Monday 13 September, 15:30-18:00 (UTC/GMT +3)
With the massive data and computing resources, Artificial Intelligence/Machine Learning has made rapid progress in terms of deep learning, which have exceeded human performance and deployed in many areas in daily life, including computer vision, natural language processing, smart decision making, and so on. The next generation wireless networks, i.e., 6G, will open a new era of “Internet of intelligence”, which uses the wireless network to support and enable AI services for the connected people and things. Native AI support will be the key of the system design, where AI will become a native feature of 6G with learning and inferring capabilities embedded everywhere in the network. The network architecture and transmission technologies need to be revisited, because the communication system will not just deliver the bits reliably, but more importantly provide the intelligent processing of data via the connectivity and computing resources in the devices, the edge, and the cloud in the network.
In other aspect, with the introduction of various types of techniques and service requirements, the communication systems become more and more complex. How to resort to AI for efficient communications is the promising solution. Beyond the popular methods in current research where AI is used to replace the modules in the communication system in a static manner, we envision that the network nodes should have the ability of continuously learning from and adapting to the environment intelligently, to use the spectrum more efficiently. Especially, the air-interface could be automated through training via the collected data from various aspects, enabling joint optimization of the transmitter and receiver capturing the specific environment, the hardware impairments, and the application requirements. While, the research is still at very early stage, where the power of DL is not fully exploited to solve the hard problems in communications, e.g., coding, multi-terminal information theory and transmissions. As the essential part for data-driven AI, the standard Wireless realistic Dataset is an emerging issue, to evaluate and compare the performance of various algorithms. The wireless AI has also its only specific requirements, e.g., labeled data not easy or impossible to get, efficient training or online learning method for Dynamic channel or environment, and Scalable/Flexible neural architecture to realize flexible wireless resource utilization, Complexity and processing latency issue especially for the baseband processing.
As the interdisciplinary of wireless communications and artificial intelligence, many fundamental problems remain open. This workshop aims at bringing together academic and industrial researchers in an effort to identify and discuss the major technical challenges, recent breakthroughs, and new applications related to native-AI wireless networks. The potential topics include, but are not limited to:
- Concept, architectures, theories, and applications of Native-AI wireless network
- Collaborative Distributed Learning (training/inferring) over wireless networks: Federated learning, split learning, in-device learning
- Theory for communication and AI: Information theory about tradeoff between capacity, computation, complexity, and machine learning communications systems
- Source coding for Native-AI wireless network
- Transmission for Native-AI wireless network
- Radio resource management for Native-AI wireless network
- Multi-agent reinforcement learning with Native-AI wireless network
- Privacy and security issues
- End-to-end communications empowered by AI: Source/channel coding, modulation, waveform, MIMO
- Semantic communications empowered by AI
- Feedback and Control empowered by AI
- Radio environment sensing empowered by AI
- Advanced AI for wireless: Neural network architecture and training, Scalable neural networks, Online training
- Efficient Hardware implementations of neural networks in communications
- Wireless Datasets
- Prototyping, demo, test-beds, and field trials
Workshop chairs / organizers
- Prof. Zhaoyang Zhang, Zhejiang University, China, firstname.lastname@example.org
- Dr. Wen Tong, Huawei Technologies Co., Ltd., Canada, email@example.com
- Prof. Geoffrey Ye Li, Imperial College London, UK, firstname.lastname@example.org
Technical Program Committee Chairs
- Prof. Caijun Zhong, Zhejiang University, China, email@example.com
- Prof. Tony Quek, Singapore University of Technology and Design, Singapore, firstname.lastname@example.org
Keynote 1: “Deep Learning for Physical Layer Communications: An Attempt towards 6G”
Monday 13 September, 15:35-16:10 (UTC/GMT +3)
Feifei Gao, Associate Professor, IEEE Fellow, Department of Automation, Tsinghua University, China.
