02/12/2014 - 1:15pm

02/12/2014 - 2:15pm

Speaker:

Makis Kappos (Department of Mathematics, Yasar University, Izmir, Turkey)

Abstract:

Traditionally, control theory had as its main aim the stabilization of the dynamical behavior of a system, or the improvement of the transient dynamics. This is already a nontrivial problem for nonlinear systems, though the linear case is well understood and elegant. In this talk, we shall go beyond this, to address the problem of designing more complicated dynamical behavior (e.g., multiple equilibria, related among themselves in a particular way, or the creation of limit cycles.) As expected, Lyapunov functions play a role, but the analysis draws from global analysis, differential geometry and topology. We shall give an introduction to these ideas, emphasizing the possibility of bifurcation-like design.

Where:

CGU, Math South Conference Room, 710 N. College Ave

Misc. Information:

Speaker's Affiliations: Department of Mathematics, Yasar University, Izmir, Turkey, on leave from Dept of Mathematics, Aristotle University of Thessaloniki, Greece; Previous appointments include Northeastern U. and Sheffield University and one-year visits to UC Berkeley and Santa Barbara

10/02/2013 - 1:15pm

10/02/2013 - 2:15pm

Speaker:

Mike McCoy (Computing and Mathematical Sciences, Caltech)

Abstract:

Demixing is the problem of disentangling multiple informative signals from a single observation. These problems appear frequently in image processing, wireless communications, machine learning, statistics, and other data-intensive fields. Convex optimization provides a framework for creating tractable demixing procedures that work right out of the box.

In this talk, we describe a geometric theory that characterizes the performance of convex demixing methods under a generic model. This theory precisely identifies when demixing can succeed, and when it cannot, and further indicates that a sharp phase transition between success and failure is a ubiquitous feature of these programs. Our results admit an elegant interpretation: Each signal has an intrinsic dimensionality, and demixing can succeed if (and only if) the number of measurements exceeds the total dimensionality in the signal.

Where:

Davidson, CMC

10/30/2013 - 1:15pm

10/30/2013 - 2:15pm

Speaker:

Marina Chugunova (CGU)

Abstract:

We study short-time existence, long-time existence, finite speed of propagation, and finite-time blow-up of non-negative solutions for long-wave unstable thin-film equations with >0.

Where:

Davidson Lecture Hll

10/23/2013 - 1:15pm

10/23/2013 - 2:15pm

Speaker:

Alexander Kurganov (Tulane University)

Abstract:

I will first give a brief description of finite-volume, Godunov-type methods for hyperbolic

systems of conservation laws. These methods consist of two types of schemes: upwind and

central. My lecture will focus on the second type -- non-oscillatory central schemes.

Godunov-type schemes are projection-evolution methods. In these methods, the solution, at

each time step, is interpolated by a (discontinuous) piecewise polynomial interpolant, which

is then evolved to the next time level using the integral form of conservation laws. Therefore,

in order to design an upwind scheme, (generalized) Riemann problems have to be

(approximately) solved at each cell interface. This however may be hard or even impossible.

The main idea in the derivation of central schemes is to avoid solving Riemann problems by

averaging over the wave fans generated at cell interfaces. This strategy leads to a family of

universal numerical methods that can be applied as a black-box-solver to a wide variety of

hyperbolic PDEs and related problems. At the same time, central schemes suffer from

(relatively) high numerical viscosity, which can be reduced by incorporating of some

upwinding information into the scheme derivation -- this leads to central-upwind schemes,

which will be presented in the lecture.

During the talk, I will show a number of recent applications of the central schemes.

Where:

Davidson, CMC

12/04/2013 - 1:15pm

12/04/2013 - 2:15pm

Speaker:

Almut Burchard (University of Toronto)

Abstract:

Delay analysis in packet networks is notoriously hard. Statistical properties of traffic, link scheduling, and subtle correlations between traffic at different nodes increase the difficulty of characterizing the variable portion of delays. Historically, performance analysis has relied on two fundamentally different tools: Classical queuing theory (to predict delay distributions in a network where nodes operate independently and time correlations can be neglected), and worst-case analysis (to understand complex scheduling algorithms in smaller networks). Beyond that, asymptotic methods have been used to determine stability regions and exponential decay rates.

In this talk, I will discuss recent progress on the end-to-end delay analysis for a traffic flow in a packet network, using a stochastic network calculus approach that has been developed over the last twenty years. I will consider the following questions: What is the relative impact of scheduling and statistical multiplexing on delays at a packet switch? How do end-to-end delays scale as the number of traversed nodes is increased? What do self-similar and heavy-tailed traffic arrival processes contribute to the delay? (Joint work with J. Liebeherr and his group.)

