Past

The standard rounding procedure in floating-point computations is round to nearest (RN). In this talk we consider an alternative rounding strategy called stochastic rounding (SR) which has the appealing property of being exact (actually exact!) in expectation. In the first part of the talk we discuss recent developments in probabilistic rounding error analysis and we show how rounding errors grow at an O(\sqrt{n}) rate rather than O(n) when SR is employed. This shows that Wilkinson's rule of thumb provably holds for this type of rounding. In the second part of the talk we consider the application of SR to parabolic PDEs admitting a steady state solution. We show that when the heat equation is solved in half precision RN fails to compute an accurate solution, while SR successfully solves the problem to decent accuracy.
 

I will explain the basics of 2-representation theory and will explain an approach to classifying 'simple' 2-representations of the Hecke 2-category (aka Soergel bimodules) for finite Coxeter types.

Consider all planar graphs on n vertices. What is the smallest graph that contains them all as induced subgraphs? We provide an explicit construction of such a graph on $n^{4/3+o(1)}$ vertices, which improves upon the previous best upper bound of $n^{2+o(1)}$, obtained in 2007 by Gavoille and Labourel.

In this talk, we will gently introduce the audience to the notion of so-called universal graphs (graphs containing all graphs of a given family as induced subgraphs), and devote some time to a key lemma in the proof. That lemma comes from a recent breakthrough by Dujmovic et al. regarding the structure of planar graphs, and has already many interesting consequences - we hope the audience will be able to derive more. This is based on joint work with Cyril Gavoille and Michal Pilipczuk.

We consider the problem of dynamically pricing multiple products on a network of resources, such as that faced by an airline selling tickets on its route network. For computational reasons this inherently stochastic problem is often approximated deterministically, however even the deterministic dynamic pricing problem can be impractical to solve. For this reason we have derived a novel iterative Quadratic Programming approximation to the deterministic dynamic pricing problem that is not only very fast to solve in practice but also exhibits a provably linear rate of convergence. This is joint work with Saksham Jain and Raphael Hauser.
 

Spatial networks are often the most natural way to represent spatial information of different kinds. One of the outstanding problems in current spatial network research is to effectively quantify the heterogeneity of the discrete-valued spatial distributions underlying a spatial graph. In this talk we will presentsome recent alternative approaches to estimate heterogeneity in spatial networks based on simple dynamical processes running on them.

The parabolic Allen-Cahn equations is the gradient flow of phase transition energy and can be viewed as a diffused version of mean curvature flows of hypersurfaces. It has been known by the works of Ilmanen and Tonegawa that the energy densities of the Allen-Cahn flows converges to mean curvature flows in the sense of varifold and the limit varifold is integer rectifiable. It is not known in general whether the transition layers have higher regularity of convergence yet. In this talk, I will report on a joint work with Huy Nguyen that under the low entropy condition, the convergence of transition layers can be upgraded to C^{2,\alpha} sense. This is motivated by the work of Wang-Wei and Chodosh-Mantoulidis in elliptic case that under the condition of stability, one can upgrade the regularity of convergence.

We will discuss some problems around independence of l in compatible systems of Galois representations, mostly focusing on the independence of l of algebraic monodromy groups. We will explain how these problems fit into the context of the Langlands program, and present results both in characteristic zero and in positive characteristic settings.

We will first construct pairs of homotopic 2-spheres smoothly embedded in a 4-manifold that are smoothly equivalent (via an ambient diffeomorphism preserving homology) but not even topologically isotopic. Indeed, these examples show that Gabai's recent "4D Lightbulb Theorem" does not hold without the 2-torsion hypothesis. We will proceed to discuss two distinct ways of obstructing such an isotopy, as well as related invariants which can be used to obstruct an isotopy between pairs of properly embedded disks (rather than spheres) in a 4-manifold.

