Upcoming Seminars

Title: What in the structure of data make them learnable?
Seminar: Algebra
Speaker: Matthieu Wyart of EPFL
Contact: Matthias Chung, matthias.chung@emory.edu
Date: 2023-12-04 at 11:30AM
Venue: N215
Download Flyer
Abstract:
Deep learning algorithms have achieved remarkable successes, yet why they work is unclear. Notably, they can learn many high-dimensional tasks, a feat generically infeasible due to the so-called curse of dimensionality. What is the structure of data that makes them learnable, and how this structure is exploited by deep neural networks, is a central question of the field. In the absence of an answer, relevant quantities such as the number of training data needed to learn a given task -the sample complexity- cannot be determined. I will show how deep neural networks trained with gradient descent can beat the curse of dimensionality when the task is hierarchically compositional, by building a good representation of the data that effectively lowers the dimension of the problem. This analysis also reveals how the sample complexity is affected by the hierarchical nature of the task. If time permits, I will also discuss how the fact that regions in the data containing information on the task can be sparse affects sample complexity.
Title: Perfect Cuboids and Magic Squares of Squares
Colloquium: Algebra
Speaker: Tony Várilly-Alvarado of Rice University
Contact: Andrew Kobin, ajkobin@emory.edu
Date: 2023-12-04 at 3:00PM
Venue: Atwood 240
Download Flyer
Abstract:
A perfect cuboid is a box such that the distance between any two corners is a positive integer. A magic square is a grid filled with distinct positive integers, whose rows, columns, and diagonals add up to the same number. To date, we don't know if there exists a perfect cuboid, or a 3 x 3 magic square whose entries are distinct squares. What do these problems have in common? Secretly, they are both problems about rational points on algebraic surfaces of general type with mild singularities. I believe there is no such thing as a perfect cuboid or a 3 x 3 magic square of squares, and I will try to convince you that geometry suggests this is so.
Title: Predicting Complex Spatiotemporal Cardiac Voltage Dynamics Using Reservoir Computing
Seminar: Numerical Analysis and Scientific Computing
Speaker: Elizabeth Cherry of Georgia Tech
Contact: Matthias Chung, matthias.chung@emory.edu
Date: 2023-12-05 at 10:00AM
Venue: MSC N306
Download Flyer
Abstract:
Disruptions to the electrical behavior of the heart caused by cardiac arrhythmias can result in complex dynamics, from period-2 rhythms in single cells to spatiotemporally complex spiral and scroll waves of electrical activity, which can inhibit contraction and may be lethal if untreated. Accurate forecasts of cardiac voltage behavior could allow new opportunities for intervention and control, but predicting complex nonlinear time series is a challenging task. In this talk, we discuss our recent work using machine-learning approaches based on reservoir computing to forecast cardiac voltage dynamics. First, we show that a novel method combining an echo state network with automated feature extraction via an autoencoder can successfully and efficiently predict time series of synthetic and experimental datasets of cardiac voltage in one cell with 20-30 action potentials in advance. Building on this work, we then demonstrate a novel method for predicting the complex spatiotemporal electrical dynamics of cardiac tissue using an echo state network integrated with a convolutional autoencoder. We show that our approach can forecast complex spiral-wave behavior, including breakup several periods in advance for time series ranging from model-derived synthetic datasets to optical-mapping recordings of explanted human hearts.
Title: On the prime Selmer ranks of cyclic prime twist families of elliptic curves over global function fields
Seminar: Algebra
Speaker: Sun Woo Park of University of Wisconsin
Contact: Andrew Kobin, ajkobin@emory.edu
Date: 2023-12-05 at 4:00PM
Venue: MSC W301
Download Flyer
Abstract:
Fix a prime number $p$. Let $\mathbb{F}_q$ be a finite field of characteristic coprime to 2, 3, and $p$, which also contains the primitive $p$-th root of unity $\mu_p$. Based on the works by Swinnerton-Dyer, Klagsbrun, Mazur, and Rubin, we prove that the probability distribution of the sizes of prime Selmer groups over a family of cyclic prime twists of non-isotrivial elliptic curves over $\mathbb{F}_q(t)$ satisfying a number of mild constraints conforms to the distribution conjectured by Bhargava, Kane, Lenstra, Poonen, and Rains with explicit error bounds. The key tools used in proving these results are the Riemann hypothesis over global function fields, the Erd\"os-Kac theorem, and the geometric ergodicity of Markov chains.
Title: A shifted convolution problem arising from physics
Seminar: Algebra
Speaker: Kim Klinger-Logan of Kansas State University/Rutgers University
Contact: Andrew Kobin, andrew.jon.kobin@emory.edu
Date: 2023-12-12 at 4:00PM
Venue: MSC W301
Download Flyer
Abstract:
Physicists Green, Russo, and Vanhove have discovered solution to differential equations involving automorphic forms appear at the coefficients to the $4$-graviton scattering amplitude in type IIB string theory. Specifically, for $\Delta$ the Laplace-Beltrami operator and $E_s(g)$ a Langlands Eisenstein series, solutions $f(g)$ of $(\Delta-\lambda) f(g) = E_a(g) E_b(g)$ for $a$ and $b$ half-integers on certain moduli spaces $G(Z)\backslash G(R)/K(R)$ of real Lie groups appear as coefficients to the analytic expansion of the scattering amplitude. We will briefly discuss different approaches to finding solutions to such equations and focus on a shifted convolution sum of divisor functions which appears as the Fourier modes associated to the homogeneous part of the solution. Initially, it was thought that, when summing over all Fourier modes, the homogeneous solution would vanish but recently we have found an exciting error term. This is joint work with Stephen D. Miller, Danylo Radchenko, and Ksenia Fedosova.