Where:

Davidson Lecture Hall

11/20/2013 - 1:15pm

11/20/2013 - 2:15pm

Speaker:

Yi Grace Wang (Duke University)

Abstract:

I will describe a method for learning a piecewise affine approximation to a mapping f : R^d → R^p given a labeled training set of examples {x1, ..., xn} = X \subset R^d and targets {y1 = f(x1), ..., yn = f(xn)} = Y \subset R^p. The method first trains a binary subdivision tree that splits via hyperplanes in X corresponding to high variance directions in Y . A fixed number K of affine regressors of rank q are then trained via a K-means like iterative algorithm on the leaves of the tree. Expereiments are followed to evaluate its performance.

Where:

Davidson, CMC

11/13/2013 - 1:15pm

11/13/2013 - 2:15pm

Speaker:

Gerard Awanou (University of Illinois, Chicago)

Abstract:

Abstract: Given an orthogonal lattice with mesh length on a bounded convex domain , we show that the Aleksandrov solution of the Monge-Amp\`ere equation is the uniform limit on compact subsets of of mesh functions which solve a discrete Monge-Amp\`ere equation with the Hessian discretized using the standard finite difference method. The result provides the mathematical foundation of methods based on the standard finite difference method and designed to numerically converge to non-smooth solutions. The numerical resolution of the Monge-Amp\`ere equation has become an active research field because of several applications of increasing importance e.g. optimal transport, reflector design etc.

Where:

CMC Davidson

02/26/2014 - 1:15pm

02/26/2014 - 2:15pm

Speaker:

Blerta Shtylla (Pomona College)

Abstract:

The appropriate localization of proteins inside a cell is vital for the maintenance of cell function. Proteins can reach specific locations in a variety of ways, however directed movement frequently requires the rectification of Brownian motion. A exotic and rich variety of mechanisms can be used to generate directed movement, here we will focus on a bacterial track-altering motor complex that is employed to segregate replicated chromosome regions to specific locations in the cell. We develop and analyze various simplified mathematical models that examine how diffusion and ATP-hydrolysis mediated monomer removal can be combined to power directed movement. Using a mean first passage approach and stochastic simulations, we discuss how to calculate the effective track cleaving velocities and effective dispersion coefficient for the bacterial track-altering motor. We conclude by giving an overview of the biological implications of our model results.

Where:

CGU, Math South Conference Room, 710 N. College Ave

10/16/2013 - 1:15pm

10/16/2013 - 2:15pm

Speaker:

Xu Jing (Zhejiang Gongshang University, Currently she is visiting UCLA)

Abstract:

Abstract:

Image restoration has drawn much attention in recent years and a surge of research has been done on variational models and their numerical studies. However, there remains an urgent need to develop fast and robust methods for solving the minimization problems and the underlying nonlinear PDEs to process image of moderate to large size. We proposed a two –level domain decomposition method, which consists of an overlapping domain decomposition technique and a coarse mesh correction, for directly solving the total variational minimization problems. What’s more, the domain decomposition method hadn't been applied directly to image deblurring because of the global character of blur operator. In order to avoid separating the blur operator, we propose an algorithm for directly solving the total variational minimization problems with domain decomposition method. Various numerical experiments and comparisons demonstrate that the proposed method is efficient and fast particularly for images of large size.

Bio:

Dr. Jing Xu is an associate professor in Zhejiang Gongshang University, Hangzhou, China. Now she is a visiting scholar of Professor Luminita Vese in mathematics in UCLA from July 1, 2013 to June 30, 2014. Her research major is the applications of partial differential equations to image processing. She got her PHD degree in applied mathematics institute of Chinese Academy Science, 2007. In that time, she studied the multigrid method in image restoration from Professor Qianshun Chang. Jing Xu has one year experience being a research fellow in Nanyang Technology University, Singapore from 2008 to 2009. During that time, she studied domain decomposition method from Professor Xuecheng Tai and lilian Wang.

Where:

CMC Campus, Adams Hall, Davidson (the largest lecture room on the first floor)

09/18/2013 - 1:15pm

09/18/2013 - 2:15pm

Speaker:

Mark Huber (CMC)

Abstract:

High dimensional integrals and sums arise often in applications coming from statistics, theoretical physics, and computer science. Consider the problem of approximating these integrals to within a factor of for some specified when the integrand (or summand) is nonnegative.. Monte Carlo methods for approximating these integrals work by drawing samples randomly from a family of distributions formed from the nonnegative integrand. Typically the value of these integrals is exponentially large in the problem input size, so let denote the natural logarithm of the integral in question. Previous Monte Carlo methods for such approximation used samples (suppressing the dependence on ), but the big-O notation hid factors on the order of .. In this work, a new method is presented that uses only samples, with a factor of about 16.7. As a bonus, this method is much simpler to implement then previous approaches. This is the first practical method for creating provably good approximations for these problems.

Where:

CMC Campus, Adams Hall, Davidson (the largest lecture room on the first floor)