The 2023 International Symposium on Learning and Information Theory was held.
Group photo of participants (Shenzhen venue)
From July 3rd to 5th, the 2023 International Seminar on Learning and Information Theory was held in Shenzhen Institute of Talent Training and Gao Kun Convention Center. This seminar was sponsored by the Institute of Data and Information of our institute (hereinafter referred to as "Digital Information Institute"), the Institute of Network Coding of the Chinese University of Hong Kong and the Shenzhen Institute of Big Data. Eighteen outstanding scholars from universities and research institutes at home and abroad, such as the University of Hong Kong, the Chinese University of Hong Kong, the Massachusetts Institute of Technology, and Columbia University, were invited to share and exchange academic knowledge. Nearly 200 participants deeply discussed the related issues in the field of machine learning theory and information theory.
Academician Zhang Xiaoping delivered the opening speech.
Academician of the Canadian Academy of Engineering, Fellow of the International Association of Electrical and Electronics Engineers, Professor Tsinghua University Peng Rui, Dean of the Institute of Mathematics and Information Technology of our Institute, Zhang Xiaoping, Co-Director of the Institute of Network Coding of the Chinese University of Hong Kong and Zhuomin Chair Professor of Information Engineering, Weihao Yang, Foreign Academician of the China Academy of Engineering, Academician of the Royal Canadian Academy of Sciences, Vice President of the Chinese University of Hong Kong (Shenzhen), Dean of Shenzhen Big Data Research Institute, Director and Chair Professor of the Institute of Data Science of the University of Hong Kong, and Ma Yi, Director of the Department of Computer Science of the University of Hong Kong, delivered opening speeches for The meeting was presided over by Huang Shaolun, an associate professor of Mathematics and Information College of our college.
Weihao Yang, Ma Yi, Lin Zhengren, Zhang Rui, Sun Ruoyu and Er Jiang (from left to right and from top to bottom) gave academic speeches.
On the morning of July 3rd, Weihao Yang demonstrated that entropy function plays a central role in information theory by summarizing the development of machine-proved entropy inequality in the past 25 years. Starting from the history of intelligence research, Ma Yi provided a more systematic and principled view on the practice of artificial intelligence. Masahito Hayashi, a fellow of the International Institute of Electrical and Electronics Engineers, a fellow of the International Society of Mathematical Statistics and a professor of the Chinese University of Hong Kong (Shenzhen), summarized an algorithm recently proposed in the context of Arimoto-Blahut algorithm, showed various convergence theorems and applied the algorithm to the objective problem of EM algorithm.
On the afternoon of July 3rd, Zhang Rui, the chief research scientist of Shenzhen Big Data Research Institute and professor of Xue Qin, President of the Chinese University of Hong Kong (Shenzhen) Institute of Technology, discussed the basic concepts and technical points of intelligent reflector (IRS), its application in wireless networks, the main design difficulties in IRS-assisted communication related fields and the new progress related to IRS in recent years. Sun Ruoyu, a senior research scientist of Shenzhen Big Data Research Institute, Shenzhen International Center for Industrial and Applied Mathematics, and an associate professor of the School of Data Science of the Chinese University of Hong Kong (Shenzhen), combined with his recent research results, demonstrated the specific behavior of Adam optimizer under different hyperparameter settings through mathematical deduction, demonstrated its convergence, and put forward suggestions for adjusting parameters in practical application. Ercan E.Kuruo?lu, a professor at the School of Mathematics and Information Science in our college, introduced the basic concept of graph signal processing and adaptive graph signal processing, and introduced a new algorithm for processing non-Gaussian data and time-varying graph data in related problems. Later, the forum guests Zhang Xiaoping, Tian Chao, Zheng Lizhong and Weihao Yang held a round-table discussion on information theory and innovation in the AI Big Bang era.
Zheng Lizhong, Tian Chao, Amin Gohari, Han Guangyue and Heim Perlmutter gave academic speeches (from left to right and from top to bottom).
