Seminars and Colloquia by Series

Vanishing cycles and almost toric fibrations by Jie Min

Series
Geometry Topology Seminar
Time
Monday, March 10, 2025 - 14:00 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
Jie MinUniversity of Massachusetts Amherst

Vanishing cycles of Lefschetz fibrations give examples of Lagrangian spheres in the fiber. A natural question, first raised by Donaldson, is whether all Lagrangian spheres arise this way. We focus on this problem for positive rational surfaces, which were shown to admit a geometric structure called almost toric fibrations. I will talk about a work-in-progress showing all Lagrangian spheres here are visible in an almost toric fibration and thus are vanishing cycles of a nodal degeneration.

Forbidden Minor Results for Flag Matroids

Series
Algebra Seminar
Time
Monday, March 10, 2025 - 13:00 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Nathaniel VaduthalaTulane University

Please Note: There will be a pre-seminar from 10:55 to 11:15 in Skiles 005.

Similar to how matroids can be viewed as a combinatorial abstraction of linear subspaces, a flag matroid can be viewed as a combinatorial abstraction of a nested sequence of linear subspaces. In this talk, we will discuss forbidden minor results that describe precisely when a flag matroid is representable and when it is graphic. 

Mathapalooza!

Series
Other Talks
Time
Sunday, March 9, 2025 - 15:00 for 4 hours (half day)
Location
Sun ATL Gallery, 399 Edgewood Ave SE, Atlanta, GA 30312
Speaker
Many

Please Note: Mathapalooza! is a math-themed event at the Atlanta Science Festival. If you want to volunteer to help on March 9, please write to Evans Harrell

The 2025 Atlanta Science Festival Mathapalooza! will take place in an art gallery at the opening of an exhibition, The Art of Math.  The show will begin with magic by Matt Baker, followed by hands-on art construction, circus acts, "Math Court" skits, and 4-dimensional dance, before ending with a performance by Tracy Woodard of Bach's Crab Canon, with a discussion of its symmetry algebra.  Find out more and get tickets at https://atlantasciencefestival.org/events-2025/1113-mathapalooza-at-the-gallery/

On the original Ulam's problem and its quantization

Series
CDSNS Colloquium
Time
Friday, March 7, 2025 - 15:30 for 1 hour (actually 50 minutes)
Location
Skiles 314
Speaker
Jing ZhouGreat Bay University (Dongguan)

Please Note: Zoom link: https://gatech.zoom.us/j/91390791493?pwd=QnpaWHNEOHZTVXlZSXFkYTJ0b0Q0UT09

In 1961 Ulam proposed a mathematical simplification extracting the essential accelerating mechanism proposed by Fermi, as to explain the cause of high-energy particles in cosmic rays. In this talk, we shall describe the typical behavior of the very model introduced by Ulam in both the classical original form as well as its quantization. In the classical model, we show that typical orbits are recurrent under resonance assumptions. Meanwhile in the quantum model, the acceleration caused by resonance gets much amplified and we point out a direct relationship between the acceleration behavior of the system and the shape of its quasi-energy spectrum. This is a joint work in progress with Changguang Dong and Disheng Xu.

Unsupervised Solution Operator Learning for Mean-Field Games

Series
Applied and Computational Mathematics Seminar
Time
Friday, March 7, 2025 - 11:00 for 1 hour (actually 50 minutes)
Location
Skiles 006 and https://gatech.zoom.us/j/98355006347
Speaker
Rongjie LaiPurdue University

Recent advances in deep learning have introduced numerous innovative frameworks for solving high-dimensional mean-field games (MFGs). However, these methods are often limited to solving single-instance MFGs and require extensive computational time for each instance, presenting challenges for practical applications.

In this talk, I will present our recent work on a novel framework for learning the MFG solution operator. Our model takes MFG instances as input and directly outputs their solutions in a single forward pass, significantly improving computational efficiency. Our method offers two key advantages: (1) it is discretization-free, making it particularly effective for high-dimensional MFGs, and (2) it can be trained without requiring supervised labels, thereby reducing the computational burden of preparing training datasets common in existing operator learning methods. If time permits, I will also explore connections between this framework and in-context learning, highlighting its broader implications and potential for further advancements.

