Seminars and Colloquia by Series

Flows (and group-connectivity) in signed graphs

Series
Graph Theory Seminar
Time
Tuesday, April 18, 2023 - 15:45 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Jessica McDonaldAuburn University

We discuss flows (and group-connectivity) in signed graphs, and prove a new result about group-connectivity in 3-edge-connected signed graphs. This is joint work with Alejandra Brewer Castano and Kathryn Nurse.

Global well-posedness for the one-phase Muskat problem

Series
PDE Seminar
Time
Tuesday, April 18, 2023 - 15:00 for
Location
Skiles 006
Speaker
Huy NguyenUniversity of Maryland, College Park

 

We will discuss the one-phase Muskat problem concerning the free boundary of Darcy fluids in porous media. It is known that there exists a class of non-graph initial boundary leading to self-intersection at a single point in finite time (splash singularity). On the other hand, we prove that the problem has a unique global-in-time solution if the initial boundary is a periodic Lipschitz graph of arbitrary size. This is based on joint work with H. Dong and F. Gancedo. 

Symplectic trisections and connected sum decompositions

Series
Geometry Topology Seminar
Time
Monday, April 17, 2023 - 16:30 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
Peter Lambert-ColeUniversity of Georgia

This talk will have two parts.  The first half will describe how to construct symplectic structures on trisected 4-manifolds. This construction is inspired by projective complex geometry and completely characterizes symplectic 4-manifolds among all smooth 4-manifolds.  The second half will address a curious phenomenon: symplectic 4-manifolds appear to not admit any interesting connected sum decompositions.  One potential explanation is that every embedded 3-sphere can be made contact-type.  I will outline some strategies to prove this from a trisections perspective, describe some of the obstructions, and give evidence that these obstructions may be overcome.

Jones diameter and crossing numbers of satellite knots

Series
Geometry Topology Seminar
Time
Monday, April 17, 2023 - 15:00 for 1 hour (actually 50 minutes)
Location
Speaker
Effie KalfagianniMichigan State University
It has been long known that the quadratic term in the degree of the colored Jones polynomial of knot provides a lower bound of the crossing number the knot.
I’ll discuss work with Lee where we determine the class of knots for which this bound is sharp and give applications to computing crossing numbers of satellite knots.
 

Uncovering the Law of Data Separation in Deep Learning

Series
Applied and Computational Mathematics Seminar
Time
Monday, April 17, 2023 - 14:00 for 1 hour (actually 50 minutes)
Location
Skiles 005 and https://gatech.zoom.us/j/98355006347
Speaker
Prof. Weijie SuUniversity of Pennsylvania (Wharton)

Please Note: The speaker will present in person.

In this talk, we will investigate the emergence of geometric patterns in well-trained deep learning models by making use of a layer-peeled model and the law of equi-separation. The former is a nonconvex optimization program that models the last-layer features and weights. We use the model to shed light on the neural collapse phenomenon of Papyan, Han, and Donoho, and to predict a hitherto-unknown phenomenon that we term minority collapse in imbalanced training.
 
The law of equi-separation is a pervasive empirical phenomenon that describes how data are separated according to their class membership from the bottom to the top layer in a well-trained neural network. We will show that, through extensive computational experiments, neural networks improve data separation through layers in a simple exponential manner. This law leads to roughly equal ratios of separation that a single layer is able to improve, thereby showing that all layers are created equal. We will conclude the talk by discussing the implications of this law on the interpretation, robustness, and generalization of deep learning, as well as on the inadequacy of some existing approaches toward demystifying deep learning.
 

Invariants of Matrices

Series
Algebra Seminar
Time
Monday, April 17, 2023 - 10:20 for 1.5 hours (actually 80 minutes)
Location
Skiles 005
Speaker
Harm DerksenNortheastern University

The group SL(n) x SL(n) acts on m-tuples of n x n matrices by simultaneous left-right multiplication.  Visu Makam and the presenter showed the ring of invariants is generated by invariants of degree at most mn^4. We will also discuss geometric aspects of this action and connections to algebraic complexity and the notion of noncommutative rank.