The deep neural network has achieved impressive results in various applications, and is involved in more and more branches of science. However, there are still few theories supporting its empirical success. In particular, we miss the mathematical tool to explain the advantage of certain structures of the network, and to have quantitive error bounds. In our recent work, we used a regularised relaxed control problem to model the deep neural network.  We managed to characterise its optimal control by the invariant measure of a mean-field Langevin system, which can be approximated by the marginal laws. Through this study we understand the importance of the pooling for the deep nets, and are capable of computing an exponential convergence rate for the (stochastic) gradient descent algorithm.

An 8-dimensional Riemannian manifold with holonomy group contained in Spin(7) is Ricci-flat, but not Kahler. The condition that the holonomy reduces to Spin(7) is equivalent to a complicated system of non-linear PDEs. In the non-compact setting, symmetries can be used to reduce this complexity. In the case of cohomogeneity one manifolds, i.e. where a generic orbit has codimension one, the non-linear PDE system
reduces to a nonlinear ODE system. I will discuss recent progress in the construction of 1-parameter families of complete cohomogeneity one Spin(7) holonomy metrics. All examples are asymptotically conical (AC) or asymptotically locally conical (ALC).

Consider a collection of particles whose state evolution is described through a system of interacting diffusions in which each particle
is driven by an independent individual source of noise and also by a small amount of noise that is common to all particles. The interaction between the particles is due to the common noise and also through the drift and diffusion coefficients that depend on the state empirical measure. We study large deviation behavior of the empirical measure process which is governed by two types of scaling, one corresponding to mean field asymptotics and the other to the Freidlin-Wentzell small noise asymptotics. 
Different levels of intensity of the small common noise lead to different types of large deviation behavior, and we provide a precise characterization of the various regimes. We also study large deviation behavior of  interacting particle systems approximating various types of Feynman-Kac functionals. Proofs are based on stochastic control representations for exponential functionals of Brownian motions and on uniqueness results for weak solutions of stochastic differential equations associated with controlled nonlinear Markov processes. 

I will discuss recently published examples of SCFTs in
two dimensions and their dual backgrounds. Aspects of the
integrability of these string backgrounds will be described in
correspondence with those of the dual SCFTs. The comparison with four and
six dimensional examples will be presented. It time allows, the case of
conformal quantum mechanics will also be addressed.

Elena Gal
Categorification, Quantum groups and TQFTs

Quantum groups are mathematical objects that encode (via their "category of representations”) certain symmetries which have been found in the last several dozens of years to be connected to several areas of mathematics and physics. One famous application uses representation theory of quantum groups to construct invariants of 3-dimensional manifolds. To extend this theory to higher dimensions we need to “categorify" quantum groups - in essence to find a richer structure of symmetries. I will explain how one can approach such problem.

Carolina Urzua-Torres
Why you should not do boundary element methods, so I can have all the fun.

Boundary integral equations offer an attractive alternative to solve a wide range of physical phenomena, like scattering problems in unbounded domains. In this talk I will give a simple introduction to boundary integral equations arising from PDEs, and their discretization via Galerkin BEM. I will discuss some nice mathematical features of BEM, together with their computational pros and cons. I will illustrate these points with some applications and recent research developments.
 

Sudden cardiac death is the most feared complication of Hypertrophic Cardiomyopathy. This inherited heart muscle disease affects 1 in 500 people. But we are poor at identifying those who really need a potentially life-saving implantable cardioverter-defibrillator. Measuring the abnormalities believed to trigger fatal ventricular arrhythmias could guide treatment. Myocardial disarray is the hallmark feature of patients who die suddenly but is currently a post mortem finding. Through recent advances, the microstructure of the myocardium can now be examined by mapping the preferential diffusion of water molecules along fibres using Diffusion Tensor Cardiac Magnetic Resonance imaging. Fractional anisotropy calculated from the diffusion tensor, quantifies the directionality of diffusion.  Here, we show that fractional anisotropy demonstrates normal myocardial architecture and provides a novel imaging biomarker of the underlying substrate in hypertrophic cardiomyopathy which relates to ventricular arrhythmia.

