AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

LMCleaner: Efficient and Certified Online Unlearning via Influence Propagation Truncation

Jie Xu, Zihan Wu, Wenbo Pan, Jiao Yin, Yong-Feng Ge, Hua Wang, Cong Wang, Xiaohua Jia

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws

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
increase truncation window K from 50 to 1000 (ablation)LMCleaner · TOFU Epoch 5, LLaMA-3.2-1Butility, forget truth, MIA: utility 0.429 -> 0.428; 20x costNo clear effect
increase truncation window K to 1250 (ablation)LMCleaner · TOFU Epoch 5, LLaMA-3.2-1BMIA: 0.516 -> 0.599Hurt
increase mini-batch size B from 8 to 256 (ablation)LMCleaner · TOFU, LLaMA-3.2-1BFG Prob / utility: FG Prob 0.282 -> 0.201; utility 0.388 -> 0.349Mixed
increase batch size B from 8 to 64 (ablation)LMCleaner · TOFU, LLaMA-3.2-1BMIA: 0.575 -> 0.509Improved
treat mini-batch as atomic unit instead of per-sample gradientsLMCleaner · storagestorage reduction: 512x reductionImproved
use online unlearning during training instead of fine-tuning-based unlearningLMCleaner vs GradDiff · TOFU Epoch 5, LLaMA-3.2-1Bmodel utility: 0.270 -> 0.429Improved
LMCleaner vs PDU on utility at 3BLLaMA-3.2-3B · TOFU Epoch 1model utility: 0.439 (PDU) vs 0.431No clear effect
use LMCleaner for membership inference resistanceLLaMA-3.2-3B · TOFU Epoch 1MIA min-k: 0.287 (PDU) -> 0.332Improved
use LMCleaner instead of NPO over epochsGPT-2 · TOFU Epoch 4utility stabilityImproved

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