.Gap: 1.29 -> 1.74Hurt
use NatMU instead of best baselineunlearning baselines · Sub-class Cattle, ideal hyperparametersAvg.Gap: 3.28 -> 2.22Improved
use NatMU instead of best baselineunlearning baselines · Sub-class Sea, practical hyperparametersAvg.Gap: 4.40 -> 3.63Improved
use NatMU instead of NegGrad+NegGrad+ · CIFAR-10 VGG16-BN, 1% sample-wiseAvg.Gap: 2.18 -> 1.55Improved
use NatMU instead of SCRUBSCRUB · CIFAR-100 ResNet-18, 10% sample-wiseAvg.Gap: 2.28 -> 1.77Improved
use NatMU instead of AMUNAMUN · TinyImageNet ResNet-34, 1% sample-wiseAvg.Gap: 4.18 -> 2.18Improved
switch from ideal to practical (transferred) hyperparametersNatMU · Class-wise unlearningAvg.GapNo clear effect
switch from ideal to practical hyperparametersrelabeling baselines such as delta-targeted · Sub-class unlearningAvg.Gap: 6.47 -> 17.56Hurt
inject correct remaining-data information into forgetting samples (ablation)relabeling-based unlearning · Forgetting accuracy curves, CIFAR-100 ResNet-18accuracy on (x^f*m, y^r) pairsImproved

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