Saved in:
Bibliographic Details
Main Authors: Kim, Hyo Seo, Han, Dongyoon, Choe, Junsuk
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.05583
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911034358366208
author Kim, Hyo Seo
Han, Dongyoon
Choe, Junsuk
author_facet Kim, Hyo Seo
Han, Dongyoon
Choe, Junsuk
contents Machine unlearning aims to selectively remove specific knowledge from a trained model. Existing approaches, such as Task Arithmetic, fine-tune the model on the forget set to create a task vector (i.e., a direction in weight space) for subtraction from the original model's weight. However, their effectiveness is highly sensitive to hyperparameter selection, requiring extensive validation to identify the optimal vector from many fine-tuned candidates. In this paper, we propose a novel method that utilizes all fine-tuned models trained with varying hyperparameters instead of a single selection. Specifically, we aggregate the computed task vectors by retaining only the elements with consistent shared signs. The merged task vector is then negated to induce unlearning on the original model. Evaluations on zero-shot and standard image recognition tasks across twelve datasets and four backbone architectures show that our approach outperforms state-of-the-art methods while requiring similar or fewer computational resources. Code is available at https://github.com/naver-ai/negmerge.
format Preprint
id arxiv_https___arxiv_org_abs_2410_05583
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle NegMerge: Sign-Consensual Weight Merging for Machine Unlearning
Kim, Hyo Seo
Han, Dongyoon
Choe, Junsuk
Machine Learning
Artificial Intelligence
Machine unlearning aims to selectively remove specific knowledge from a trained model. Existing approaches, such as Task Arithmetic, fine-tune the model on the forget set to create a task vector (i.e., a direction in weight space) for subtraction from the original model's weight. However, their effectiveness is highly sensitive to hyperparameter selection, requiring extensive validation to identify the optimal vector from many fine-tuned candidates. In this paper, we propose a novel method that utilizes all fine-tuned models trained with varying hyperparameters instead of a single selection. Specifically, we aggregate the computed task vectors by retaining only the elements with consistent shared signs. The merged task vector is then negated to induce unlearning on the original model. Evaluations on zero-shot and standard image recognition tasks across twelve datasets and four backbone architectures show that our approach outperforms state-of-the-art methods while requiring similar or fewer computational resources. Code is available at https://github.com/naver-ai/negmerge.
title NegMerge: Sign-Consensual Weight Merging for Machine Unlearning
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2410.05583