We present a simple algorithm that successfully re-constructs a sine wave, sampled vastly below the Nyquist rate, but with sampling time intervals having small random perturbations. We show how the fact that it works is just common sense, but then go on to discuss how the procedure relates to Compressed Sensing. It is not exactly Compressed Sensing as traditionally stated because the sampling transformation is not linear.  Some published results do exist that cover non-linear sampling transformations, but we would like a better understanding as to what extent the relevant CS properties (of reconstruction up to probability) are known in certain relatively simple but non-linear cases that could be relevant to industrial applications.

One can study periods of algebraic varieties by a process of "fibering out" in which the variety is fibred over a punctured curve $f:X->U$. I will explain this process and how it leads to the classical Picard Fuchs (or Gauss-Manin) differential equations. Periods are computed by integrating solutions of Picard Fuchs over suitable closed paths on $U$. One can also couple (i.e.tensor) the Picard Fuchs connection to given connections on $U$. For example, $t^s$ with $t$ a unit on $U$ and $s$ a parameter is a solution of the connection on $\mathscr{O}_U$ given by $\nabla(1) = sdt/t$. Our "periods" become integrals over suitable closed chains on $U$ of $f(t)t^sdt/t$. Golyshev called the resulting functions of $s$ "motivic Gamma functions". 
Golyshev and Zagier studied certain special Picard Fuchs equations for their proof of the Gamma conjecture in mirror symmetry in the case of Picard rank 1. They write down a generating series, the Apéry series, the knowledge of the first few terms of which implied the gamma conjecture. We show their Apéry series is the Taylor series of a product of the motivic Gamma function times an elementary function of $s$. In particular, the coefficients of the Apéry series are periods up to inverting $2\pi i$. We relate these periods to periods of the limiting mixed Hodge structure at a point of maximal unipotent monodromy. This is joint work with M. Vlasenko. 
 

We adopt Deep Reinforcement Learning algorithms to design trading strategies for continuous futures contracts. Both discrete and continuous action spaces are considered and volatility scaling is incorporated to create reward functions which scale trade positions based on market volatility. We test our algorithms on the 50 most liquid futures contracts from 2011 to 2019, and investigate how performance varies across different asset classes including commodities, equity indices, fixed income and FX markets. We compare our algorithms against classical time series momentum strategies, and show that our method outperforms such baseline models, delivering positive profits despite heavy transaction costs. The experiments show that the proposed algorithms can follow large market trends without changing positions and can also scale down, or hold, through consolidation periods.
The full-length text is available at https://arxiv.org/abs/1911.10107.
 

Discussion of https://arxiv.org/abs/1911.07895

The task of solving large scale linear algebraic problems such as factorising matrices or solving linear systems is of central importance in many areas of scientific computing, as well as in data analysis and computational statistics. The talk will describe how randomisation can be used to design algorithms that in many environments have both better asymptotic complexities and better practical speed than standard deterministic methods.

The talk will in particular focus on randomised algorithms for solving large systems of linear equations. Both direct solution techniques based on fast factorisations of the coefficient matrix, and techniques based on randomised preconditioners, will be covered.

Note: There is a related talk in the Random Matrix Seminar on Tuesday Feb 25, at 15:30 in L4. That talk describes randomised methods for computing low rank approximations to matrices. The two talks are independent, but the Tuesday one introduces some of the analytical framework that supports the methods described here.

Sustainability is a highly complex topic, containing interwoven economic, ecological, and social aspects.  Simply defining the concept of sustainability is a challenge in itself.  Assessing the impact of sustainability efforts and generating effective policy requires analyzing the interactions and challenges presented by these different aspects. To address this challenge, it is necessary to develop methods that bridge fields and take into account phenomena that range from physical analysis of climate to network analysis of societal phenomena. In this talk, I will give an insight into areas of mathematical research that try to account for these inter-dependencies. The aim of this talk is to provide a critical discussion of the challenges in a joint discussion and emphasize the importance of multi-disciplinary approaches.