 

Why are the logits of trained models distorted? A theory of overfitting for imbalanced classification

Series
Stochastics Seminar
Time
Thursday, March 6, 2025 - 15:30 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
Yiqiao ZhongUniversity of Wisconsin–Madison

Data imbalance is a fundamental challenge in data analysis, where minority classes account for a small fraction of the training data compared to majority classes. Many existing techniques attempt to compensate for the underrepresentation of minority classes, which are often critical in applications such as rare disease detection and anomaly detection. Notably, in empirical deep learning, the large model size exacerbates the issue. However, despite extensive empirical heuristics, the statistical foundations of these methods remain underdeveloped, which poses an issue to the reliability of these machine learning models.

In this talk, I will examine imbalanced classification problems in high dimensions, focusing on support vector machine (SVMs) and logistic regression. I will introduce a "truncation" phenomenon---which we verifed across single-cell tabular data, image data, and text data---where overfitting in high dimensions distorts the distribution of logits on training data. I will provide a theoretical foundation by characterizing the asymptotic distribution via a variational formulation. This analysis formalizes the intuition that overfitting disproportionately harms minority classes and reveals how margin rebalancing---a widely used deep learning heuristic---mitigates data imbalance. As a consequence, the theory offers both qualitative and quantitative insights into generalization errors and uncertainty measures such as calibration.

This talk is based on a joint work with Jingyang Lyu (3rd-year Stats PhD student) and Kangjie Zhou (Columbia Statistics): arXiv:2502.11323.

Towards an algorithmic model of the neuron for Neuroscience and AI

Series
School of Mathematics Colloquium
Time
Thursday, March 6, 2025 - 11:00 for 1 hour (actually 50 minutes)
Location
Skiles 005 and Zoom: https://gatech.zoom.us/j/98474702488?pwd=2CiHNben05BqfpbikKkCuzzdr0MjdZ.1
Speaker
Dmitri Chklovskii NYU and the Flatiron Institute

Modern Artificial Intelligence (AI) systems, such as ChatGPT, rely on artificial neural networks (ANNs), which are historically inspired by the human brain. Despite this inspiration, the similarity between ANNs and biological neural networks is largely superficial. For instance, the foundational McCulloch-Pitts-Rosenblatt unit of ANNs drastically oversimplifies the complexity of real neurons.Recognizing the intricate temporal dynamics in biological neurons and the ubiquity of feedback loops in natural networks, we suggest reimagining neurons as feedback controllers. A practical implementation of such controllers within biological systems is made feasible by the recently developed Direct Data-Driven Control (DD-DC). We find that DD-DC neuron models can explain various neurophysiological observations, affirming our theory.

Dual Lyapunov exponents and sharp arithmetic spectral transitions for quasiperiodic operators

Series
School of Mathematics Colloquium
Time
Thursday, March 6, 2025 - 09:30 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Svetlana JitomirskayaUC Berkeley

We will describe a recently discovered object, dual Lyapunov exponents, that has emerged as a powerful tool in the spectral analysis of  quasiperiodic operators with analytic potentials, leading to solutions of several long outstanding problems. Based on papers joint with L. Ge, J. You, and Q. Zhou

A retract of a Banach manifold is a Banach manifold

Series
Geometry Topology Student Seminar
Time
Wednesday, March 5, 2025 - 14:00 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
John StavroulakisGeorgia Tech

We discuss the proof of the following Theorem

 

Assume $E$ is a $C^{p}$ real Banach manifold, and $f:E\circlearrowleft$, $f\circ f=f$ is a $C^{p}$ retraction, where $1\leq p\leq +\infty$. Then the retract $f(E)$ is a $C^{p}$ sub Banach manifold of $E$.

 

If time allows, we will also discuss how this fact is related to the study of smoothness and structural stability of attractors, along the intersection of topology and dynamics. We will be focusing on the proof and perspective of Oliva 1975, who was interested in Banach manifolds as phase-spaces of delay equations.

Construction of multi-soliton solutions for semilinear equations in dimension 3

Series
PDE Seminar
Time
Tuesday, March 4, 2025 - 15:30 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Istvan KadarPrinceton University

The existence of multi black hole solutions in General Relativity is one of the expectations from the final state conjecture, the analogue of soliton resolution. In this talk, I will present preliminary works in this direction via a semilinear model, the energy critical wave equation, in dimension 3. In particular, I show 1) an algorithm to construct approximate solutions to the energy critical wave equation that converge to a sum of solitons at an arbitrary polynomial rate in (t-r); 2) a robust method to solve the remaining error terms for the nonlinear equation. The methods apply to energy supercritical problems.

Pages