L2

Tensors are higher dimensional analogues of matrices, used to record data with multiple changing variables. Interpreting tensor data requires finding multi-linear stucture that depends on the application or context. I will describe a tensor-based clustering method for multi-dimensional data. The multi-linear structure is encoded as algebraic constraints in a linear program. I apply the method to a collection of experiments measuring the response of genetically diverse breast cancer cell lines to an array of ligands. In the second part of the talk, I will discuss low-rank decompositions of tensors that arise in statistics, focusing on two graphical models with hidden variables. I describe how the implicit semi-algebraic description of the statistical models can be used to obtain a closed form expression for the maximum likelihood estimate.

We present some regularity estimates for viscosity solutions to a class of possible degenerate and singular integro-differential equations whose leading operator switches between two different types of fractional elliptic phases, according to the zero set of a modulating coefficient a = a(·, ·). The model case is driven by the following nonlocal double phase operator,

$$\int \frac{|u(x) − u(y)|^{p−2} (u(x) − u(y))} {|x − y|^{n+sp}} dy+ \int a(x, y) \frac{|u(x) − u(y)|^{ q−2} (u(x) − u(y))} {|x − y|^{n+tq}} dy$$

where $q ≥ p$ and $a(·, ·) = 0$. Our results do also apply for inhomogeneous equations, for very general classes of measurable kernels. By simply assuming the boundedness of the modulating coefficient, we are able to prove that the solutions are Hölder continuous, whereas similar sharp results for the classical local case do require a to be Hölder continuous. To our knowledge, this is the first (regularity) result for nonlocal double phase problems.

Numerous researchers recently applied empirical spectral analysis to the study of modern deep learning classifiers. We identify and discuss an important formal class/cross-class structure and show how it lies at the origin of the many visually striking features observed in deepnet spectra, some of which were reported in recent articles and others unveiled here for the first time. These include spectral outliers and small but distinct bumps often seen beyond the edge of a "main bulk". The structure we identify organizes the coordinates of deepnet features and back-propagated errors, indexing them as an NxC or NxCxC array. Such arrays can be indexed by a two-tuple (i,c) or a three-tuple (i,c,c'), where i runs across the indices of the train set; c runs across the class indices and c' runs across the cross-class indices. This indexing naturally induces C class means, each obtained by averaging over the indices i and c' for a fixed class c. The same indexing also naturally defines C^2 cross-class means, each obtained by averaging over the index i for a fixed class c and a cross-class c'. We develop a formal process of spectral attribution, which is used to show the outliers are attributable to the C class means; the small bump next to the "main bulk" is attributable to between-cross-class covariance; and the "main bulk" is attributable to within-cross-class covariance. Formal theoretical results validate our attribution methodology.
We show how the effects of the class/cross-class structure permeate not only the spectra of deepnet features and backpropagated errors, but also the gradients, Fisher Information matrix and Hessian, whether these are considered in the context of an individual layer or the concatenation of them all. The Kronecker or Khatri-Rao product of the class means in the features and the class/cross-class means in the backpropagated errors approximates the class/cross-class means in the gradients. These means of gradients then create C and C^2 outliers in the spectrum of the Fisher Information matrix, which is the second moment of these gradients. The outliers in the Fisher Information matrix spectrum then create outliers in the Hessian spectrum. We explain the significance of this insight by proposing a correction to KFAC, a well known second-order optimization algorithm for training deepnets.

We study nearly singular PDEs that arise in the solution of linear elasticity and incompressible flow. We will demonstrate, that due to the nearly singular nature, standard methods for the solution of the linear systems arising in a finite element discretisation for these problems fail. We motivate two key ingredients required for a robust multigrid scheme for these equations and construct robust relaxation and prolongation operators for a particular choice of discretisation.
 

