Past

The focus of this talk is how to tackle huge linear least square problems via sketching, a dimensionality reduction technique from randomised numerical linear algebra. The technique allows us to project the huge problem to a smaller dimension that captures essential information of the original problem. We can then solve the projected problem directly to obtain a low accuracy solution or using the projected problem to construct a preconditioner for the original problem to obtain a high accuracy solution. I will survey the existing projection techniques and evaluate the performance of sketching for linear least square problems by comparing it to the state-of-the-art traditional solution methods. More than ten-fold speed-up has been observed in some cases.

There have been two main attempts so far to forecast the level of development of artificial intelligence (or ‘computerisation’) over time, Frey and Osborne (2013, 2017) and Manyika et al (2017). Unfortunately, their methodology seems to be flawed. Their results depend upon expert predictions of which occupations will be automatable in 2050, but these predictions are notoriously unreliable. Therefore, we develop an alternative which does not depend upon these expert predictions. We build a dataset of all the start-ups, firms, and university research laboratories working on automating different types of tasks, and use this to build a dynamic network model of them and how they interact. How automatable each type of task is ‘emerges’ from the model. We validate it, predicting the level of development of supervised learning in 2017 using data from the year 2000, and use it to forecast of the automatability of each of these task types from 2018 to 2050. Finally, we discuss extensions for our model; how it could be used to test the impact of public policy decisions or forecast developments in other high-technology industries.

Progress in developing a consistent theory that describes physical phenomena
at scales where quantum and general relativistic effects are large is
hindered by the lack of experiments. In this talk, we present a proposal
that would overcome this experimental obstacle by using a Bose-Einstein
condensate (BEC) to test for possible conflicts between quantum theory and
general relativity. Recent developments in large BEC systems allows us to
verify if gravitationally-induced wave function collapse occurs at the
timescales predicted by Roger Penrose. BECs with high particle numbers
(N>10^9) can also be used to demonstrate quantum field theory in curved
spacetime by observing how changes in the spacetime affect the phononic
quantum field of a BEC. These effects will enable the development of a new
generation of instruments that will be able to probe scales where new
physics might emerge, with applications including gravitational wave
detectors, gravimeters, gradiometers and dark energy probes.

Abstract: We use groupoids to describe a geometric framework which can host a generalisation of Hairer's regularity structures to manifolds. In this setup, Hairer's re-expansionmap (usually denoted \Gamma) is a (direct) connection on a gauge groupoid and can therefore be viewed as a groupoid counterpart of a (local) gauge field. This definitions enables us to make the link between re-expansion maps (direct connections), principal connections and path connections, to understand the flatness of the direct connection in terms of that of the manifold and, finally, to easily build a polynomial regularity structure which we compare to the one given by Driver, Diehl and Dahlquist. (Join work with Sara Azzali, Alessandra Frabetti and Sylvie Paycha).

Quantum Yang-Mills theory is an important part of the Standard model built by physicists to describe elementary particles and their interactions. One approach to the mathematical substance of this theory consists in constructing a probability measure on an infinite-dimensional space of connections on a principal bundle over space-time. However, in the physically realistic 4-dimensional situation, the construction of this measure is still an open mathematical problem. The subject of this talk will be the physically less realistic 2-dimensional situation, in which the construction of the measure is possible, and fairly well understood.

In probabilistic terms, the 2-dimensional Yang-Mills measure is the distribution of a stochastic process with values in a compact Lie group (for example the unitary group U(N)) indexed by the set of continuous closed curves with finite length on a compact surface (for example a disk, a sphere or a torus) on which one can measure areas. It can be seen as a Brownian motion (or a Brownian bridge) on the chosen compact Lie group indexed by closed curves, the role of time being played in a sense by area.

In this talk, I will describe the physical context in which the Yang-Mills measure is constructed, and describe it without assuming any prior familiarity with the subject. I will then present a set of results obtained in the last few years by Antoine Dahlqvist, Bruce Driver, Franck Gabriel, Brian Hall, Todd Kemp, James Norris and myself concerning the limit as N tends to infinity of the Yang-Mills measure constructed with the unitary group U(N). 

 

After discussing a general excision technique for constructing canonical orientations for moduli spaces that derive from an elliptic equation, I shall
explain how to carry out this program in the case of G2-instantons and the 7-dimensional real Dirac operator. In many ways our approach can
be regarded as a categorification of the Atiyah-Singer index theorem. (Based on joint work with Dominic Joyce.)

The E-string theory is usually considered as the simplest among 6D (1,0) superconformal field theories. Nonetheless, we still have little information about its spectrum of operators. In this talk, I'm going to describe our recent geometric approach using F-theory compactification on an elliptic Calabi-Yau threefold. The elliptic fibration is non-flat, which means that there are complex surface components in the fiber direction. From the geometry of non-flat fiber, we read out an infinite tower of particle states in the E-string theory. I will also discuss its relevance to 4D standard model building, which is a main motivation of this work.
 

