[Lecture One] Edge Intelligence and Lightweight Collaborative Learning
Speaker: Yu Dongxiao
Date: April 19, 2023 (Wednesday)
Time: 10:00 am
Venue: Room 321, Zhangheng Building
Lecture content: With the widespread adoption of mobile devices, their computing power is increasing as they generate huge amounts of data. Edge computing integrates ubiquitous resources and is coupled with artificial intelligence. The formed edge agent can indeed implement AI close to the terminal. Collaborative learning is an essential way to achieve edge intelligence. In the context of edge computing, on the one hand, the composition of edge devices is relatively complex, and the computing capacity of devices, the corresponding speed, etc. are different. The second challenge comes from the complex network environment and data environment, which leads to potential malicious nodes, network topology changes, and the possible presence of a large number of outliers in the data set. It is important to design lightweight, highly robust and adaptive collaborative learning algorithms based on the heterogeneous nature of edge devices and the complexity challenges of network and data environments. The report reviews and discusses the development, classification, latest research results and trends in edge intelligence and lightweight collaborative learning, as well as the goals and visions, and discusses how integrated scheduling of upload-deposit-computing resources can enable a new generation of green and efficient collaborative learning frameworks.
About the speaker: Yu Dongxiao, deputy dean of the School of Computer Science and Technology of Shandong University. His research interests include edge intelligence, the Internet of Things and data mining. He has published more than a hundred academic papers in recent years. He has presided over more than a dozen national and provincial projects, such as key research and development projects and natural science foundation projects, and received a second prize in natural science from Shandong Province.
[Lecture Two] On Reinforcement Learning Methods for Wireless Security
Speaker: Xiao Liang
Date: April 22, 2023 (Saturday)
Time: 10:00 am
Venue: Online (Tencent Conference: 299-314-239)
Lecture content: In recent years, wireless network security incidents have been frequent, and the demand for communication anti-jamming and privacy protection is growing rapidly. There is an urgent need to design efficient security mechanisms for wireless networks to defend against hostile interference and other attacks. This report will discuss reinforcement learning approaches for wireless security, using hierarchical architecture, security exploration, and transfer learning to address existing technical issues such as slow optimization speed, poor security performance, and short-term security risks. In this approach, a hierarchical structure based on policy priority is designed to compress the policy space, and a risk avoidance mechanism based on safety constraints is constructed to avoid exploring dangerous strategies that lead to task failure. On this basis, a deep neural network based on expected benefit and risk value is designed to further improve the learning efficiency. Then, we apply the proposed method to optimize the channel selection and transmission power of UAVs for anti-jamming communication, as an example, and show that the proposed method can significantly reduce the performance in terms of bit error rate, packet loss rate, and energy consumption.
About the speaker: Xiao Liang, professor and PhD supervisor at the School of Information at Xiamen University. He works in cybersecurity, big data and machine learning. He has conducted and participated in a number of research projects funded by the National Natural Science Foundation of China and the Fujian Provincial Natural Science Foundation. He has published four academic monographs/chapters in English.
