Title: | Identifiability of PDEs from Trajectory Data and Some Novel Methods based on Group Projected Subspace Pursuit |
Speaker: | Dr. Yuchen He, Institute of Natural Sciences, Shanghai Jiao Tong University, China |
Time/Place: | 10:00 - 11:00 FSC1110, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Thanks to the advanced data acquisition techniques, data-driven PDE identification has become a popular topic in various areas of science and industry. One can regard a wide class of PDEs as polynomials of partial derivatives of the functions of interest, and the unknown PDE is identified if the involved monomials are determined. This point of view fosters applications and developments of sparse-regression algorithms as well as model selection techniques. In the first part of the talk, we will introduce some recently proposed methods for PDE identification based on a newly developed group projected subspace pursuit algorithm. In the second part of the talk, we will provide some theoretical insights on the identifiability problem related to different types of PDEs. Specifically, we will consider when PDE can be exactly recovered and why certain types of PDEs are harder to identify than the others. |
Title: | Towards resolving information veracity challenge in finance |
Speaker: | Dr. Houping Xiao, J. Mack Robinson College of Business, Georgia State University, USA |
Time/Place: | 11:30 - 12:30 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Recently, investors can obtain earnings forecast information through traditional venues, such as Wall Street, Institutional Brokers’ Estimate System (IBES) as well as modern social media platforms like Estimize which generates consensus estimates based on the forecasts from individuals with different backgrounds. As a result, this will inevitably lead to conflicts in the earnings forecast. This paper presents a novel and effective optimization-based approach to resolve such conflicts in earnings forecast data and further generates accurate and robust earnings forecast consensus. Consistent with the wisdom-of-crowds effect, the new earnings forecast consensus is more accurate than Wall Street consensus (67.5% of estimations with error less than the Wall Street) and IBES consensus (67.4% of estimations with error less than the IBES) of the time. Moreover, the new earnings forecast consensus can provide incrementally useful information in forecasting earnings and the incremental information is further priced in the market after the earning announcement. |
Title: | Estimation and model selection for nonparametric function-on-function regression |
Speaker: | Professor Yuedong Wang, Department of Statistics and Applied Probability, University of California at Santa Barbara, USA |
Time/Place: | 10:00 - 11:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Regression models with functional response and functional covariates have recently received significant attention. While various nonparametric and semiparametric models have been developed, there is an urgent need for model selection and diagnostic methods. We will present a unified framework for estimation and model selection in nonparametric function-on-function regression. We consider a general nonparametric functional regression model with the model space constructed through smoothing spline analysis of variance (SS ANOVA). The proposed model reduces to some existing models when selected components in the SS ANOVA decomposition are eliminated. We propose new estimation procedures under either L1 or L2 penalty and show that combining the SS ANOVA decomposition and the L1 penalty provides powerful tools for model selection and diagnostics. We establish consistency and convergence rates for estimates of the regression function and each component in its decomposition under both the L1 and L2 penalties. Simulation studies and real examples show that the proposed methods perform well. |
Title: | Component Selection for Exponential Power Mixture Models |
Speaker: | Dr. FENG Zhenghui, School of Science, Harbin Institute of Technology, Shenzhen, China |
Time/Place: | 11:00 - 12:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Exponential Power (EP) family is a much flexible distribution family including Gaussian family as a sub-family. In this article, we study component selection and estimation for EP mixture models and regressions. The assumption on zero component mean in [1] is relaxed. To select components and estimate parameters simultaneously, we propose a penalized likelihood method, which can shrink mixing proportions to zero to achieve components selection. Modified EM algorithms are proposed, and the consistency of estimated component number is obtained. Simulation studies show the advantages of the proposed methods on accuracies of component number selection, parameter estimation, and density estimation. Analysis of value at risk of SHIBOR and a climate change data are given as illustration. |
Title: | Generalized Nash equilibrium problems and polynomial optimization |
Speaker: | Dr. Xindong Tang, Department of Applied Mathematics, Hong Kong Polytechnic University |
Time/Place: | 15:00 - 15:30 Zoom, (Meeting ID: 916 5380 5384) |
