Georgia Institute of Technology College of Sciences

Numerical algebraic geometry revolves around the study of solutions to polynomial systems via numerical method. Two of the fundamental tools in this field are the polyhedral homotopy of Huber and Sturmfels for computing isolated solutions and the concept of witness sets put forth by Sommese and Wampler as numerical representations for non-isolated solution components. In this talk, we will describe a stratified polyhedral homotopy method that will bridge the gap between these two largely independent area. Such a homotopy method will discover numerical representations of non-isolated solution components as by-products from the process of computing isolated solutions. We will also outline the pipeline of numerical algorithms necessary to implement this homotopy method on modern massively parallel computing architecture.

Given a collection of functions in a Banach space, typically called a dictionary in machine learning, we study the approximation properties of its convex hull. Specifically, we develop techniques for bounding the metric entropy and n-widths, which are fundamental quantities in approximation theory that control the limits of linear and non-linear approximation. Our results generalize existing methods by taking the smoothness of the dictionary into account, and in particular give sharp estimates for shallow neural networks. Consequences of these results include: the optimal approximation rates which can be attained for shallow neural networks, that shallow neural networks dramatically outperform linear methods of approximation, and indeed that shallow neural networks outperform all continuous methods of approximation on the associated convex hull. Next, we discuss greedy algorithms for constructing approximations by non-linear dictionary expansions. Specifically, we give sharp rates for the orthogonal greedy algorithm for dictionaries with small metric entropy, and for the pure greedy algorithm. Finally, we give numerical examples showing that greedy algorithms can be used to solve PDEs with shallow neural networks.

Many real life processes that we would like to model have a self-exciting property, i.e. the occurrence of one event causes a temporary spike in the probability of other events occurring nearby in space and time.  Examples of processes that have this property are earthquakes, crime in a neighborhood, or emails within a company.  In 1971, Alan Hawkes first used what is now known as the Hawkes process to model such processes.  Since then much work has been done on estimating the parameters of a Hawkes process given a data set and creating variants of the process for different applications.

In this talk, we propose a new variant of a Hawkes process, called a self-limiting Hawkes process, that takes into account the effect of police activity on the underlying crime rate and an algorithm for estimating its parameters given a crime data set.  We show that the self-limiting Hawkes process fits real crime data just as well, if not better, than the standard Hawkes model.  We also show that the self-limiting Hawkes process fits real financial data at least as well as the standard Hawkes model.

Baker and Lorscheid have developed a theory of foundation that characterize the representability of matroids. Justin Chen and I are developing a computer program that computes representations of matroids based on the theory of foundation. In this talk, I will introduce backgrounds on matroids and the foundation, then I will talk about the key algorithms in computing the morphisms of pastures. If possible, I will also show some examples of the program.

Teams link: https://teams.microsoft.com/l/meetup-join/19%3a3a9d7f9d1fca4f5b991b4029b09c69a1%40thread.tacv2/1648750292956?context=%7b%22Tid%22%3a%22482198bb-ae7b-4b25-8b7a-6d7f32faa083%22%2c%22Oid%22%3a%2206706002-23ff-4989-8721-b078835bae91%22%7d