Probing then Editing: A Push-Pull Framework for Retain-Free Machine Unlearning in Industrial IoT

The introduction of the Probing then Editing (PTE) framework marks a significant advancement in machine unlearning, particularly within dynamic Industrial Internet of Things (IIoT) environments. Traditional methods often depend on retain data, which not only increases computational and energy costs but also poses challenges regarding data privacy and compliance. PTE innovatively reframes unlearning as a probe-edit process, allowing models to selectively forget outdated or erroneous knowledge. By probing the decision boundary of the model and generating editing instructions based on its own predictions, PTE effectively dismantles the decision region of the target class while preserving the integrity of retained classes through masked knowledge distillation. Experimental results indicate that PTE achieves a commendable balance between unlearning effectiveness and model utility, showcasing its potential to enhance operational efficiency and privacy compliance in industrial applications.