Merging artificial intelligence in to the system design has appeared as a new trend in wireless communications areas and has been deemed as one of the 6G technologies. In this talk, we will present how to apply the deep neural network (DNN) for various aspects of physical layer communications design, including the channel estimation, channel prediction, channel feedback, data detection, and beamforming, etc. We will also present a promising new approach that is driven by both the communications data and the communication models. It will be seen that the DNN can be used to enhance the performance of the existing technologies once there is model mismatch. More interestingly, we will show that applying DNN can deal with the conventionally unsolvable problems, thanks to the universal approximation capability of DNN. With the well-defined propagation model in communication areas, we also attempt to explain the DNN under the scenario of channel estimation and reach a strong conclusion that DNN can always provide the asymptotically optimal channel estimations. We have also build test-bed to show the effectiveness of the AI aided wireless communications. In all, DNN is shown to be a very powerful tool for communications and would make the communications protocols more intelligently. Nevertheless, as a new born stuff, one should carefully select suitable scenarios for applying DNN rather than simply spreading it everywhere.
Prof. Gao’s research interest include signal processing for communications, array signal processing, convex optimizations, and artificial intelligence assisted communications. He has authored/ coauthored more than 150 refereed IEEE journal papers and more than 150 IEEE conference proceeding papers that are cited more than 10000 times in Google Scholar. Prof. Gao has served as an Editor of IEEE Transactions on Wireless Communications, IEEE Journal of Selected Topics in Signal Processing (Lead Guest Editor), IEEE Transactions on Cognitive Communications and Networking, IEEE Signal Processing Letters, IEEE Communications Letters, IEEE Wireless Communications Letters, and China Communications. He has also serves as the symposium co-chair for 2019 IEEE Conference on Communications (ICC), 2018 IEEE Vehicular Technology Conference Spring (VTC), 2015 IEEE Conference on Communications (ICC), 2014 IEEE Global Communications Conference (GLOBECOM), 2014 IEEE Vehicular Technology Conference Fall (VTC), as well as Technical Committee Members for more than 50 IEEE conferences.
Keynote 2: “How can a classic communication algorithm help deep learning – integrate message-passing-algorithm into a deep neural network”
Monday 13 September, 16:23-16:58 (UTC/GMT +3)
Mr. Yiqun GE, Huawei Technology Canada
A lot of effort has been made to use deep neural network (autoencoder) to learn a wireless transceiver. A wireless channel is time-varying, its noise and other hostility are random, and offsets are always there between true channel and estimated one. Moreover, a modern wireless system usually exploits multi-dimensional channel, such as multiple antennas, multiple codes and so on, to improve its radio efficiency. Changes and properties on various channel dimensions may be very different from each other over the time. Although a DNN-based auto-encoder could train a transceiver to fit into a known and static multi-dimensional channel by a specific type of the information to be transmitted, this transceiver would suffer from an innate poor generalization of a DNN, when a real channel condition is an outlier for the training data set. It is mainly due to that the transceiver (all of its neurons) is frozen during its inference. To improve the generalization and reduce its vulnerability against varying hostile channels, we introduce a classic communication algorithm, message-passing-algorithm (MPA), into an auto-encoder. Thanks to the efficient iteration of MPA, the transmitter could adjust itself in real time according to a roughly estimated channel, leaving the receiver (DNN) unchanged. We can also show that this concept could widen DNN-based transceiver’s applicability by its MPA-enhanced generalization. At last, this example lights up a potential direction that some classic algorithms widely used in communication system could be integrated into DNN and improve it.
Mr. Yiqun GE obtained his bachelor degree from Shanghai Jiaotong University and master degree from Ecole Nationale Superieure des telecommunication de Bretagne. He joined STMicroelectronics as research engineer. Then, he joined Huawei Technology Canada as researcher in the field of 5G, mainly focusing on channel code and chip architecture. Recently, he is a distinguished research engineer working on machine learning and its application on the future wireless system.
Keynote 3: “Semantic Communications: Beyond Transmitting Bits”
Monday 13 September, 17:11-17:46 (UTC/GMT +3)
Dr. Zhijin Qin, Queen Mary University of London, UK
In the past decades, communications primarily focus on how to accurately and effectively transmit symbols (measured by bits) from the transmitter to the receiver, in which bit-error rate or symbol-error rate is usually taken as the performance metrics. With the development of cellular communication systems, the achieved transmission rate has been improved tens of thousands of times and the system capacity is gradually approaching to the Shannon limit. Inspired by powerful deep learning technologies, semantic communications have been regarded as a promising solution to further improve the system efficiency, which is recognized as the second level of communications by Shannon and Weaver. Semantic communications aim to realize the successful semantic information exchange rather than receive the transmitted bit sequences or symbols accurately. In this talk, the concept of the semantic communication and its key difference from typical communications will be introduced first. Afterwards, the recent work on deep learning enabled semantic communications will be discussed.
