Edge Learning for 6G-Enabled Internet of Things: A Comprehensive Survey of Vulnerabilities, Datasets, and Defenses

The deployment of the fifth-generation (5G) wireless networks in Internet of Everything (IoE) applications and future networks (e.g., sixth-generation (6G) networks) has raised a number of operational challenges and limitations, for example in terms of security and privacy. Edge learning is an emerging approach to training models across distributed clients while ensuring data privacy. Such an approach when integrated in future network infrastructures (e.g., 6G) can potentially solve challenging problems such as resource management and behavior prediction. However, edge learning (including distributed deep learning) are known to be susceptible to tampering and manipulation. This survey article provides a holistic review of the extant literature focusing on edge learning-related vulnerabilities and defenses for 6G-enabled Internet of Things (IoT) systems. Existing machine learning approaches for 6G–IoT security and machine learning-associated threats are broadly categorized based on learning modes, namely: centralized, federated, and distributed. Then, we provide an overview of enabling emerging technologies for 6G–IoT intelligence. We also provide a holistic survey of existing research on attacks against machine learning and classify threat models into eight categories, namely: backdoor attacks, adversarial examples, combined attacks, poisoning attacks, Sybil attacks, byzantine attacks, inference attacks, and dropping attacks. In addition, we provide a comprehensive and detailed taxonomy and a comparative summary of the state-of-the-art defense methods against edge learning-related vulnerabilities. Finally, as new attacks and defense technologies are realized, new research and future overall prospects for 6G-enabled IoT are discussed.

I. Introduction

We are experiencing accelerated development, increasing adoption, and innovative combinations of information and communication technologies, such as cloud and edge computing, the Internet of Things (IoT), massive data analytics, and Artificial Intelligence (AI). This widespread endorsement is driven by many factors, including the widespread availability of broadband communications, which are expected to move the world into an all-connected zone. One instance of such a combination is the integration of AI into the fifth-generation (5G) wireless networks. However, it is only intended to operate in specific areas under specific conditions (massive data and robust computing) [1], [2]. This alliance is expected to be much tighter in future generations, starting with the upcoming sixth-generation (6G) wireless networks. AI is expected to be a core component in it [3], [4]. In addition, Multi-Access Edge Computing (MEC) provides the possibility of processing large volumes of data by edge devices and distributed learning paradigms such as Federated Learning (FL) which enable multiple parties to collaborate in building shared Machine Learning models without sharing their data. There is a lot of interest in making the edge intelligent, an emerging research area known as distributed edge learning [5], [5], [6].