A Durable Machine Unlearning Framework to Nullify Recall of Sensitive Data on Incremental Training

Cao, Qingqing; Hu, Liang; Liu, Dora D.; Miao, Jiaxing; Cao, Jian; Lai, Zhongyuan; Cao, Wei

doi:10.24963/ijcai.2026/715

A Durable Machine Unlearning Framework to Nullify Recall of Sensitive Data on Incremental Training

Qingqing Cao, Liang Hu, Dora D. Liu, Jiaxing Miao, Jian Cao, Zhongyuan Lai, Wei Cao

Download BibTeX Cite

Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence. AI4Tech: AI Enabling Technologies. Pages 6426-6435.

https://doi.org/10.24963/ijcai.2026/715

Keywords

Advanced AI4Tech: AI4Tech foundations
Advanced AI4Tech: Data-driven AI4Tech
Advanced AI4Tech: Deep AI4Tech

Abstract

The advancement of data privacy regulations has spurred the development of Machine Unlearning (MU), which is designed to remove the influence of sensitive data from a trained model and results in an unlearned model (ULM). Despite rapid progress in MU techniques, their vulnerabilities remain underexplored, which poses risks due to potential leakage of unlearned information. In realistic scenarios, ULMs always need to be incrementally trained with the newly collected data samples, which can lead to the consequences of recalling sensitive information if the new dataset contains similar or even the same unlearned samples. To address this issue, we devise a Durable Unlearning Enhancement (DUE) framework to avoid restoring unwanted sensitive information from incremental training data samples. The DUE framework has three key components that identify sensitive samples and suppress their gradients to update ULMs. Extensive experiments on state-of-the-art MU methods across multiple real-world datasets show that the proposed DUE framework can effectively nullify the recall of sensitive information after MU, and even improve the performance of ULMs. Consequently, our work establishes a new fundamental research direction in safe training against MU vulnerabilities.

Qingqing Cao, Liang Hu, Dora D. Liu, Jiaxing Miao, Jian Cao, Zhongyuan Lai, and Wei Cao. 2026. A Durable Machine Unlearning Framework to Nullify Recall of Sensitive Data on Incremental Training. In Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26. IJCAI, 6426–6435. https://doi.org/10.24963/ijcai.2026/715

@inproceedings{ijcai2026p715,
  title     = {A Durable Machine Unlearning Framework to Nullify Recall of Sensitive Data on Incremental Training},
  author    = {Cao, Qingqing and Hu, Liang and Liu, Dora D. and Miao, Jiaxing and Cao, Jian and Lai, Zhongyuan and Cao, Wei},
  booktitle = {Proceedings of the Thirty-Fifth International Joint Conference on
               Artificial Intelligence, {IJCAI-26}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Diego Calvanese},
  pages     = {6426--6435},
  year      = {2026},
  month     = {8},
  note      = {AI4Tech: AI Enabling Technologies},
  doi       = {10.24963/ijcai.2026/715},
  url       = {https://doi.org/10.24963/ijcai.2026/715},
}

TY  - CPAPER
TI  - A Durable Machine Unlearning Framework to Nullify Recall of Sensitive Data on Incremental Training
AU  - Cao, Qingqing
AU  - Hu, Liang
AU  - Liu, Dora D.
AU  - Miao, Jiaxing
AU  - Cao, Jian
AU  - Lai, Zhongyuan
AU  - Cao, Wei
T2  - Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence, IJCAI-26
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2026
SP  - 6426
EP  - 6435
DO  - 10.24963/ijcai.2026/715
UR  - https://doi.org/10.24963/ijcai.2026/715
ER  -