Label Smoothing Improves Machine Unlearning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Di, Zonglin, Zhu, Zhaowei, Jia, Jinghan, Liu, Jiancheng, Takhirov, Zafar, Jiang, Bo, Yao, Yuanshun, Liu, Sijia, Liu, Yang
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929382614892544
author Di, Zonglin
Zhu, Zhaowei
Jia, Jinghan
Liu, Jiancheng
Takhirov, Zafar
Jiang, Bo
Yao, Yuanshun
Liu, Sijia
Liu, Yang
author_facet Di, Zonglin
Zhu, Zhaowei
Jia, Jinghan
Liu, Jiancheng
Takhirov, Zafar
Jiang, Bo
Yao, Yuanshun
Liu, Sijia
Liu, Yang
contents The objective of machine unlearning (MU) is to eliminate previously learned data from a model. However, it is challenging to strike a balance between computation cost and performance when using existing MU techniques. Taking inspiration from the influence of label smoothing on model confidence and differential privacy, we propose a simple gradient-based MU approach that uses an inverse process of label smoothing. This work introduces UGradSL, a simple, plug-and-play MU approach that uses smoothed labels. We provide theoretical analyses demonstrating why properly introducing label smoothing improves MU performance. We conducted extensive experiments on six datasets of various sizes and different modalities, demonstrating the effectiveness and robustness of our proposed method. The consistent improvement in MU performance is only at a marginal cost of additional computations. For instance, UGradSL improves over the gradient ascent MU baseline by 66% unlearning accuracy without sacrificing unlearning efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2406_07698
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Label Smoothing Improves Machine Unlearning
Di, Zonglin
Zhu, Zhaowei
Jia, Jinghan
Liu, Jiancheng
Takhirov, Zafar
Jiang, Bo
Yao, Yuanshun
Liu, Sijia
Liu, Yang
Machine Learning
The objective of machine unlearning (MU) is to eliminate previously learned data from a model. However, it is challenging to strike a balance between computation cost and performance when using existing MU techniques. Taking inspiration from the influence of label smoothing on model confidence and differential privacy, we propose a simple gradient-based MU approach that uses an inverse process of label smoothing. This work introduces UGradSL, a simple, plug-and-play MU approach that uses smoothed labels. We provide theoretical analyses demonstrating why properly introducing label smoothing improves MU performance. We conducted extensive experiments on six datasets of various sizes and different modalities, demonstrating the effectiveness and robustness of our proposed method. The consistent improvement in MU performance is only at a marginal cost of additional computations. For instance, UGradSL improves over the gradient ascent MU baseline by 66% unlearning accuracy without sacrificing unlearning efficiency.
title Label Smoothing Improves Machine Unlearning
topic Machine Learning
url https://arxiv.org/abs/2406.07698