iPET: Privacy Enhancing Traffic Perturbations for Secure IoT Communications

IoT devices constantly communicate with servers over the Internet, allowing an attacker to extract sensitive information by passively monitoring network traffic. Recent research works have shown that a trained machine learning (ML) model can accurately fingerprint learned from (encrypted) traffic flows of devices. Such fingerprinting attacks are capable revealing make and devices, which further be used detailed user activities. In this work, we develop propose iPET, novel adversarial perturbation-based modification system defends against attacks. iPET design employs GAN (Generative Adversarial Networks) in tuneable way, users specify maximum bandwidth overhead they willing tolerate for defense. A fundamental idea is deliberately introduce stochasticity between instances. This approach limits counter attack, as it inhibits recreating identical perturbation using fingerprinting. We evaluate effectiveness our defense state-of-the-art models three different capabilities. Our evaluations on synthetic real-world datasets demonstrate decreases accuracy even potent attackers. also show perturbations generated generalize well schemes may deploy.