Dr. Zhijin Qin is a lecturer (assistant professor) at Queen Mary University of London. She was with Lancaster University and Imperial College London as a lecturer and research associate, respectively, from 2016 to 2018. Her research interests include semantic communications, compressive sensing and low-power wide-area networks for IoT applications. She serves as an area editor of IEEE JSAC Series on Machine learning in Communications and Networks, an associate editor of IEEE Transactions on Communications, IEEE Transactions on Cognitive Communications and Networking, and IEEE Communications Letters. She served as the symposium co-chair for IEEE VTC Fall 2019 and Globecom 2020, 2021. She received the Best Paper Award from IEEE Globecom 2017, and the IEEE Signal Processing Society Young Author Best Paper Award 2018.
A Kalman-Based Autoencoder Framework for End-To-End Communication Systems
Bin Hu (Huawei Techonologies, China); Jian Wang (Huawei Techologies, China); Chen Xu (Huawei Technologies Co., Ltd., China); Gongzheng Zhang and Rong Li (Huawei Technologies, Co. Ltd., China)
A Signal Detection Scheme Based on Deep Learning in OFDM Systems
Guangliang Pan, Zitong Liu and Wei Wang (Nanjing University of Aeronautics and Astronautics, China); Minglei Li (China University of Petroleum (East China), China)
Adaptive Modulation for Wireless Federated Learning
Xinyi Xu, Guanding Yu and Shengli Liu (Zhejiang University, China)
AoI Optimal UAV Trajectory Planning: A Deep Recurrent Reinforcement Learning Approach
Mengjie Wu, Huijia Chi and Shuying Gan (Northwest A&F University, China); Xijun Wang (Sun Yat-sen University, China); Chao Xu (Northwest A&F University, China)
Client Selection Based on Label Quantity Information for Federated Learning
Jiahua Ma and Xinghua Sun (Sun Yat-sen University, China); Wenchao Xia (Nanjing University of Posts and Telecommunications, China); Xijun Wang and Xiang Chen (Sun Yat-sen University, China); Hongbo Zhu (Nanjing University of Posts and Telecommunications, China)
Deep Reinforcement Learning Based Caching Placement and User Association for Dynamic Cellular Networks
Yue Wang, Chunyan Feng and Tiankui Zhang (Beijing University of Posts and Telecommunications, China)
Deep Reinforcement Learning-Based Multi-Panel Beam Management in Massive MIMO Systems: Algorithm Design and System-Level Simulation
Yang Li (China Academy of Information and Communications Technology, China); Jiamo Jiang (China Academy of Information and Communications Technology (CAICT), China); Chao Jia (Beijing University of Posts and Telecommunications, China); Yifei Yuan (China Mobile Research Institute, China); Zhongyuan Zhao (Beijing University of Posts and Telecommunications, China); Ying Du and Zhiqin Wang (China Academy of Information and Communications Technology, China)
Fast Convergence for Federated Learning in OFDMA Systems
Deshi Ye, Songyang Chen and Can Wang (Zhejiang University, China)
GPAE-LSTMnet: A Novel Learning Structure for Mobile MIMO Channel Prediction
Xiao Zhuoran, Zhaoyang Zhang, Chongwen Huang, Caijun Zhong and Xiaoming Chen (Zhejiang University, China)
Memetic Algorithm Based on Community Detection for Energy-Efficient Service Migration Optimization in 5G Mobile Edge Computing
Guo Li, Ling Liu, Zhengping Liang, Xiaoliang Ma and Zexuan Zhu (Shenzhen University, China)
Relevance-Based Wireless Resource Allocation for a Machine Learning-Based Centralized Control System
Afsaneh Gharouni and Peter Rost (Nokia Bell Labs, Germany); Andreas Maeder (Nokia Networks, Germany); Hans D. Schotten (University of Kaiserslautern, Germany)
Smart Scheduling Based on Deep Reinforcement Learning for Cellular Networks
Jian Wang (Huawei Techologies, China); Chen Xu (Huawei Technologies Co., Ltd., China); Rong Li (Huawei Technologies, Co. Ltd., China); Yiqun Ge (Huawei Technologies Canada Inc., Canada); Jun Wang (Huawei Technologies Co. Ltd, China)