Teaching ethics to the mathematicians who need it most
For the last 20 years it has become increasingly obvious, and increasingly pressing, that mathematicians should be taught some ethical awareness so as to realise the impact of their work. This extends even to those more highly trained, like graduate students and postdocs. But which mathematicians should we be teaching this to, what should we be teaching them, and how should we do it? In this talk I’ll explore the idea that all mathematicians will, at some stage, be faced with ethical challenges stemming from their work, and yet few are ever told beforehand.
 

Mathematics is both the language and the instrument that connects our abstract understanding with the physical world, thus knowledge of mathematics quickly translates to substantial knowledge and influence on the way the world works.  But those who have the greatest ability to understand and manipulate the world hold the greatest capacity to do damage and inflict harm.  In this talk I'll explain that yes, there is ethics in mathematics, and that it is up to us as mathematicians to make good ethical choices in order to prevent our work from becoming harmful.

In this work we will attempt, via virtual models, to interpret how structure and body positioning impact upon the outcomes of Multi-Breath-Washout tests. 

By extrapolating data from CT images, a virtual reduced dimensional airway/vascualr network will be constructed. Using this network both airway and blood flow profiles will be calculated. These profiles will then be used to model gas transport within the lungs. The models will allow us to investigate the role of airway restriction, body position during testing and washout gas choice have on MBW measures. 
 

Many scientists, and in particular mathematicians, report difficulty in understanding thermodynamics. So why is thermodynamics so difficult? To attempt an answer, we begin by looking at the components in an exposition of a scientific theory. These include a mathematical core, a motivation for the choice of variables and equations, some historical remarks, some examples and a discussion of how variables, parameters, and functions (such as equations of state) can be inferred from experiments. There are other components too, such as an account of how a theory relates to other theories in the subject.

It will be suggested that theories of thermodynamics are hard to understand because (i) many expositions appear to argue from the particular to the general (ii) there are several different thermodynamic theories that have no obvious logical or mathematical equivalence (iii) each theory really is subtle and requires intense study (iv) in most expositions different theories are mixed up, and the different components of a scientific exposition are also mixed up. So, by presenting one theory at a time, and by making clear which component is being discussed, we might reduce the difficulty in understanding any individual thermodynamic theory. The key is perhaps separation of the mathematical core from the physical motivation. It is also useful to realise that a motivation is not generally the same as a proof, and that no theory is actually true.

By way of illustration we will attempt expositions of two of the simplest thermodynamic theories – reversible and then irreversible thermodynamics of homogeneous materials – where the mathematical core and the motivation are discussed separately. In conclusion we’ll relate these two simple theories to other, foundational and generalised, thermodynamic theories.

In the past decade, deep learning methods have achieved unprecedented performance on a broad range of problems in various fields from computer vision to speech recognition. So far research has mainly focused on developing deep learning methods for Euclidean-structured data. However, many important applications have to deal with non-Euclidean structured data, such as graphs and manifolds. Such data are becoming increasingly important in computer graphics and 3D vision, sensor networks, drug design, biomedicine, high energy physics, recommendation systems, and social media analysis. The adoption of deep learning in these fields has been lagging behind until recently, primarily since the non-Euclidean nature of objects dealt with makes the very definition of basic operations used in deep networks rather elusive. In this talk, I will introduce the emerging field of geometric deep learning on graphs and manifolds, overview existing solutions and outline the key difficulties and future research directions. As examples of applications, I will show problems from the domains of computer vision, graphics, high-energy physics, and fake news detection. 

Clustering is a very important task in data analytics and is usually addressed using (i) statistical tools based on maximum likelihood estimators for mixture models, (ii) techniques based on network models such as the stochastic block model, or (iii) relaxations of the K-means approach based on semi-definite programming (or even simpler spectral approaches). Statistical approaches of type (i) often suffer from not being solvable with sufficient guarantees, because of the non-convexity of the underlying cost function to optimise. The other two approaches (ii) and (iii) are amenable to convex programming but do not usually scale to large datasets. In the big data setting, one usually needs to resort to data subsampling, a preprocessing stage also known as "coreset selection". We will present this last approach and the problem of selecting a coreset for the special cases of K-means and spectral-type relaxations.

Seminal work of Hida tells us that if a modular eigenform is ordinary at p then we can always find other eigenforms, of different weights, that are congruent to our given form. Even better, it says that we can find q-expansions with coefficients in p-adic analytic function of the weight variable k that when evaluated at positive integers give the q-expansion of classical eigenforms. His construction of these families uses mainly the geometry of the modular curve and its ordinary locus.
In a joint work with Marc-Hubert Nicole, we obtained similar results for Drinfeld modular forms over function fields. After an extensive introduction to Drinfeld modules, their moduli spaces, and Drinfeld modular forms, we shall explain how to construct Hida families for ordinary Drinfeld modular forms.

 One of the most defining features of modern financial networks is their inherent complex and intertwined structure. In particular the often observed core-periphery structure plays a prominent role. Here we study and quantify the impact that the complexity of networks has on contagion effects and system stability, and our focus is on the channel of default contagion that describes the spread of initial distress via direct balance sheet exposures. We present a general approach describing the financial network by a random graph, where we distinguish vertices (institutions) of different types - for example core/periphery - and let edge probabilities and weights (exposures) depend on the types of both the receiving and the sending vertex. Our main result allows to compute explicitly the systemic damage caused by some initial local shock event, and we derive a complete characterization of resilient respectively non-resilient financial systems in terms of their global statistical characteristics. Due to the random graphs approach these results bear a considerable robustness to local uncertainties and small changes of the network structure over time. Applications of our theory demonstrate that indeed the features captured by our model can have significant impact on system stability; we derive resilience conditions for the global network based on subnetwork conditions only. 

