dorsal/arxiv
View SchemaA Robust Certified Machine Unlearning Method Under Distribution Shift
| Authors | Jinduo Guo, Yinzhi Cao |
|---|---|
| Categories | |
| ArXiv ID | 2601.06967vv1 |
| URL | https://arxiv.org/abs/2601.06967 |
| License | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
Abstract
The Newton method has been widely adopted to achieve certified unlearning. A critical assumption in existing approaches is that the data requested for unlearning are selected i.i.d.(independent and identically distributed). However,the problem of certified unlearning under non-i.i.d. deletions remains largely unexplored. In practice, unlearning requests are inherently biased, leading to non-i.i.d. deletions and causing distribution shifts between the original and retained datasets. In this paper, we show that certified unlearning with the Newton method becomes inefficient and ineffective under non-i.i.d. unlearning sets. We then propose a better certified unlearning approach by performing a distribution-aware certified unlearning framework based on iterative Newton updates constrained by a trust region. Our method provides a closer approximation to the retrained model and yields a tighter pre-run bound on the gradient residual, thereby ensuring efficient (epsilon, delta)-certified unlearning. To demonstrate its practical effectiveness under distribution shift, we also conduct extensive experiments across multiple evaluation metrics, providing a comprehensive assessment of our approach.
{
"annotation_id": "09120a85-310a-4af6-bbe5-3425b0fc4419",
"date_created": "2026-02-17T05:53:08.155000Z",
"date_modified": "2026-02-17T05:53:08.155000Z",
"file_hash": "ab365e67770bf99d8fd4024657726d3d6b7f955bd09b9f83b8d213694ac24a03",
"private": false,
"record": {
"abstract": "The Newton method has been widely adopted to achieve certified unlearning. A critical assumption in existing approaches is that the data requested for unlearning are selected i.i.d.(independent and identically distributed). However,the problem of certified unlearning under non-i.i.d. deletions remains largely unexplored. In practice, unlearning requests are inherently biased, leading to non-i.i.d. deletions and causing distribution shifts between the original and retained datasets. In this paper, we show that certified unlearning with the Newton method becomes inefficient and ineffective under non-i.i.d. unlearning sets. We then propose a better certified unlearning approach by performing a distribution-aware certified unlearning framework based on iterative Newton updates constrained by a trust region. Our method provides a closer approximation to the retrained model and yields a tighter pre-run bound on the gradient residual, thereby ensuring efficient (epsilon, delta)-certified unlearning. To demonstrate its practical effectiveness under distribution shift, we also conduct extensive experiments across multiple evaluation metrics, providing a comprehensive assessment of our approach.",
"arxiv_id": "2601.06967",
"authors": [
"Jinduo Guo",
"Yinzhi Cao"
],
"categories": [
"cs.LG",
"cs.CR"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"title": "A Robust Certified Machine Unlearning Method Under Distribution Shift",
"url": "https://arxiv.org/abs/2601.06967",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "42581d69-a5aa-4bf0-896a-57d5c94de41d",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}