On-Device Intelligent Computing for 5G Mobile Ultra High Definition Video Applications: Techniques and Solutions

Overview: With the vigorous development of the mobile internet, innovative breakthroughs in video codec technologies, and the popularization of intellectual mobile terminals, the requirement for content-rich mobile video applications is constantly increasing, and the consumption of mobile video is continuing to climb rapidly. Mobile video is transferring from standard definition and high definition to ultra high definition (UHD), especially under the power of fifth-generation mobile communication technology (5G). In the future, more and more UHD videos applications, including UHD video, AR/VR, cloud games, and high resolution 3D holographic images, will “enter” mobile devices. Mobile UHD video services will become the main battlefield of mobile traffic in the future. The application scenarios, service quality, and industrial development of mobile UHD video services are depended on the intelligent computing power. The intelligent computing power is gradually expanded from the on-cloud (center) to the on-edge and the on-device with the deepening of the ubiquitous degree of the intelligent computing power. In particular, intelligent data processing (such as feature calculation, model inference, and model training) on mobile terminal devices has become a new trend with the significant improvement of the intelligence and the intelligent computing power of mobile terminal devices. As a powerful supplement to “on-cloud intelligent computing”, “on-device intelligent computing” has many advantages such as low latency, data privacy protection, and cloud computing resources conservation. It provides the on-device intelligent computing support for improving the service quality of mobile UHD video applications. Meanwhile, on-device intelligent computing also has many challenges, such as (i) mobile terminal devices have limited resources and vary greatly in performance. When the inference and the training of complex models occur on large-scale terminal devices, the resource limitation and the heterogeneity of devices need to be considered; (ii) High data heterogeneity: data is distributed on massive terminal devices, and this can bring problems such as uneven data distribution, quite differences in data characteristics, and irregular data availability; (iii) Low reliability of terminal devices: different from the characteristics of cloud servers that are completely controlled by the data center and have high reliability, the terminal devices have low controllability and poor reliability, and are vulnerable to attacks and become malicious nodes. The aim of this Special Issue is to collate original research and review articles discussing the state of the art in the field. We hope that this Special Issue will provide a forum for researchers and practitioners in all aspects of mobile multimedia technologies.