IEEE Wireless Communications Magazine deals with all technical and policy issues related to personalization, location-independent communications in all media. Papers highlight such topics as portable telephones, communicating palmtop computers, protocols, messaging, communications, and personalized traffic filtering. It also covers such policy issues as spectrum allocation, industry structure, and technology evolution.

Special Issue on Sustainable Big AI Model for Wireless Networks
Submission Date: 2023-10-01

With the continuous growth of mobile devices and rapid development of wireless communications, we are witnessing a vast amount of data transmitted over or generated by the fifth-generation (5G)-and-Beyond wireless networks. The advancements in machine learning are providing new approaches to explore these wireless data with certain artificial intelligence (AI) to accomplish a wide variety of large-scale, computation or communication-oriented tasks, from intelligent mobile edge computing to environment/object sensing and intelligent wireless communication. Traditional approaches require that each individual task has one specific AI model, which results in high HW/SW overheads and prevents the deep exploration of the inherent correlation within data and among tasks.

Recently, big AI model (or foundation model) has received a lot of attention, which is an emerging paradigm for building a unified machine learning system based on a generic class of AI models. As an example, the generative pre-trained transformer (GPT), has been successfully applied to natural language processing and many other computational tasks. Big AI model faces three interrelated crucial challenges: the large-scale model parameters, the large amount of training data, and the large computing power requirement for model training. However, due to the distributive nature of data and computing resources and for the security concerns, building big AI model over wireless networks requires a large number of wireless devices and edge servers properly coordinated by cloud centers to complete the joint training using their local data and distributed computing power. Moreover, the big AI model training process involves repeated and asynchronous downloading and uploading of high-dimensional (millions to billions) model parameters or their updates by tens to thousands of devices at a time. This will generate enormous data traffic and consume huge energy of both communication and computation procedures. The training problem cannot be efficiently solved using traditional wireless techniques targeting rate maximization and decoupled from learning. How to make the training and deployment of big AI model in a wireless network to be robust, efficient and sustainable, is a big question to be answered in the foreseeable future, which calls for a brand-new design of wireless techniques based on a communication-and-learning integration approach. Thereby, with the observation of the recent surge in relevant research, this Special Issue (SI) seeks to bring together researchers from both the academia and industry to introduce to the communication community the latest advancements in big AI models and point to readers many promising interdisciplinary research opportunities. The prospective topics of this special issue are listed below but are not limited to:

- Impact of big AI model on energy and network economics.
- Architecture design of big AI model suitable for future wireless networks.
- Wireless network protocol design for efficient deployment of big AI model.
- Distributed and power-efficient training of big AI model in wireless networks.
- Comprehensive performance evaluation of big AI model in wireless networks.
- Experiments, testbeds, and applications of big AI model over wireless networks.
- Explainability, privacy, and security related issues of big AI model in wireless networks.
- Big AI model for semantic communications over wireless networks.
- Big AI model for joint wireless sensing and communications.
- Big AI model for multi-task intelligent mobile edge computing.