We consider CFTs with large gap in the spectrum of operators and a large number of degrees of freedom (large central charge). We analytically study a Heavy-Heavy-Light-Light correlation function, where Heavy, refers to an operator with conformal dimension which scales like the central charge and Light, refers to an operator whose dimension is of order unity in the large central charge limit. In certain regimes, the correlation function can be examined analytically leading to very simple and suggestive expressions.

In this talk I discuss how we can simulate stochastic chemical kinetics when there is a memory component. This can occur when there is spatial crowding within a cell or part of a cell, which acts to constrain the motion of the molecules which then in turn changes the dynamics of the chemistry. The counterpart of the Law of Mass Action in this setting is through replacing the first derivative in the ODE description of the Law of Mass Action by a time-­fractional derivative, where the time-­fractional index is between 0 and 1. There has been much discussion in the literature, some of it wrong, as to how we model and simulate stochastic chemical kinetics in the setting of a spatially-­constrained domain – this is sometimes called anomalous diffusion kinetics.

In this presentation, I discuss some of these issues and then present two (equivalent) ways of simulating fractional stochastic chemical kinetics. The key here is to either replace the exponential waiting time used in Gillespie’s SSA by Mittag-­Leffler waiting times (MacNamara et al. [2]), which have longer tails than in the exponential case. The other approach is to use some theory developed by Jahnke and Huisinga [1] who are able towrite down the underlying probability density function for any set of mono-­molecular chemical reactions (under the standard Law of Mass Action) as a convolution of either binomial probability density functions or binomial and Poisson probability density functions). We can then extend the Jahnke and Huisinga formulation through the concept of iterated Brownian Motion paths to produce exact simulations of the underlying fractional stochastic chemical process. We demonstrate the equivalence of these two approaches through simulations and also by computing the probability density function of the underlying fractional stochastic process, as described by the fractional chemical master equation whose solution is the Mittag-­Lefflermatrix function. This is computed based on a clever algorithm for computing matrix functions by Cauchy contours (Weideman and Trefethen [3]).

This is joint work with Manuel Barrio (University of Vallodolid, Spain), Kevin Burrage (QUT), Andre Leier (University of Alabama), Shev MacNamara(University of Technology Sydney)and T. Marquez-­Lago (University of Alabama).

[1]T. Jahnke and W. Huisinga, 2007, Solving the chemical master equation for monomolecular reaction systems analytically, J. Math. Biology 54, 1, 1—26.[2]S. MacNamara, B. Henry and W. McLean, 2017, Fractional Euler limits and their applications, SIAM J. Appl. Math. 77, 2, 447—469.[3]J.A.C. Weideman and L.N. Trefethen, 2007, Parabolic and hyperbolic contours for computing the Bromwich integral, Math. Comp. 76, 1341—1356.

ACORN generators represents an approach to generating uniformly distributed pseudo-random numbers which is straightforward to implement for arbitrarily large order $k$ and modulus $M=2^{30t}$ (integer $t$). They give long period sequences which can be proven theoretically to approximate to uniformity in up to $k$ dimensions, while empirical statistical testing demonstrates that (with a few very simple constraints on the choice of parameters and the initialisation) the resulting sequences can be expected to pass all the current standard tests .

The standard TestU01 Crush and BigCrush Statistical Test Suites are used to demonstrate for ACORN generators with order $8≤k≤25$ that the statistical performance improves as the modulus increases from $2^{60}$ to $2^{120}$. With $M=2^{120}$ and $k≥9$, it appears that ACORN generators pass all the current TestU01 tests over a wide range of initialisations; results are presented that demonstrate the remarkable consistency of these results, and explore the limits of this behaviour.

This contrasts with corresponding results obtained for the widely-used Mersenne Twister MT19937 generator, which consistently failed on two of the tests in both the Crush and BigCrush test suites.

