Speaker: XU Yan, Associate Professor of Nanyang Technological University

Speaker: BU Kaifeng, Postdoctoral Fellow, Department of Physics, Harvard UniversityVenue: Room 200-9, Run Run Shaw Business Administration Building, Yuquan CampusAbstract:One of the central problems in the study of deep learning theory is to understand complexity of the neural networks. In this talk, we will introduce the topological entropy from dynamic system as a measure of complexity of neural networks. We will also discuss the depth-width trade-offs of neural networks via topological entropy.Contact: LIU Weihua(lwh.math@zju.edu.cn)

Speaker: LI Mo, Associate Professor, School of Computer Engineering, Nanyang Technological UniversityVenue: Online

Speaker: Professor Frank Kemper, RWTH Aachen University Venue: Room A601, Anzhong Building, Zijingang CampusAbstract:In the field of wind engineering, various aspects of scientific fields are interconnected. First, the randomness of the natural wind needs to be handled with respect to a realistic model for the location dependent gust velocity profile for the necessary return period. Second, aerodynamic aspects of the structure of interest, its geometries and its neighbouring buildings need to be considered. Third, the vulnerability towards different kinds of wind induced vibrations needs to be predicted. Finally, reliability analyses are based on profound knowledge of the structure itself and the extreme values distributions. So, methods for - and knowledge about - meteorology, aerodynamics, mechanics, and statistics are needed. All this should follow the goal to find predictions of the wind induced load situation in an extreme case which is as realistic as possible.  The quality appraisement is an analysis, to confirm the level of reliability with respect to the design requirements. Basically, wind engineers are working with same approaches and the international scientific community is actively communicating. However, due to some specific particularities in different countries, there seem to be some differences with respect to the application of wind engineering. Research in the field of engineering in Germany is always very much related to application. So, in this lecture, some newly developed methods/findings and some project related applications will be presented.

Speaker: Professor Ming Yan（Michigan State University）Venue: Room 200-9, Run Run Shaw Business Administration Building, Yuquan CampusAbstract:Large-scale machine learning models are trained by parallel stochastic gradient descent algorithms on distributed or decentralized systems. The communications for gradient aggregation and model synchronization become the major obstacles for efficient learning as the number of nodes and the model's dimension scale up. In this talk, I will introduce several ways to compress the transferred data and reduce the overall communication such that the obstacles can be immensely mitigated.

Speaker: LI Qirui, University of TorontoVenue：1st floor of east 7th teaching building, Lecture Hall of Institute for Advanced Study in Mathematics, Zijingang CampusAbstract: The central values and derivatives of L-functions are studied in the global Gan-Gross-Prasad conjecture and its arithmetic generalizations. These type of problems originally started from the study of Waldspurger furmula and Gross-Zagier formula. Recently, in the function field case, Yun and Zhang have discovered a higher Gross--Zagier formula, relating higher derivatives of L-functions to intersection numbers of special cycles on the moduli space of Drinfeld Shtukas of rank two.The Guo-Jacquet Fundamental Lemma is a higher dimensional generalization of the local field analogue of the Waldspurger formula. It has an arithmetic generalization called the Linear Arithmetic Fundamental Lemma. We will introduce the known results for the linear Arithmetic Fundamental Lemma, and the intersection number formula for Lubin-Tate spaces. After a joint work with Ben Howard, we also discovered a bi-quadratic generalization of the conjecture.Contact: Yifeng Liu (liuyf0719@zju.edu.cn)

Venue: Faculty Club Hall No.4, International Campus

Speaker: WANG Zigan, Assistant Professor of bussiness school, the University of Hong KongVenue: Room 530, School of Economics, West Part of Zijingang CampusAbstract:We hypothesize and provide evidence that the effect of political connections on corporate taxation depends on firms' financial constraints. After an arguably exogenous increase in political connections arising from close congressional elections, financially unconstrained firms increase their tax planning, while constrained firms decrease it. The decrease (increase) in the tax planning of financially constrained (unconstrained) firms is more pronounced when the connected politicians serve on the banking-related (tax-related) committees. Constrained firms also experience a significant decrease in cost of capital and receive more government contracts after obtaining more political connections, supporting the idea that alleviating firms’ financial constraints is the mechanism through which political connections affect the tax planning of constrained firms. Collectively, our findings highlight the importance of financial constraints in understanding the tax planning of politically connected firms.

ECR Program SDG 15, Session on Sustainable Development Goal 15 Life on Land.pdf