Abstract: | We consider generalized Nash equilibrium problems (GNEPs) given by polynomial functions. Based on the Karush-Kuhn-Tucker optimality conditions and Lagrange multiplier expressions, we formulate polynomial optimization problems for finding candidate solutions to GNEPs. Then, we introduce the feasible extensions to preclude KKT points that are not solutions when the GNEP is nonconvex, and we apply the Moment-Sum-of-Squares hierarchy for solving occurring polynomial optimization problems. We show that this approach guarantees to find a solution to the GNEP or detect the nonexistence of solutions after solving finitely many semidefinite programs under some generic assumptions. We also investigate several other approaches for solving GNEPs given by polynomials. Besides that, we study some research topics related to polynomial optimization, such as sparse polynomial optimization, completely positive tensors, variational inequality problems of polynomials, etc. |
Title: | Image Vectorization by Affine Shortening Flow |
Speaker: | Dr. Yuchen He, Institute of Natural Sciences, Shanghai Jiao Tong University, China |
Time/Place: | 16:30 - 17:00 Zoom, (Meeting ID: 916 5380 5384) |
Abstract: | Image vectorization is an important parameterization process that converts raster images to Scalable Vector Graphics (SVG), which becomes resolution-free. In this talk, I will introduce a technique based on affine shortening flow, and discuss its extension from binary images to color images. Examples are presented to illustrate its effectiveness and efficiency compared to other state-of-art software and algorithms. If time allows, I will briefly present my other research projects. |
Title: | A Unified Approach to Synchronization Problems over Subgroups of the Orthogonal Group |
Speaker: | Dr. Huikang Liu, School of Information Management & Engineering, Shanghai University of Finance and Economics, China |
Time/Place: | 17:00 - 17:30 Zoom, (Meeting ID: 916 5380 5384) |
Abstract: | The problem of synchronization over a group aims to estimate a collection of group elements based on noisy observations of a subset of all pairwise ratios. Such a problem has gained much attention recently and finds many applications across a wide range of scientific and engineering areas. In this talk, we consider the class of synchronization problems in which the group is a closed subgroup of the orthogonal group. This class covers many group synchronization problems that arise in practice. First, we propose a unified approach for solving this class of group synchronization problems, which consists of a suitable initialization step and an iterative refinement step based on the generalized power method, and show that it enjoys a strong theoretical guarantee on the estimation error under certain assumptions on the group, measurement graph, noise, and initialization. Second, we verify the assumptions on the measurement graph and noise for standard random graph and random matrix models. Thirdly, based on the classic notion of metric entropy, we develop and analyze a novel spectral-type estimator. Finally, we show via extensive numerical experiments that our proposed non-convex approach outperforms existing approaches in terms of computational speed, scalability and estimation error. |
Title: | A Flexible and Parsimonious Modelling Strategy for Clustered Data Analysis |
Speaker: | Prof. ZHANG Wenyang, Department of Mathematics, The University of York, UK |
Time/Place: | 11:00 - 12:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Statistical modelling strategy is the key for success in data analysis. The trade-off between flexibility and parsimony plays a vital role in statistical modelling. In clustered data analysis, in order to account for the heterogeneity between the clusters, certain flexibility is necessary in the modelling, yet parsimony is also needed to guard against the complexity and account for the homogeneity among the clusters. In this talk, I will introduce a flexible and parsimonious modelling strategy for clustered data analysis. The strategy strikes a nice balance between flexibility and parsimony, and accounts for both heterogeneity and homogeneity well among the clusters, which often come with strong practical meanings. In fact, its usefulness has gone beyond clustered data analysis, it also sheds promising lights on transfer learning. An estimation procedure is developed for the unknowns in the resulting model, and asymptotic properties of the estimators are established. Intensive simulation studies are conducted to demonstrate how well the proposed methods work, and a real data analysis is also presented to illustrate how to apply the modelling strategy and associated estimation procedure to answer some real problems arising from real life. |
Title: | Towards Efficient and Explainable Deep Learning: A Training Dynamics Perspective |
Speaker: | Dr Wei HUANG, RIKEN Center for Advanced Intelligence Project, Japan |
Time/Place: | 11:00 - 12:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Deep learning theory research aims to elucidate the complex learning dynamics of neural networks via (stochastic) gradient descent, thereby enhancing our comprehension of the deep learning process. This presentation will illuminate the shift from the neural tangent kernel (NTK) regime to feature learning dynamics, offering a comprehensive outlook on the subject matter. We will initially explore the interplay between learning dynamics and the generalization properties of deep active learning and neural architecture search. Leveraging over-parameterization, we aim to bridge the chasm between theoretical understanding and practical methodologies in these areas, characterizing theoretical properties while concurrently proposing pragmatic methods for improvement. Moreover, we delve into the dynamics of Graph Neural Networks (GNNs) through the lens of feature learning. By examining the role of graph convolution in the context of feature learning theory in neural networks utilizing gradient descent training, we highlight the superior performance of GNNs over Multi-Layer Perceptrons (MLPs). |
