报告主题：Federated Learning and Analysis with Multi-access Edge Computing

报 告 人：韩竹

时    间：2024年10月15日 9:00-10:00

地    点：腾讯会议号568-2474-3091

报告摘要：

With the maturity of edge computing and the large amount of data generated by loT devices, we have witnessed an increasing number of intelligent applications in wireless networks. The growing awareness of privacy further motivates the wide study and deployment of federated learning, a collaborative distributed model training framework for predictive tasks, However, a wide range of applications, more broadly relevant to data analytics and query in wireless networks, cannot be well supported by this framework. These applications usually require more complex and diverse aggregation methods, instead of the simple weight aggregations, and are broadly nourished by statistics, information theory, and signa processing, besides machine learning. This talk aims to present the recent advances in federated analytics at the intersection of data science, wireless communication, and security and privacy, We will present the definition, taxonomy, and architecture of the federated analytics techniques. It will also cover several practical and important data analytics tasks in wireless networks, including federated anomaly detection, federated frequent pattern analysis, federated distribution estimation and skewness analytics. Finally, we will present important challenges, open problems, and future directions at the intersection of federated learning/analysis and wireless networks.

报告主题：The Transformative lmpact of Generative Al: strategies, Applications, and innovations

报 告 人：韩竹

时    间：2024年10月15日 10:00-11:00

地    点：腾讯会议号568-2474-3091

报告摘要：

This speech explores the transformative potential of Generative Al (GAl) across various domains. Firstly, we examine the use of Large Language Models (LLMs)in repeated games to develop practical strategies aligning with the folk theorem's equilibrium conditions, enhancing cooperative behavior through future payoff considerations. Secondly, we address lowlight image enhancement in teleoperation using diffusion-based Al generated content (AlGC) models. A Vision Language Model (VLM)empowered contract theory framework optimizes AlGC task allocation and pricing under information asymmetry, improving resource management for teleoperators and edge servers. In the realm of autonomous driving, we integrate Federated Learning (FL)with Vision-language models (VLMs) in Graph Visual Question Answering (GVQA), highlighting advancements in privacy preservation, reduced communication costs, and maintained model performance. Lastly, an LLM-based semantic communication (SC) framework for underwater communication is presented, demonstrating efficient data transmission and resilience against noise and signal loss by performing semantic compression and prioritization of image data. These innovations collectively illustrate the broad impact of Al technologies, shaping strategies, and enhancing applications across various fields.

报告人简介：

韩竹，IEEE Fellow、ACM Fellow、AAAS Fellow、美国艺术与科学院院士以及Web of Science全球高被引进科学家，IEEE通信协会杰出讲者。1997年获清华大学电子工程学士学位，1999年和2003年分别获马里兰大学学院帕克分校电子与计算机工程硕士学位和博士学位。2000年至2002年，他在马里兰州日耳曼敦的JDSU担任研发工程师；2003年至2006年，他在马里兰州立大学担任助理研究员；2006年至2008年，他在爱荷华州博伊西大学担任助理教授。目前，他是美国德克萨斯州休斯顿大学电气与计算机工程系以及计算机科学系的John and Rebecca Moores教授。他是2010年国家自然科学基金CAREER award 获得者，2021年lEEE Kiyo Tomiyasu奖获得者。他自2014年起成为lEEE研究员，2020年起成为AAAS研究员，2024年起成为ACM研究员，2015年至2018年成为lEEE杰出讲师，2022年至2025年成为ACMD杰出演讲者。自2017年以来，他一直是全球被引用次数最高的前1%研究人员之一。

报告主题：The Third Wave of Artificial Intelligence: Neurosymbolic Al

报 告 人：宋厚冰

时    间：2024年10月17日 9:00-10:00

地    点：腾讯会议号568-2474-3091

报告摘要：

There are three waves of Artificial Intelligence, The first Wave of AI is Crafted Knowledge, which includes rule-based AI systems. The second wave of AI is Statistical Learning, which includes machine becoming intelligent by using statistical methods. The third wave of AI is contextual adaptation. In the third wave, instead of learning from data intelligent machines will understand and perceive the world on its own, and learn by understanding the world and reason with it. Neurosymbolic Al, which combines neural networks with symbolic representations, has emerged as a promising solution of the third wave of AI. In this talk, I will my perspective on the emerging area.

报告主题：Al for Cybersecurity and Security of Al

报 告 人：宋厚冰

时    间：2024年10月17日 10:00-11:00

地    点：腾讯会议号568-2474-3091

报告摘要：

The mutual needs and benefits of Al and cybersecurity have been widely recognized. AI techniques are expected to enhance cybersecurity by assisting human system managers with automated monitoring, analysis, and responses to adversarial attacks. Conversely, it is essential to guard AI technologies from unintended uses and hostile exploitation by leveraging cybersecurity practices. The interplay between AI/machine learning, and cybersecurity introduces new opportunities and challenges in the security of AI as well as AI for cybersecurity. In this talk I will present my perspective on AI for cybersecurity and the security of AI.