There are other pseudo-random number generators available which will also pass all the TestU01 tests. However, for the ACORN generators it is possible to go further: we assert that an ACORN generator might also be expected to pass any more demanding tests for $p$-dimensional uniformity that may be required in the future, simply by choosing the order $k>p$, the modulus $M=2^{30t}$ for sufficiently large $t$, together with any odd value for the seed and an arbitrary set of initial values. We note that there will be $M/2$ possible odd values for the seed, with each such choice of seed giving rise to a different $k$-th order ACORN sequence satisfying all the required tests.

This talk builds on and extends results presented at the recent discussion meeting on “Numerical algorithms for high-performance computational science” at the Royal Society London, 8-9 April 2019, see download link at bottom of web page http://acorn.wikramaratna.org/references.html.

This meeting is being held in celebration of Prof Philip Maini's 60th birthday. Prof Maini has been an internationally leading researcher in mathematical biology for decades. He is currently the Director of the Wolfson Centre for Mathematical Biology, a position he has held since 1998. In the past 20 years he has grown the group significantly. He has established countless interdisciplinary collaborations, has over 400 publications in numerous areas of mathematical biology, with major contributions in mathematical modelling of tumours, wound healing and embryonic pattern formation. He has been elected Fellow of the Royal Society (FRS), Fellow of the Academy of Medical Sciences (FMedSci), and Foreign Fellow of the Indian National Science Academy (FNA). He has served or is serving on editorial board of a large number of journals, and was Editor-in-Chief of the Bulletin of Mathematical Biology [2002-15]. And yet his service to the community cannot be captured just by numbers and titles. Anyone who has met him and worked with him cannot but notice and be touched by his unfailing generosity and the many sacrifices he has made and continues to make day in and day out to help students, early career researchers, and fellow faculty alike.

This meeting provides an opportunity to celebrate Prof Maini's many accomplishments; to thank him for all of his sacrifices; and to bring together the large number of researchers – mathematicians, biologists, physiologists, and clinicians – that he has worked with and interacted with over the years. More broadly, the meeting provides a unique opportunity to reflect on mathematical biology, to provide perspectives on the trajectory of a field that was scarcely recognised and had very few dedicated researchers in the days of Prof Maini's own DPhil; yet a field that has grown tremendously since then. Much of this growth is attributable to the work of Prof Maini, so that today the value of mathematics in biology is increasingly recognized by biologists and clinicians, and with theoretical predictions of mathematical models having cemented a role in advancing biological understanding. 

Speakers

David Sumpter, Uppsala University (Public lecture), Derek Moulton, University of Oxford, Hans Othmer, Minnesota University, Jen Flegg, University of Melbourne, Jim Murray, University of Washington, Jonathan Sherratt, Heriot-Watt University, Kevin Painter, Heriot-Watt University, Linus Schumacher, University of Edinburgh, Lucy Hutchinson, Roche, Mark Chaplain, University of St Andrews, Mark Lewis, University of Alberta, Mary Myerscough, University of Sydney, Natasha Martin, University of Bristol, Noemi Picco, Swansea University, Paul Kulesa, Stowers Institute for Medical Research, Ruth Baker, University of Oxford, Santiago Schnell, University of Michigan, Tim Pedley, University of Cambridge

Organising committee

Ruth Baker (University of Oxford)

Derek Moulton (University of Oxford)

Helen Byrne (University of Oxford)

Santiago Schnell (University of Michigan)

Mark Chaplain (University of St Andrews)

The disruptive drone activity at airports requires an early warning system and Aveillant make a radar system that can do the job. The main problem is telling the difference between birds and drones where there may be one or two drones and 10s or 100s of birds. There is plenty of data including time series for how the targets move and the aim is to improve the discrimination capability of tracker using machine learning.

Specifically, the challenge is to understand whether there can be sufficient separability between birds and drones based on different features, such as flight profiles, length of the track, their states, and their dominance/correlation in the overall discrimination. Along with conventional machine learning techniques, the challenge is to consider how different techniques, such as deep neural networks, may perform in the discrimination task.