With the rise of various AIoT services, edge intelligence has emerged as a promising solution for implementing new applications in B5G/6G mobile communication systems. This lecture will provide an overview of the key features of B5G/6G mobile networks and focus on distributed AI learning driven by edge intelligence. Firstly, we will discuss the vulnerabilities associated with the edge intelligence framework. We will introduce model poisoning attacks, model inversion attacks triggered by malicious end devices, and an abnormal aggregator during the training phase. The former can compromise the model’s security and mislead the global model into producing incorrect inference results. While the latter leaks sensitive information by reversing the model weights to users' raw data. Then, we propose feasible defense approaches to combat privacy leaks and security threats mentioned above. Next, we will present state-of-the-art AI booster techniques, such as semi-supervised learning (SSL) and knowledge distillation, to optimize edge intelligence systems. We will introduce the latest SSL technique, Meta Pseudo Labels (MPL), which advocates a feedback strategy in the Teacher-Student architecture to mitigate performance degradation. Moreover, we present our proposed MPL-edge intelligence framework, which fully leverages unlabeled data to improve model generalization across various non-IID scenarios. Finally, we will examine different resource perspectives, including non-IID data, system heterogeneity, and model heterogeneity in edge intelligence environments. In real-world scenarios, limited and heterogeneous hardware resources among users threaten the overall training efficiency in edge intelligence. Existing research adopts model-heterogeneity frameworks to mitigate idle time on powerful devices, overlooking the possibility that powerful devices may negatively impact final results. Therefore, we propose extracting diverse features and balancing the contributions of strong and weak devices to mitigate straggler effects. In conclusion, we will explore potential future research directions in edge intelligence.

Dr. Te-Chuan Chiu is currently an assistant professor at the Department of Computer Science, National Tsing Hua University (NTHU), Taiwan. Before that, he has served as a postdoctoral research scholar at the Research Center for Information Technology Innovation (CITI), Academia Sinica, Taiwan, from 2018 to 2022. He has been a research scholar at the Department of Electrical and Computer Engineering, University of California, Davis (UCD), USA in 2022. He received the Ph.D. degree in Computer Science and Information Engineering from National Taiwan University (NTU), Taiwan. Recently, Dr. Chiu has cooperated with several industrial companies such as Delta, ITRI and NICS to realize edge AI in commercialized products. His research interests include B5G/6G communications, edge intelligence/AI, fog/edge computing, and AIoT.