The Internet of Things (IoT) has continued gaining in popularity and importance in everyday life in recent years. However, this development does not only present advantages. Indeed, due to the number of sensitive and private data produced by IoT systems, they have become the new privileged targets for cyberattackers. At the same time, Machine Learning (ML) has gained a phenomenal success in various fields like telecommunications, transport or cybersecurity. Nonetheless, the application of ML can cause significant damage when put in the hands of an attacker. Contrary to many previous works, we do not focus on the potential contributions of ML in the IoT security systems. Indeed, this survey aims to provide a comprehensive overview of ML approaches to enable more effective and less detectable attacks. Thereby, the purpose of this article is to identify and discuss the advantages of the elaboration of ML attacks and the possible solutions already evoked in the literature. Firstly, we provide an identification of the main threats and potential attacks on IoT networks. Then, we investigate on cyberattacks integrating machine learning algorithms during the last few years and we provide future research directions, especially for jamming, side channel, false data injection and adversarial machine learning attacks.

中文翻译：

---

机器学习如何改变物联网网络攻击的性质：一项调查

近年来，物联网 (IoT) 在日常生活中的受欢迎程度和重要性不断提高。然而，这种发展不仅具有优势。事实上，由于物联网系统产生了大量敏感和私人数据，它们已成为网络攻击者的新特权目标。与此同时，机器学习 (ML) 在电信、运输或网络安全等各个领域都取得了惊人的成功。尽管如此，机器学习的应用在落入攻击者手中时可能会造成重大损害。与许多以前的工作相反，我们不关注 ML 在物联网安全系统中的潜在贡献。事实上，本次调查旨在提供对 ML 方法的全面概述，以实现更有效且不易检测的攻击。从而，本文的目的是识别和讨论详细描述 ML 攻击的优势以及文献中已经提出的可能解决方案。首先，我们提供了对物联网网络的主要威胁和潜在攻击的识别。然后，我们调查了过去几年集成机器学习算法的网络攻击，并提供了未来的研究方向，特别是针对干扰、侧信道、虚假数据注入和对抗性机器学习攻击。