Seminars and Colloquia by Series

Sparse Singular Value Decomposition in High Dimensions

Series
Stochastics Seminar
Time
Tuesday, April 24, 2012 - 16:05 for 1 hour (actually 50 minutes)
Location
skyles 006
Speaker
Zongming Ma – The Wharton School, Department of Statistics, University of Pennsylvania
Singular value decomposition is a widely used tool for dimension reduction in multivariate analysis. However, when used for statistical estimation in high-dimensional low rank matrix models, singular vectors of the noise-corrupted matrix are inconsistent for their counterparts of the true mean matrix. In this talk, we suppose the true singular vectors have sparse representations in a certain basis. We propose an iterative thresholding algorithm that can estimate the subspaces spanned by leading left and right singular vectors and also the true mean matrix optimally under Gaussian assumption. We further turn the algorithm into a practical methodology that is fast, data-driven and robust to heavy-tailed noises. Simulations and a real data example further show its competitive performance. This is a joint work with Andreas Buja and Dan Yang.

Mathematics of Crime

Series
School of Mathematics Colloquium
Time
Tuesday, April 24, 2012 - 16:00 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Prof. Andrea Bertozzi – UCLA Math
There is an extensive applied mathematics literature developed for problems in the biological and physical sciences. Our understanding of social science problems from a mathematical standpoint is less developed, but also presents some very interesting problems. This lecture uses crime as a case study for using applied mathematical techniques in a social science application and covers a variety of mathematical methods that are applicable to such problems. We will review recent work on agent based models, methods in linear and nonlinear partial differential equations, variational methods for inverse problems and statistical point process models. From an application standpoint we will look at problems in residential burglaries and gang crimes. Examples will consider both "bottom up" and "top down" approaches to understanding the mathematics of crime, and how the two approaches could converge to a unifying theory.

Overconvergent Lattices and Berkovich Spaces

Series
Algebra Seminar
Time
Tuesday, April 24, 2012 - 14:00 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
Andrew Dudzik – UC Berkeley
The construction of the Berkovich space associated to a rigid analytic variety can be understood in a general topological framework as a type of local compactification or uniform completion, and more generally in terms of filters on a lattice. I will discuss this viewpoint, as well as connections to Huber's theory of adic spaces, and draw parallels with the usual metric completion of $\mathbb{Q}$.

Universality of the global fluctuations for the eigenvectors of Wigner random matrices

Series
Stochastics Seminar
Time
Tuesday, April 24, 2012 - 11:00 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
F. Benaych-Georges – Universite Pierre et Marie Curie
Many of the asymptotic spectral characteristics of a symmetric random matrix with i.i.d. entries (such a matrix is called a "Wigner matrix") are said to be "universal": they depend on the exact distribution of the entries only via its first moments (in the same way that the CLT gives the asymptotic fluctuations of the empirical mean of i.i.d. variables as a function of their second moment only). For example, the empirical spectral law of the eigenvalues of a Wigner matrix converges to the semi-circle law if the entries have variance 1, and the extreme eigenvalues converge to -2 and 2 if the entries have a finite fourth moment. This talk will be devoted to a "universality result" for the eigenvectors of such a matrix. We shall prove that the asymptotic global fluctuations of these eigenvectors depend essentially on the moments with orders 1, 2 and 4 of the entries of the Wigner matrix, the third moment having surprisingly no influence.

Perturbation Theory and its Application to Complex Biological Networks --A quantification of systematic features of biological networks

Series
Dissertation Defense
Time
Tuesday, April 24, 2012 - 11:00 for 2 hours
Location
Skiles 006
Speaker
Yao Li – School of Mathematics, Georgia Tech
The primary objective of this thesis is to make a quantitative study of complex biological networks. Our fundamental motivation is to obtain the statistical dependency between modules by injecting external noise. To accomplish this, a deep study of stochastic dynamical systems would be essential. The first part is about the stochastic dynamical system theory. The classical estimation of invariant measures of Fokker-Planck equations is improved by the level set method. Further, we develop a discrete Fokker-Planck-type equation to study the discrete stochastic dynamical systems. In the second part, we quantify systematic measures including degeneracy, complexity and robustness. We also provide a series of results on their properties and the connection between them. Then we apply our theory to the JAK-STAT signaling pathway network.

Optimization of two-link and three-link snake-like locomotion

Series
Applied and Computational Mathematics Seminar
Time
Monday, April 23, 2012 - 14:00 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
Fangxu Jing – GT Math
We analyze two-link (or three-link) 2D snake like locomotions and discuss the optimization of the motion. The snake is modeled as two (or three) identical links connected via hinge joints and the relative angles between the links are prescribed as periodic actuation functions. An essential feature of the locomotion is the anisotropy of friction coefficients. The dynamics of the snake is analyzed numerically, as well as analytically for small amplitude actuations of the relative angles. Cost of locomotion is defined as the ratio between distance traveled by the snake and the energy expenditure within one period. Optimal conditions of the highest efficiency in terms of the friction coefficients and the actuations are discussed for the model.

On a weak form of Arnold diffusion in arbitrary degrees of freedom

Series
CDSNS Colloquium
Time
Monday, April 23, 2012 - 11:05 for 1 hour (actually 50 minutes)
Location
Skiles 006
Speaker
Vadim Kaloshin – Univ. of Maryland
Consider a generic perturbation of a nearly integrable system of {\it arbitrary degrees of freedom $n\ge 2$ system}\[H_0(p)+\eps H_1(\th,p,t),\quad \th\in \T^n,\ p\in B^n,\ t\in \T=\R/\T,\]with strictly convex $H_0$. Jointly with P.Bernard and K.Zhang we prove existence of orbits $(\th,p)(t)$ exhibiting Arnold diffusion \[\|p(t)-p(0) \| >l(H_1)>0 \quad \textup{independently of }\eps.\]Action increment is independent of size of perturbation$\eps$, but does depend on a perturbation $\eps H_1$.This establishes a weak form of Arnold diffusion. The main difficulty in getting rid of $l(H_1)$ is presence of strong double resonances. In this case for $n=2$we prove existence of normally hyperbolic invariant manifolds passing through these double resonances. (joint with P. Bernard and K. Zhang)

Discrete Mathematical Biology Working Seminar

Series
Other Talks
Time
Monday, April 23, 2012 - 11:00 for 1 hour (actually 50 minutes)
Location
Skiles 114
Speaker
Will Perkins – Georgia Tech
A discussion of the paper "RNA folding with soft constraints: reconciliation of probing data and thermodynamic secondary structure prediction" by Washietl et al (NAR, 2012).

Matchings in hypergraphs

Series
Combinatorics Seminar
Time
Friday, April 20, 2012 - 15:05 for 1 hour (actually 50 minutes)
Location
Skiles 005
Speaker
Tomasz Luczak – Emory University and Adam Mickiewicz University, Poznan –
Let H_k(n,s) be a k-uniform hypergraphs on n vertices in which the largest matching has s edges. In 1965 Erdos conjectured that the maximum number of edges in H_k(n,s) is attained either when H_k(n,s) is a clique of size ks+k-1, or when the set of edges of H_k(n,s) consists of all k-element sets which intersect some given set S of s elements. In the talk we prove this conjecture for k = 3 and n large enough. This is a joint work with Katarzyna Mieczkowska.

Pages