Dormant Neural Trojans

TL;DR

This paper introduces a new type of neural network backdoor, called dormant Trojans, which remain inactive until triggered by a specific weight perturbation, making them harder to detect.

Contribution

The paper proposes a novel dormant Trojan methodology that activates via weight perturbation, evading current detection techniques.

Findings

Dormant Trojans can be effectively activated with specific weight perturbations.

Dormant Trojans evade detection by state-of-the-art methods.

The approach demonstrates high attack success rates in experiments.

Abstract

We present a novel methodology for neural network backdoor attacks. Unlike existing training-time attacks where the Trojaned network would respond to the Trojan trigger after training, our approach inserts a Trojan that will remain dormant until it is activated. The activation is realized through a specific perturbation to the network's weight parameters only known to the attacker. Our analysis and the experimental results demonstrate that dormant Trojaned networks can effectively evade detection by state-of-the-art backdoor detection methods.