Machine learning backdoors have the property that the machine learning model should work as expected on normal inputs, but when the input contains a specific trigger, it behaves as the attacker desires. Detecting such triggers has been proven to be extremely difficult. In this paper, we present a novel and explainable approach to detect and eliminate such backdoor triggers based on active paths found in neural networks. We present promising experimental evidence of our approach, which involves injecting backdoors into a machine learning model used for intrusion detection.
@article{arxiv.2603.10641,
title = {Detecting and Eliminating Neural Network Backdoors Through Active Paths with Application to Intrusion Detection},
author = {Eirik Høyheim and Magnus Wiik Eckhoff and Gudmund Grov and Robert Flood and David Aspinall},
journal= {arXiv preprint arXiv:2603.10641},
year = {2026}
}