Saved in:
Bibliographic Details
Main Authors: Hsu, Hsiang, Niroula, Pradeep, He, Zichang, Brugere, Ivan, Lecue, Freddy, Chen, Chun-Fu
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.22359
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917233296408576
author Hsu, Hsiang
Niroula, Pradeep
He, Zichang
Brugere, Ivan
Lecue, Freddy
Chen, Chun-Fu
author_facet Hsu, Hsiang
Niroula, Pradeep
He, Zichang
Brugere, Ivan
Lecue, Freddy
Chen, Chun-Fu
contents Machine unlearning offers a practical alternative to avoid full model re-training by approximately removing the influence of specific user data. While existing methods certify unlearning via statistical indistinguishability from re-trained models, these guarantees do not naturally extend to model outputs when inputs are adversarially perturbed. In particular, slight perturbations of forget samples may still be correctly recognized by the unlearned model - even when a re-trained model fails to do so - revealing a novel privacy risk: information about the forget samples may persist in their local neighborhood. In this work, we formalize this vulnerability as residual knowledge and show that it is inevitable in high-dimensional settings. To mitigate this risk, we propose a fine-tuning strategy, named RURK, that penalizes the model's ability to re-recognize perturbed forget samples. Experiments on vision benchmarks with deep neural networks demonstrate that residual knowledge is prevalent across existing unlearning methods and that our approach effectively prevents residual knowledge.
format Preprint
id arxiv_https___arxiv_org_abs_2601_22359
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples
Hsu, Hsiang
Niroula, Pradeep
He, Zichang
Brugere, Ivan
Lecue, Freddy
Chen, Chun-Fu
Machine Learning
Artificial Intelligence
Machine unlearning offers a practical alternative to avoid full model re-training by approximately removing the influence of specific user data. While existing methods certify unlearning via statistical indistinguishability from re-trained models, these guarantees do not naturally extend to model outputs when inputs are adversarially perturbed. In particular, slight perturbations of forget samples may still be correctly recognized by the unlearned model - even when a re-trained model fails to do so - revealing a novel privacy risk: information about the forget samples may persist in their local neighborhood. In this work, we formalize this vulnerability as residual knowledge and show that it is inevitable in high-dimensional settings. To mitigate this risk, we propose a fine-tuning strategy, named RURK, that penalizes the model's ability to re-recognize perturbed forget samples. Experiments on vision benchmarks with deep neural networks demonstrate that residual knowledge is prevalent across existing unlearning methods and that our approach effectively prevents residual knowledge.
title The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2601.22359