Securing IoT Networks with Adversarial Learning: A Defense Framework Against Cyber Threats

This paper responds to this increasing issue in the IoT domain by utilizing adversarial training methods to develop strong security solutions. Thus, to defend against the defense, we propose a defense framework that carry out Adversarial Training and Feature Squeeze to counteract adversarial attacks, FGSM, PGD. We validate the performances of these countermeasures for machine learning models (Decision Trees, SVM) and for a Deep Learning model (CNN) with a popular IoT-23 dataset. Results show that Adversarial training improves model robustness considerably when models are complex like CNN and SVM and they are brittle to these attacks. Such a dual-defensive model reduces risk and showcases pliability in handling a range of IoT provisioning environments. Our work highlights the opportunity for augmenting such defenses in practical IoT environments to improve resilience and security to dynamic and advanced cyber threats.