报告人简介：

宋厚冰，IEEE Fellow，ACM杰出会员和杰出讲者，2012年8月获得弗吉尼亚州夏洛茨维尔弗吉尼亚大学电子工程博士学位。他目前是马里兰大学巴尔的摩分校（UMBC）的终身副教授、国家自然科学基金航空大数据分析中心（规划）主任和网络全球安全与优化实验室（SONG Lab, www.SONGLab.us）主任。之前他是美国安伯瑞德航空航天大学电气工程与计算机科学终身副教授。他担任IEEE Transactions on Artificial Intelligence (TAI) (2023年至今)、IEEE Internet of Things Journal (2020年至今)、IEEE Transactions on Intelligent Transportation Systems (2021年至今)和IEEE Journal on Miniaturization for Air and Space Systems (J-MASS) (2020年至今)的副主编。他曾任《IEEE 通信杂志》副技术编辑（2017-2020年）。他是8本书籍的编辑，100多篇文章的作者和2项专利的发明人。他的研究兴趣包括网络物理系统/物联网、网络安全和隐私以及人工智能/机器学习/大数据分析。

报告主题：Digital Twin Enhanced Federated Reinforcement Learning with Lightweight Knowledge Distillation in Mobile Networks

报 告 人：周晓康

时    间：2024年10月18日 14:00-16:30

地    点：工科E座1716会议室

报告摘要：

In this study, in order to facilitate the lightweight model training and real-time processing in high-speed mobile networks, we design and introduce an end-edge-cloud structured three-layer Federated Reinforcement Learning (FRL) framework, incorporated with an edge-cloud structured DT system. A dual-Reinforcement Learning (dual-RL) scheme is devised to support optimizations of client node selection and global aggregation frequency during FL via a cooperative decision-making strategy, which is assisted by a two-layer DT system deployed in the edge-cloud for real-time monitoring of mobile devices and environment changes. A model pruning and federated bidirectional distillation (Bi-distillation) mechanism is then developed locally for the lightweight model training, while a model splitting scheme with a lightweight data augmentation mechanism is developed globally to separately optimize the aggregation weights based on a splitted neural network structure (i.e., the encoder and classifier) in a more targeted manner, which can work together to effectively reduce the overall communication cost and improve the non-IID problem.

报告人简介：

周晓康，现任日本关西大学商务数据科学学院副教授。2014年毕业于日本早稻田大学，获人类信息科学博士学位。2012至2015年，于早稻田大学人间科学学术院任助教。2016至2024年，就职于日本国立滋贺大学数据科学学院(讲师/副教授)。2017年起，于日本理化研究所革新知能综合研究中心（AIP）兼职任客员研究员。研究领域覆盖计算机科学，数据科学和社会人类信息学，主要关注大数据、机器学习、行为认知、普适计算智能与安全等方面。发表学术期刊/会议论文200余篇，其中SCI期刊论文130余篇 (中科院1区，IEEE/ACM Trans 90余篇，入选ESI高被引17篇，ESI热点8篇)。入选2022-2024斯坦福大学发布全球前2%顶尖科学家。荣获多项国际性奖励与荣誉，如2023, 2020 IEEE SMC Society Andrew P. Sage Best Transactions Paper Award最佳汇刊论文奖, 2023 IEEE Industrial Electronics Society TC-II Best Paper年度最佳期刊论文，2022 IEEE HITC Award for Excellence in Hyper-Intelligence (Early Career Researcher)优秀青年科学家，2021滋贺大学校长奖，2020 IEEE TCSC Award for Excellence for Early Career Researchers优秀青年科学家等。目前在AIHC担任区域编委，TCSS, TCE, IoTJ, Big Data Mining and Analytics, JCSC, CAEE, HCIS等担任副编委，并于多个IEEE重要国际学术会议担任程序委员会主席。目前为美国IEEE CS, ACM，日本IPSJ, JSAI，中国CCF会员。

报告主题：Visual Sensing for Human-Machine Interaction

报 告 人：Hui Yu

时    间：2024年10月28日 10:00-11:00

地    点：工科E1716

报告摘要：

With the increasing demand of machine intelligence across a wide range of application scenarios, human-machine interaction (HMI) emerges as another essential communication, whereby facial-expression-aware is one of the principal features for natural interaction.  The principal branch of research in my group has been driven by the understanding of facial expression and the causative mechanism of emotion combining knowledge of visual computing with multiple disciplines, such as cognitive computing, as well as machine learning. Multimodal information including visual and biometric signals can record the facial muscle activity or brain activity closely related to facial movements and the internal emotional states. These multiple sensing channels would help provide an insight into the emotion and perception of facial expression, to develop widely accessible HMI solutions able to track facial motions and recognise affective states in a highly efficient and precise manner. I will discuss the development of visual capture of facial expression along with multiple sensing technologies for affective analysis. This talk will also discuss research on the development and challenges of image/video clustering as well as our recent development on this topic.

报告人简介：

Hui Yu，格拉斯哥大学教授。他是该校视觉计算和社交机器人小组的负责人。他的研究兴趣在于视觉和认知计算以及机器学习，应用于社会信号分析的四维面部表情重建和跟踪、人机交互以及智能车辆和视频分析余教授的研究成果为他赢得了多个奖项，并成功与世界各地的机构和企业建立了合作关系。他是IEEE系统、人与控制论学会的副主席，也是英国高科技公司的科学顾问。他是多个不同资金来源项目的主要研究者，这些资金来自英国工程和物理科学研究理事会、欧盟第七框架计划、英国皇家工程院、英国皇家学会、英国创新署及产业界。他荣获了英国皇家工程院的工业奖学金。他还担任《IEEE人类-机器系统汇刊》、《IEEE智能交通系统汇刊》和《IEEE/CAA自动化学报》的副编辑。