Dr Rachel Philip will discuss her experiences working at the interface between academic mathematics and industry. Oxford University Innovation will discuss how they can help academics when interacting with industry. 

Speaker: Daniel Woodhouse (North)
Title: Generalizing Leighton's Graph Covering Theorem
Abstract: Before he ran off and became a multimillionaire, exploiting his knowledge of network optimisation, the computer scientist F. Thomas Leighton proved an innocuous looking result about finite graphs. The result states that any pair of finite graphs with isomorphic universal covers have isomorphic finite covers. I will explain what all this means, and why this should be of tremendous interest to group theorists and topologists.

Speaker: Benjamin Fehrman (South)
Title: Large deviations for particle processes and stochastic PDE
Abstract: In this talk, we will introduce the theory of large deviations through a simple example based on flipping a coin.  We will then define the zero range particle process, and show that its diffusive scaling limit solves a nonlinear diffusion equation.  The large deviations of the particle process about its scaling limit formally coincide with the large deviations of a certain ill-posed, singular stochastic PDE.  We will explain in what sense this relationship has been made mathematically precise.

Speaker: Joseph Keir (North)
Title: Dispersion (or not) in nonlinear wave equations
Abstract: Wave equations are ubiquitous in physics, playing central roles in fields as diverse as fluid dynamics, electromagnetism and general relativity. In many cases of these wave equations are nonlinear, and consequently can exhibit dramatically different behaviour when their solutions become large. Interestingly, they can also exhibit differences when given arbitrarily small initial data: in some cases, the nonlinearities drive solutions to grow larger and even to blow up in a finite time, while in other cases solutions disperse just like the linear case. The precise conditions on the nonlinearity which discriminate between these two cases are unknown, but in this talk I will present a conjecture regarding where this border lies, along with some conditions which are sufficient to guarantee dispersion.

Speaker: Priya Subramanian (South)
Title: What happens when an applied mathematician uses algebraic geometry?
Abstract: A regular situation that an applied mathematician faces is to obtain the equilibria of a set of differential equations that govern a system of interest. A number of techniques can help at this point to simplify the equations, which reduce the problem to that of finding equilibria of coupled polynomial equations. I want to talk about how homotopy methods developed in computational algebraic geometry can solve for all solutions of coupled polynomial equations non-iteratively using an example pattern forming system. Finally, I will end with some thoughts on what other 'nails' we might use this new shiny hammer on.

Persistent homology has been applied to graph classification problems as a way of generating vectorizable features of graphs that can be fed into machine learning algorithms, such as neural networks. A key ingredient of this approach is a filter constructor that assigns vector features to nodes to generate a filtration. In the case where the filter constructor is smoothly tuned by a set of real parameters, we can train a neural network graph classifier on data to learn an optimal set of parameters via the backpropagation of gradients that factor through persistence diagrams [Leygonie et al., arXiv:1910.00960]. We propose a flexible, spectral-based filter constructor that parses standalone graphs, generalizing methods proposed in [Carrière et al., arXiv: 1904.09378]. Our method has an advantage over optimizable filter constructors based on iterative message passing schemes (`graph neural networks’) [Hofer et al., arXiv: 1905.10996] which rely on heuristic user inputs of vertex features to initialise the scheme for datasets where vertex features are absent. We apply our methods to several benchmark datasets and demonstrate results comparable to current state-of-the-art graph classification methods.

Spreading processes on geometric networks are often influenced by a network’s underlying spatial structure, and it is insightful to study the extent to which a spreading process follows that structure. In particular, considering a threshold contagion on a network whose nodes are embedded in a manifold and which has both 'geometric edges' that respect the geometry of the underlying manifold, as well as 'non-geometric edges' that are not constrained by the geometry of the underlying manifold, one can ask whether the contagion propagates as a wave front along the underlying geometry, or jumps via long non-geometric edges to remote areas of the network. 
Taylor et al. developed a methodology aimed at determining the spreading behaviour of threshold contagion models on such 'noisy geometric networks' [1]. This methodology is inspired by nonlinear dimensionality reduction and is centred around a so-called 'contagion map' from the network’s nodes to a point cloud in high dimensional space. The structure of this point cloud reflects the spreading behaviour of the contagion. We apply this methodology to a family of noisy-geometric networks that can be construed as being embedded in a torus, and are able to identify a region in the parameter space where the contagion propagates predominantly via wave front propagation. This consolidates contagion map as both a tool for investigating spreading behaviour on spatial network, as well as a manifold learning technique. 
[1] D. Taylor, F. Klimm, H. A. Harrington, M. Kramar, K. Mischaikow, M. A. Porter, and P. J. Mucha. Topological data analysis of contagion maps for examining spreading processes on networks. Nature Communications, 6(7723) (2015)

We will discuss what it means to study the homology of a group via the construction of the classifying space. We will look at some examples of this construction and some of its main properties. We then use this to define and study the homology of the mapping class group of oriented surfaces, focusing on the approach used by Harer to prove his Homology Stability Theorem.

When Gromov defined non-positively curved cube complexes no one knew what they would be useful for.
Decades latex they played a key role in the resolution of the Virtual Haken conjecture.
In one of the early forays into experimenting with cube complexes, Aitchison, Matsumoto, and Rubinstein produced some nice results about certain "cubed" manifolds, that in retrospect look very prescient.
I will define non-positively curved cube complexes, what it means for a 3-manifold to be cubed, and discuss what all this Haken business is about.
 

A graph of groups decomposition is a way of splitting a group into smaller and hopefully simpler groups. A natural thing to try and do is to keep splitting until you can't split anymore, and then argue that this decomposition is unique. This is the idea behind JSJ decompositions, although, as we shall see, the strength of the uniqueness statement for such a decomposition varies depending on the class of groups that we restrict our edge groups to

Machine learning for scientific applications faces the challenge of limited data. To reduce overfitting, we propose a framework to leverage as much as possible a priori-known physics for a problem. Our approach embeds a deep neural network in a partial differential equation (PDE) system, where the pre-specified terms in the PDE capture the known physics and the neural network will learn to describe the unknown physics. The neural network is estimated from experimental and/or high-fidelity numerical datasets. We call this approach a “deep learning PDE model” (DPM). Once trained, the DPM can be used to make out-of-sample predictions for new physical coefficients, geometries, and boundary conditions. We implement our approach on a classic problem of the Navier-Stokes equation for turbulent flows. The numerical solution of the Navier-Stokes equation (with turbulence) is computationally expensive and requires a supercomputer. We show that our approach can estimate a (computationally fast) DPM for the filtered velocity of the Navier-Stokes equations.