Title: | Kernel-based least-square collocation methods for solving PDEs on surfaces |
Speaker: | Dr. Meng CHEN, School of Mathematics and Computer Sciences, Nanchang University, China |
Time/Place: | 14:30 - 15:30 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | There are plenty of applications and analysis for time-independent elliptic partial differential equations in the literature hinting at the benefits of overtesting by using more collocation conditions than the number of basis functions. Overtesting not only reduces the problem size, but is also known to be necessary for stability and convergence of widely used unsymmetric Kansa-type strong-form collocation methods. We consider kernel-based meshfree methods, which is a method of lines with collocation and overtesting spatially, for solving parabolic partial differential equations on surfaces without parametrization. In this work, we extend the time-independent convergence theories for overtesting techniques to the parabolic equations on smooth and closed surfaces. |
Title: | Applying statistics in social science research: an epistemological examination through citation network analysis |
Speaker: | Dr. LU Ying, Department of Applied Statistics, Social Science and Humanities NYU Steinhardt, New York University, USA |
Time/Place: | 11:00 - 12:00 FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University |
Abstract: | Quantitative social science research emphasizes the importance of empirical evidence and applies scientific methods to study social phenomena. This approach is grounded in positivism thinking that assumes that social reality can be objectively measured and observed, and that there are universal laws governing social behavior. In particular, statistical and mathematical methods are used to analyze data and test hypotheses with respect to the universal laws. In contrast, qualitative methods such as interviews, field observations take a different epistemological approach that is grounded in experientialism thinking. It believes that individuals actively construct their own realities through their experiences, and rejects the ideas universal laws of social behavior. In this paper, we examine the philosophical underpinning of the various statistical approaches commonly used in social science research, such as experiments and quasi-experiments, measurements, surveys and social statistics. We will also discuss the limitations of these approaches in understanding the constructive nature of the society and the complex human behavior in cultural and historical contexts. As a case study, we conducted a citation network analysis by examining the bibliography of ~5,000 peer reviewed articles in the field of education published between 2010 and 2020. Our study demonstrates an unfortunate separation in research citations between the two methodological approaches. Keyword and author analyses further reveal the distinct patterns of the two social science knowledge discovery traditions. |
Title: | PhD Oral Defense: Selected Topics in Approximate Bayesian Inference |
Speaker: | Mr DONG Qishi, Department of Mathematics, Hong Kong Baptist University, Hong Kong |
Time/Place: | 14:30 - 16:30 FSC1217, or Zoom (Meeting ID: 963 9746 5834 Password: 887752) |
Abstract: | In many machine learning tasks, probability plays an important role as the data from real-world scenarios is not deterministic and its variation can be characterized using probabilistic models. One of the key problems in its application is inferring the posterior distributions of latent variables. However, for most models of interest, exact inference is intractable. Thus, researchers seek to approximate the posterior distributions in various ways, among which the most popular two are Monte Carlo techniques and variational inference. The focus of this presentation is variational inference (VI), where we try to provide some new understandings of VI from both theoretical and practical perspectives. Firstly, we conduct a statistical analysis of VI on the variable selection using Bayesian linear model and prove the estimation and selection consistency. We then discuss probabilistic clustering and show the objective of variational inference for the Gaussian mixture model reduces to a regularized optimization problem and has a deep connection with a distributionally robust optimization. The second half focuses on the application part by first considering the change point detection. We propose the Bayesian Change Point Model associated with a scalable variational EM algorithm that can automatically do change point detection, subsequent inference, and hyperparameter learning. In the second application, we build a model zoo of PixelCNN++ to improve the OoD image compression ratio. A von-Mises Fisher-based Kalman filter is proposed with a new sequential Bayesian inference algorithm based on stein variational gradient descent. Extensive experiments and simulations show both methods demonstrate our advantages over existing methods in both accuracy and efficiency. |
We organize conferences and workshops every year. Hope we can see you in future.
Learn MoreProf. M. Cheng, Dr. Y. S. Hon, Dr. K. F. Lam, Prof. L. Ling, Dr. T. Tong and Prof. L. Zhu have been awarded research grants by Hong Kong Research Grant Council (RGC) — congratulations!
Learn MoreFollow HKBU Math