On the morning of July 4th, Zheng Lizhong, a fellow of the International Institute of Electrical and Electronics Engineers and a professor at the Massachusetts Institute of Technology, showed the latest progress of modal decomposition based on neural network, which proved that this result can be used for problems such as designing receivers that can use parameters as auxiliary information. Tian Chao, an associate professor at Texas A&M University, introduced that the generalization error boundary of information theory is helpful to solve the problem that the existing methods are insufficient in understanding the generalization behavior of machine learning algorithms. Han Guangyue, a professor at the University of Hong Kong, studied the continuous-time additive Gaussian white noise channel with possible feedback, quantitatively strengthened the mutual information convergence between the associated discrete-time channel and the original continuous-time channel, and revealed how oversampling can approximate the true mutual information of the continuous-time Gaussian channel with bandwidth constraints.
On the afternoon of July 4th, Haim Permuter, a professor at Ben-Gurion University, introduced the basic concepts and calculation formulas of information measure and the application of related theories in capacity area estimation of network channel number. Chandra Nair, a professor at the Chinese University of Hong Kong, went deep into nonconvex optimization problems in information theory and proved that these problems have unique local maxima and can therefore be solved by gradient descent method. Amin Gohari, distinguished associate professor of the President of the Chinese University of Hong Kong, discussed the relationship between Shannon inequality and generalized relative entropy, and proposed a new divergence to satisfy the specific Shannon inequality.
Student poster display link
On that day, the students of our college also presented posters with 36 students from Hong Kong University, Chinese University of Hong Kong (Shenzhen) and south university of science and technology of china. After evaluation by the jury, Zhang Anping of our hospital won the first prize of poster display, Liu Fan of south university of science and technology of china won the second prize of poster display, and Jerry Lee and Wang Yongheng of our hospital won the third prize of poster display.
Niz Gonduz, Cynthia Rush, Nicholas pappas, Uri Cohen, Fitios Stavrou and Hummer gave academic speeches (from left to right and from top to bottom).
On the morning of July 5th, Deniz Gunduz, a fellow of the International Institute of Electrical and Electronics Engineers and a professor in Imperial College London, UK, discussed how to design a semantic and practical communication system to meet the needs of modern applications, and reduce the communication load by identifying and sending the most relevant information only for basic tasks. Cynthia Rush, an associate professor at Columbia University, introduced the application of L1 regularization in high-dimensional statistical estimation, focusing on the role of SLOPE estimator in linear regression and the optimal compromise between FDP and TPP. Nikolaos Pappas, an associate professor at Linxueping University in Sweden, discussed the methods to meet the requirements of real-time communication, independent decision-making and efficient distributed processing of emerging Internet of Things applications, emphasizing the freshness and value of information.
On the afternoon of July 5th, Uri Cohen, a postdoctoral researcher at Cambridge University, introduced how to relate the geometric shape of object manifold with the classification ability of neuron population by statistical physical methods. It was found that the change of geometric shape at different levels is the key to improve the classification ability. Photios Stavrou, an assistant professor at the European Institute of Telecommunications, introduced the application of rate distortion theory and its variants in information theory in target-oriented semantic communication, including the design of selective decoder, rate distortion perception function, rate distortion representation calculation method and unpredictable rate distortion function. Hummer, a postdoctoral researcher at Austrian Institute of Science and Technology, discussed how to extract information components with small effective dimensions from a large number of noises, and introduced a new proof technique, using approximate message passing theory to systematically characterize the spectral characteristics of spike random matrices.
Group photo of participants (Hong Kong venue)
Text | Chen Yanyan, Chen Bohan, Guo Jia
Figure | Zhang Mengwen
Editor | Dai Yujing
Audit | Chen Chaoqun
Typesetting | Li Xinyu
Declaration: All article resources on this website, unless otherwise specified or labeled, are collected from online resources. If the content on this website infringes on the legitimate rights and interests of the original author, you can contact this website to delete it.
