Upcoming MAMS Seminar Series

Spring 2026

2/17/2026, Tue. 2:30-3:30 pm in Rockefeller 303
Speaker:
Fabio Isaza (Case Western Reserve University)
Title:
Depth, Compositional Structure, and the Curse of Dimensionality in ReLU Networks
Abstract: Classical approximation theory shows that generic smooth functions on [0, 1]^d require a number of parameters that grows exponentially with dimension, reflecting the so-called “curse of dimensionality”. I will discuss recent results of Poggio and collaborators showing that deep ReLU neural networks can avoid this phenomenon for hierarchically compositional function classes, achieving approximation rates governed by intrinsic structural dimension rather than ambient dimension. These results clarify that the expressive advantage of depth in neural networks arises from its ability to exploit compositional structure, rather than from universality alone.

2/10/2026, Tue. 2:30-3:30 pm in Rockefeller 303
Speaker:
Bartłomiej Zawalski (Case Western Reserve University)
Title: 
On the homothety conjecture for the body of flotation and the body of buoyancy on a plane
Abstract: Let K be a convex body in R^n, i.e., a compact convex set with non-empty interior. For any $\delta \in (0, vol_n(K))$, the convex body of flotation of K is the set of points that remain above the water level when a solid body of shape K and uniform density $\delta / vol_n(K)$ floats in any orientation.

For each orientation, the centroid of the underwater part is known as the center of buoyancy. The hypersurface of buoyancy is defined as the geometric locus of all the centers of buoyancy. The hypersurface of buoyancy encloses a strongly convex body $B_\delta(K)$, called the body of buoyancy. We will investigate questions related to the homothety conjecture [3] for convex bodies on a plane, which asks whether an ellipse is the only convex body K that is homothetic to the body of flotation $F_\delta(K)$ for some $\delta \in (0, vol_n(K))$. We systematically derive all the fundamental properties of bodies of flotation and bodies of buoyancy and show that if the body of flotation is homothetic to the body of buoyancy, and if every chord of flotation cuts off from the boundary exactly 1/3 of its total affine arc length, then K is an ellipse. As a result, we get an affine counterpart of Zindler carousels introduced by J. Bracho, L. Montejano, and D. Oliveros [2]. Our result is also related to the classical floating body problem of Ulam [4, 19. Problem: Ulam] from the Scottish Book, and the work of H. Auerbach [1].

2/3/2026, Tue. 2:30-3:30 pm in Rockefeller 303
Speaker:
Mengchun Cai (Case Western Reserve University)
Title:
Convergence Rate for the Least Eigenvalue of the Laguerre Unitary Ensemble
Abstract: In this talk, we will derive a rate of convergence regarding the least eigenvalue of the Laguerre Unitary Ensemble (LUE) by discussing its corresponding point process. We will show that with a rate of order $N^\frac{2}{3}$, the scaled least eigenvalue of an LUE converges toward the Tracy-Widom law in distribution. To this end, we will establish necessary large $N$ expansions for the Laguerre polynomial by the approximation of the Whittaker function. Some tools will be introduced in this talk to facilitate the analysis of the Whittaker equation.

Spring 2026

2/18/2026, Wed. 4-5 pm in Sear 439
Speaker:
Dr. Doosung Choi (Case Western Reserve University)
Title:
Mathematical modeling of neutral Inclusions and shape recovery
Abstract: This talk is about inverse conductivity imaging: how to recover an inclusion’s shape and material contrast from boundary (multistatic) measurements, and how to design inclusions that are effectively “invisible.” We use generalized polarization tensors (GPTs) to link measurement data to geometry through multipole expansions, then reconstruct the target boundary in the planar case by recovering the exterior conformal map via Faber polynomials and Grunsky coefficients. This viewpoint leads to concrete formulas for estimating both conductivity and shape from low-order GPTs. We also discuss neutral and semi-neutral inclusions, where the relevant polarization information (in particular, Faber polynomial polarization tensors) vanishes to a prescribed order. Two constructions are highlighted: multi-coated structures obtained as conformal images of concentric annuli, and imperfect interface models that can enforce vanishing conditions. Numerical examples illustrate how these ideas work in practice and how coatings or interface effects can significantly suppress the perturbation of the background field for non-circular targets.

Spring 2026

Spring 2026

2/16/2026, Mon. 2:15 -3:15 pm in Sears 439
Speaker: Reeve Johnson (Case Western Reserve University)
Title: Virtual double categories
Abstract: Multicategories have been generalized in a variety of ways, with two prominent approaches being the span construction and the profunctor construction. These generalizations can be placed into a clean unifying framework: monads on virtual double categories. In this talk, we will learn what a virtual double category is in relation to enriched or internal categories (You will get to pick! A choose your own adventure talk!) and then we will discover how this perspective provides a template for understanding generalized multicategories.

Spring 2026

2/6/2026, Fri. 3:30-4:30 pm in Sears 541
Speaker:
Jake Hinds (Case Western Reserve University)
Title: A fine way to do Linear Algebra without matrices
Abstract: Did you feel like Shiv’s talk had TOO many matrices? Were you annoyed that Shiv didn’t fully explain what Affine transformation means? Do you wish to shear but do not own sheep? This talk will walk you through an undergraduate research project that at one point you will realize this is just linear algebra from a mathematician who refuses to use a matrix at every turn, cause something something euclidean geometry doesn’t need a grid. Said mathematician is me.

1/30/2026, Fri. 3:30-4:30 pm in Nord 356
Speaker:
Saravana Mauree (Case Western Reserve University)
Title: How I solved racism using Calculus and Linear Algebra
Abstract: This talk will walk you through an undergraduate project that started off as a sign language translation app and unexpectedly became a “social justice movement “(extreme hyperbole). The problem ultimately boiled down to classifying points in a finite Euclidean space using composition of linear maps and nonlinearity optimization.

Using this case study, I will explore one of the most popular AI systems and argue that, at its core, is nothing more than a thoughtful application of very classical tools from multivariate calculus and linear algebra.