Physics-Aware Machine Unlearning for Cyber-Physical Systems

License: See LICENSE file · Updated 2026-09-29
My rating

This paper presents a physics-guided, gradient-ascent-based machine unlearning method for cyber-physical systems. The physical residual of the target system acts as a safety fence during unlearning, directing weight updates away from poisoned behavioral regions and toward physically feasible regions of the weight space.

Tags: