Saved in:
Bibliographic Details
Main Authors: Guo, Feng, Wen, Yuntao, Gao, Shen, Zhang, Junshuo, Shang, Shuo
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.11667
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909919220858880
author Guo, Feng
Wen, Yuntao
Gao, Shen
Zhang, Junshuo
Shang, Shuo
author_facet Guo, Feng
Wen, Yuntao
Gao, Shen
Zhang, Junshuo
Shang, Shuo
contents Machine unlearning, which selectively removes harmful knowledge from a pre-trained model without retraining from scratch, is crucial for addressing privacy, regulatory compliance, and ethical concerns in Large Language Models (LLMs). However, existing unlearning methods often struggle to thoroughly remove harmful knowledge, leaving residual harmful knowledge that can be easily recovered. To address these limitations, we propose Knowledge Density-Guided Unlearning via Blocks Reinsertion (KUnBR), a novel approach that first identifies layers with rich harmful knowledge and then thoroughly eliminates the harmful knowledge via re-insertion strategy. Our method introduces knowledge density estimation to quantify and locate layers containing the most harmful knowledge, enabling precise unlearning. Additionally, we design a layer re-insertion strategy that extracts and re-inserts harmful knowledge-rich layers into the original LLM, bypassing gradient obstruction caused by cover layers and ensuring effective gradient propagation during unlearning. Extensive experiments conducted on several unlearning and general capability benchmarks demonstrate that KUnBR achieves state-of-the-art forgetting performance while maintaining model utility.
format Preprint
id arxiv_https___arxiv_org_abs_2511_11667
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Superficial Forgetting: Thorough Unlearning through Knowledge Density Estimation and Block Re-insertion
Guo, Feng
Wen, Yuntao
Gao, Shen
Zhang, Junshuo
Shang, Shuo
Machine Learning
Artificial Intelligence
Machine unlearning, which selectively removes harmful knowledge from a pre-trained model without retraining from scratch, is crucial for addressing privacy, regulatory compliance, and ethical concerns in Large Language Models (LLMs). However, existing unlearning methods often struggle to thoroughly remove harmful knowledge, leaving residual harmful knowledge that can be easily recovered. To address these limitations, we propose Knowledge Density-Guided Unlearning via Blocks Reinsertion (KUnBR), a novel approach that first identifies layers with rich harmful knowledge and then thoroughly eliminates the harmful knowledge via re-insertion strategy. Our method introduces knowledge density estimation to quantify and locate layers containing the most harmful knowledge, enabling precise unlearning. Additionally, we design a layer re-insertion strategy that extracts and re-inserts harmful knowledge-rich layers into the original LLM, bypassing gradient obstruction caused by cover layers and ensuring effective gradient propagation during unlearning. Extensive experiments conducted on several unlearning and general capability benchmarks demonstrate that KUnBR achieves state-of-the-art forgetting performance while maintaining model utility.
title Beyond Superficial Forgetting: Thorough Unlearning through Knowledge Density Estimation and Block Re-insertion
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.11667