Two graph classes with bounded chromatic number

Series
Dissertation Defense
Time
Monday, April 17, 2023 - 09:30 for 1 hour (actually 50 minutes)
Location
Skiles 114 (or Zoom)
Speaker
Joshua SchroederGeorgia Tech

Please Note: Zoom: https://gatech.zoom.us/j/98256586748?pwd=SkJLZ3ZKcjZsM0JkbGdyZ1Y3Tk9udz09 Meeting ID: 982 5658 6748 Password: 929165

A class of graphs is said to be $\chi$-bounded with binding function $f$ if for every such graph $G$, it satisfies $\chi(G) \leq f(\omega(G)$, and polynomially $\chi$-bounded if $f$ is a polynomial. It was conjectured that chair-free graphs are perfectly divisible, and hence admit a quadratic $\chi$-binding function. In addition to confirming that chair-free graphs admit a quadratic $\chi$-binding function, we will extend the result by demonstrating that $t$-broom free graphs are polynomially $\chi$-bounded for any $t$ with binding function $f(\omega) = O(\omega^{t+1})$. A class of graphs is said to satisfy the Vizing bound if it admits the $\chi$-binding function $f(\omega) = \omega + 1$. It was conjectured that (fork, $K_3$)-free graphs would be 3-colorable, where fork is the graph obtained from $K_{1, 4}$ by subdividing two edges. This would also imply that (paw, fork)-free graphs satisfy the Vizing bound. We will prove this conjecture through a series of lemmas that constrain the structure of any minimal counterexample.

Meeting on Applied Algebraic Geometry

Series
Time
Saturday, April 15, 2023 - 09:15 for 8 hours (full day)
Location
Skiles 005/006 and Atrium
Speaker

The Meeting on Applied Algebraic Geometry (MAAG 2023) is a regional gathering which attracts participants primarily from the South-East of the United States. Previous meetings took place at Georgia Tech in 2015, 2018, and 2019, and at Clemson in 2016.

For more information and to register, please visit https://sites.google.com/view/maag-2023. The registration is free until February 28th, 2023, and the registration fee will become $50 after that. 

MAAG will be followed by a Macaulay2 Day on April 16.

Organizers: Abeer Al Ahmadieh, Greg Blekherman, Anton Leykin, and Josephine Yu.

Toward algorithms for linear response and sampling

Series
CDSNS Colloquium
Time
Friday, April 14, 2023 - 15:30 for 1 hour (actually 50 minutes)
Location
Skiles 006 and Online
Speaker
Nisha ChandramoorthyGeorgia Tech

Zoom Link: Link: https://gatech.zoom.us/j/91390791493?pwd=QnpaWHNEOHZTVXlZSXFkYTJ0b0Q0UT09

Abstract: Linear response refers to the smooth change in the statistics of an observable in a dynamical system in response to a smooth parameter change in the dynamics. The computation of linear response has been a challenge, despite work pioneered by Ruelle giving a rigorous formula in Anosov systems. This is because typical linear perturbation-based methods are not applicable due to their instability in chaotic systems. Here, we give a new differentiable splitting of the parameter perturbation vector field, which leaves the resulting split Ruelle's formula amenable to efficient computation. A key ingredient of the overall algorithm, called space-split sensitivity, is a new recursive method to differentiate quantities along the unstable manifold.

In the second part, we discuss a new KAM method-inspired construction of transport maps. Transport maps are transformations between the sample space of a source (which is generally easy to sample) and a target (typically non-Gaussian) probability distribution. The new construction arises from an infinite-dimensional generalization of a Newton method to find the zero of a "score operator". We define such a score operator that gives the difference of the score -- gradient of logarithm of density -- of a transported distribution from the target score. The new construction is iterative, enjoys fast convergence under smoothness assumptions, and does not make a parametric ansatz on the transport map.

Pages