EUVIMEDEUVIMED entdecken
Crossref

Certified Removal-Based Machine Unlearning for Data Poisoning in Cybermanufacturing

Xueling Tang, Wenmeng Tian

Journal of Manufacturing Science and Engineering · 01.10.2026 · 148 (10)

DOI: 10.1115/1.4072314 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Manufacturing Science and Engineering, online erschienen am 14.08.2026. Verfasst von Xueling Tang und 1 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Gefördert von: Directorate for Engineering.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Modern manufacturing systems heavily rely on process data and machine learning (ML) frameworks in process monitoring and quality control. Cyber-attacks (e.g., data poisoning) can maliciously modify these data, resulting in compromised performance of the ML models. Although there is extensive research on cyber-attack detections in manufacturing systems, very limited studies have focused on how to respond to the cyber-attacks to remove the impact of the identified poisoned data. Therefore, the certified removal (CR) framework for machine unlearning is proposed to remove the impact of the poisoned data in anomaly detection applications. For L2-regularized linear models, CR integrates randomized loss perturbation during training to mask small residuals as well as a fast one-step Newton update to remove the effect of the poisoned samples, providing provable removal guarantees without retraining from scratch. After unlearning, the resulting model is (ε, δ)-indistinguishable from a model trained from scratch on data that never contained the poisoned samples. A case study leveraging time-series data in rotary machine fault detection was used to evaluate the effectiveness of the proposed framework. Data poisoning was simulated by flipping labels in the training set. The CR framework was implemented to unlearn (or forget) a series of poisoned samples, and the results show strong capability of removing the impact of the forget set and consistent performance on the remaining set with significantly lower computation cost than retraining from scratch. This will result in significantly more timely responses to data poisoning attacks in cybermanufacturing systems.

Förderung

  • Directorate for Engineering

Zitieren

APA

Tang, X., & Tian, W. (2026). Certified Removal-Based Machine Unlearning for Data Poisoning in Cybermanufacturing. Journal of Manufacturing Science and Engineering, 148(10). https://doi.org/10.1115/1.4072314

Vancouver

Tang X, Tian W. Certified Removal-Based Machine Unlearning for Data Poisoning in Cybermanufacturing. Journal of Manufacturing Science and Engineering. 2026;148(10). doi:10.1115/1.4072314

BibTeX
@article{Tang2026,
  author = {Tang, Xueling and Tian, Wenmeng},
  title = {Certified Removal-Based Machine Unlearning for Data Poisoning in Cybermanufacturing},
  journal = {Journal of Manufacturing Science and Engineering},
  year = {2026},
  volume = {148},
  number = {10},
  doi = {10.1115/1.4072314}
}
Kostenlos starten →