The Strominger-Yau-Zaslow (SYZ) conjecture postulates that mirror dual Calabi-Yau manifolds carry dual special Lagrangian fibrations. Within the study of Mirror Symmetry the SYZ conjecture has provided a particularly fruitful point of convergence of ideas from Riemannian, Symplectic, Tropical, and Algebraic geometry over the last twenty years. I will attempt to provide a brief overview of this aspect of Mirror Symmetry.

Blisters form when a thin surface layer of a solid body separates/delaminates from the underlying bulk material over a finite, bounded region. It is ubiquitous in a range of industrial applications, e.g. blister test is applied to assess the strength of adhesion between thin elastic films and their solid substrates, and during natural processes, such as formation and spreading of laccoliths or retinal detachment.

We study a special case of blistering, in which a thin elastic membrane is adhered to the substrate by a thin layer of viscous fluid. In this scenario, the expansion of the newly formed blister by fluid injection occurs via a displacement flow, which peels apart the adhered surfaces through a two-way interaction between flow and deformation. If the injected fluid is less viscous than the fluid already occupying the gap, patterns of short and stubby fingers form on the propagating fluid interface in a radial geometry. This process is regulated by membrane compliance, which if increased delays the onset of fingering to higher flow rates and reduces finger amplitude. We find that the morphological features of the fingers are selected in a simple way by the local geometry of the compliant cell. In contrast, the local geometry itself is determined from a complex fluid–solid interaction, particularly in the case of rectangular blisters. Furthermore, changes to the geometry of the channel cross-section in the latter case lead to a rich variety of possible interfacial patterns. Our experiments provide a link between studies of airway reopening, Saffman-Taylor fingering and printer’s instability.   

We discuss the key role that bespoke linear algebra plays in modelling PDEs with random coefficients using stochastic Galerkin approximation methods. As a specific example, we consider nearly incompressible linear elasticity problems with an uncertain spatially varying Young's modulus. The uncertainty is modelled with a finite set of parameters with prescribed probability distribution.  We introduce a novel three-field mixed variational formulation of the PDE model and and  assess the stability with respect to a weighted norm. The main focus will be  the efficient solution of the associated high-dimensional indefinite linear system of equations. Eigenvalue bounds for the preconditioned system can be  established and shown to be independent of the discretisation parameters and the Poisson ratio.  We also  discuss an associated a posteriori error estimation strategy and assess proxies for the error reduction associated with selected enrichments of the approximation spaces.  We will show by example that these proxies enable the design of efficient  adaptive solution algorithms that terminate when the estimated error falls below a user-prescribed tolerance.

This is joint work with Arbaz Khan and Catherine Powell

Tanut Treetanthiploet
---------------------
Exploration vs Exploitation under Statistical Uncertainty

The exploration vs Exploitation trade-off can be quantified and studied through the notion of statistical uncertainty using the theory of nonlinear expectations. The dynamic allocation problem of multi-armed bandits will be discussed. In the case of a finite state space in discrete time, we can describe the value function in terms of the solution to a discrete BSDE and obtain a similar notion to the Bellman equation. We also give an approximation scheme to evaluate decisions in the simple setting.


Julien Vaes
-----------
Optimal Execution Strategy Under Price and Volume Uncertainty

In the seminal paper on optimal execution of portfolio transactions, Almgren and Chriss define the optimal trading strategy to liquidate a fixed volume of a single security under price uncertainty. Yet there exist situations, such as in the power market, in which the volume to be traded can only be estimated and becomes more accurate when approaching a specified delivery time. To meet the need of efficient strategies in these situations, we have developed  a model that accounts for volume uncertainty and show that a risk-averse trader has benefit in delaying their trades. We show that the optimal strategy is a trade-off between early and late trades to balance risk associated to both price and volume. With the incorporation of a risk term for the volume to trade, the static optimal strategies obtained with our model avoid the explosion in the algorithmic complexity associated to dynamic programming solutions while yielding to competitive performance.
 

In this talk I will discuss recent work with P. Daskalopoulos on sufficient conditions to prove uniqueness of complete graphs evolving by mean curvature flow. It is interesting to remark that the behaviour of solutions to mean curvature flow differs from the heat equation, where non-uniqueness may occur even for smooth initial conditions if the behaviour at infinity is not prescribed for all times. 

NSOP1 is a class of first order theories containing simple theories, which contains many natural examples that somehow slip-out of the simple context.

As in simple theories, NSOP1 theories admit a natural notion of independence dubbed Kim-independence, which generalizes non-forking in simple theories and satisfies many of its properties.

In this talk I will explain all these notions, and in particular talk about recent progress (joint with Nick Ramsey) in the study of Kim-independence, showing transitivity and several consequences.