In this talk I explain the fertile relationship between the foundations of inference and learning and combinatorial geometry.

My presentation contains several powerful examples where famous theorems in discrete geometry answered natural  questions from machine learning and statistical inference:

In this tasting tour I will include the problem of deciding the existence of Maximum likelihood estimator in multiclass logistic regression, the variability of behavior of k-means algorithms with distinct random initializations and the shapes of the clusters, and the estimation of the number of samples in chance-constrained optimization models. These obviously only scratch the surface of what one could do with extra free time. Along the way we will see fascinating connections to the coupon collector problem, topological data analysis, measures of separability of data, and to the computation of Tukey centerpoints of data clouds (a high-dimensional generalization of median). All new theorems are joint work with subsets of the following wonderful folks: T. Hogan, D. Oliveros, E. Jaramillo-Rodriguez, and A. Torres-Hernandez.

Two relevant papers published/ to appear are

https://arxiv.org/abs/1907.09698https://arxiv.org/abs/1907.09698

https://arxiv.org/abs/2205.05743https://arxiv.org/abs/2205.05743

At the intersection of Topological Data Analysis and machine learning, the field of cellular signal processing has advanced rapidly in recent years. In this context, each signal on the cells of a complex is processed using the combinatorial Laplacian and the resulting Hodge decomposition. Meanwhile, discrete Morse theory has been widely used to speed up computations by reducing the size of complexes while preserving their global topological properties. In this talk, we introduce an approach to signal compression and reconstruction on complexes that leverages the tools of discrete Morse theory. The main goal is to reduce and reconstruct a cell complex together with a set of signals on its cells while preserving their global topological structure as much as possible. This is joint work with Stefania Ebli and Kelly Maggs.

The tumor microenvironment (TME) is a complex milieu around the tumor, whereby cancer cells interact with stromal, immune, vascular, and extracellular components. The TME is being increasingly recognized as a key determinant of tumor growth, disease progression, and response to therapies. We build a generalizable and robust tensor-based framework capable of integrating dissociated single-cell and spatially resolved RNA-seq data for a comprehensive analysis of the TME. Tensors are a generalization of matrices to higher dimensions. Tensor methods are known to be able to successfully incorporate data from multiple sources and perform a joint analysis of heterogeneous high-dimensional data sets. The methodologies developed as part of this effort will advance our understanding of the TME in multiple directions. These include cellular heterogeneity within the TME, crosstalks between cells, and tumor-intrinsic pathways stimulating tumor growth and immune evasion.

Here is the program.

Models of animal brains are increasingly common and mapped in increasing detail. To simplify analysis of their function, we consider subregions and show that they perform well as classifiers of overall activity, with only a fraction of the neurons. The uniqueness of such ''reliable'' regions seems to be related to the types of connections that pairs of neurons form in them. By focusing on topologically significant structures and reciprocally connected neurons we find even stronger classification results. This is ongoing work across several institutions, including EPFL, the Blue Brain Project, and the University of Aberdeen.

The recent curation of large-scale databases with 3D surface scans of shapes has motivated the development of tools that better detect global patterns in morphological variation. Studies which focus on identifying differences between shapes have been limited to simple pairwise comparisons and rely on pre-specified landmarks (that are often known). In this talk, we present SINATRA: a machine learning pipeline for analyzing collections of shapes without requiring any correspondences. Our method takes in two classes of shapes and highlights the physical features that best describe the variation between them. We develop a rigorous simulation framework to assess our approach, which themselves are a novel contribution to 3D image and shape analyses. Lastly, as case studies with real data, we use SINATRA to (1) analyze mandibular molars from four different suborders of primates and (2) facilitate the visual identification of structural signatures differentiating between the trajectories of two protein ensembles resulting from molecular dynamics simulations. Together, these results highlight a promising future for interpretable machine learning to facilitate the non-trivial task of pattern recognition in evolutionary and structural biology with substantially increased resolution.

A common problem in data science is "use this function defined over this small set to generate predictions over that larger set." Extrapolation, interpolation, statistical inference and forecasting all reduce to this problem. The Kan extension is a powerful tool in category theory that generalizes this notion. In this work we explore several applications of Kan extensions to data science. We begin by deriving simple classification and clustering algorithms as Kan extensions and experimenting with these algorithms on real data. Next, we build more complex and resilient algorithms from these simple parts.

Discrete structures have a long history of use in applied mathematics. Graphs and hypergraphs provide models of social networks, biological systems, academic collaborations, and much more. Network science, and more recently hypernetwork science, have been used to great effect in analyzing these types of discrete structures. Separately, the field of applied topology has gathered many successes through the development of persistent homology, mapper, sheaves, and other concepts. Recent work by our group has focused on the convergence of these two areas, developing and applying topological concepts to study discrete structures that model real data.

This talk will survey our body of work in this area showing our work in both the theoretical and applied spaces. Theory topics will include an introduction to hypernetwork science and its relation to traditional network science, topological interpretations of graphs and hypergraphs, and dynamics of topology and network structures. I will show examples of how we are applying each of these concepts to real data sets.

What sets the size and form of living organisms is still, by large, an open question. During this talk, I will illustrate how we are addressing this question by examining the links between spatial scales, from subcellular to organ, both experimentally and theoretically. First, I will present how we are deriving continuous plant growth mechanical models using homogenisation. Second, I will discuss how directionality of organ growth emerges from cell level. Last, I will present predictions of fluctuations at multiple scales and experimental tests of these predictions, by developing a data analysis approach that is broadly relevant to geometrically disordered materials.

What is the largest deterministic amount of time T∗ that a suitably normalized martingale X can be kept inside a convex body K in Rd? We show, in a viscosity framework, that T∗ equals the time it takes for the relative boundary of K to reach X(0) as it undergoes a geometric flow that we call (positive) minimum curvature flow. This result has close links to the literature on stochastic and game representations of geometric flows. Moreover, the minimum curvature flow can be viewed as an arrival time version of the Ambrosio–Soner codimension-(d − 1) mean curvature flow of the 1-skeleton of K. We present very preliminary sampling-based numerical approximations to the solution of the corresponding PDE. The numerical part is work in progress.

This work is based on a collaboration with Camilo Garcia Trillos, Martin Larsson, and Yufei Zhang.