Past

Classifying line arrangements on the plane is a problem that has been around for a long time. There has been a lot of work from the perspective of incidence geometry, but after a paper of Hirzebruch in in 80's, it has also attracted the attention of algebraic geometers for the applications that it has on classifying complex algebraic surfaces of general type. In this talk I will present various results around this problem, I will show some specific questions that are still open, and I will explain how it relates to complex surfaces of general type. 
 

It is well known that the theory of differentially closed fields of characteristic 0 has prime models and that they are unique up to isomorphism. One can ask the same question for the theory ACFA of existentially closed difference fields (recall that a difference field is a field with an automorphism).

In this talk, I will first give the trivial reasons of why this question cannot have a positive answer. It could however be the case that over certain difference fields prime models (of the theory ACFA) exist and are unique. Such a prime model would be called a difference closure of the difference field K. I will show by an example that the obvious conditions on K do not suffice.

I will then consider the class of aleph-epsilon saturated models of ACFA, or of kappa-saturated models of ACFA. There are natural notions of aleph-epsilon prime model and kappa-prime model. It turns out that for these stronger notions, if K is an algebraically closed difference field of characteristic 0, with fixed subfield F aleph-epsilon saturated, then there is an aleph-epsilon prime model over K, and it is unique up to K-isomorphism. A similar result holds for kappa-prime when kappa is a regular cardinal.

None of this extends to positive characteristic.
 

We consider a generalization of degeneracy loci of morphisms between vector bundles based on orbit closures of algebraic groups in their linear representations. Using a certain crepancy condition on the orbit closure we gain some control over the canonical sheaf in a preferred class of examples. This is notably the case for Richardson nilpotent orbits and partially decomposable skew-symmetric three-forms in six variables. We show how these techniques can be applied to construct Calabi-Yau manifolds and Fano varieties of dimension three and four.

This is a joint work with Vladimiro Benedetti, Laurent Manivel and Fabio Tanturri.

We provide the rate of convergence of general monotone numerical schemes for parabolic Hamilton-Jacobi-Bellman equations in bounded domains with Dirichlet boundary conditions. The so-called "shaking coefficients" technique introduced by Krylov is used. This technique is based on a perturbation of the dynamics followed by a regularization step by convolution. When restricting the equation to a domain, the perturbed problem may not satisfy such a restriction, so that a special treatment near the boundary is necessary. 

In my talk I will explain how to relate 1-dimensional representations of finite W-algebras with multiplicity free primitive ideals of universal enveloping algebras and representations of minimal dimension of the corresponding reduced enveloping algebras (Humphreys' conjecture). I will also mention some open problems in the field.

The phenomenon of poor algorithmic scalability is a critical problem in large-scale machine learning and data science. This has led to a resurgence in the use of first-order (Hessian-free) algorithms from classical optimisation. One major drawback is that first-order methods tend to converge extremely slowly. However, there exist techniques for efficiently accelerating them.
    
The topic of this talk is the Dual Regularisation Nonlinear Acceleration algorithm (DRNA) (Geleta, 2017) for nonconvex optimisation. Numerical studies using the CUTEst optimisation problem set show the method to accelerate several nonconvex optimisation algorithms, including quasi-Newton BFGS and steepest descent methods. DRNA compares favourably with a number of existing accelerators in these studies.
    
DRNA extends to the nonconvex setting a recent acceleration algorithm due to Scieur et al. (Advances in Neural Information Processing Systems 29, 2016). We have proven theorems relating DRNA to the Kylov subspace method GMRES, as well as to Anderson's acceleration method and family of multi-secant quasi-Newton methods.
 

Superradiance in black hole spacetimes is a phenomenon by which a field of spin 0 or 1 can extract energy from the background. Typically, one can imagine sending a wave packet with a given energy towards a black hole and receiving in return a superposition of wave packets carrying a total amount of energy that is larger than the energy sent in. It can be caused by rotation or by interaction between the charges of the black hole and the field. In the first case, the region where superradiance takes place (the ergoregion) has a clear geometrical localization depending only on the physical parameters of the black hole. For charge induced superradiance, this is not the case and we have a generalized ergoregion depending also on the physical properties of the field (mass, charge, angular momentum). In the most severe cases, the generalized ergoregion may cover the whole exterior of the black hole. We focus on charge-induced superradiance for spin 0 fields in spherically symmetric situations. Alain Bachelot wrote a thorough theoretical study of this question in 2004, which, to my knowledge, is the only work of its kind. When I was in Bordeaux, he and I discussed the possibility of investigating superradiance numerically. Over the years it became an actual research project, involving Laurent Di Menza and more recently Mathieu Pellen, of which this talk is an account. The idea was to observe numerically some superradiant behaviours and gain a more precise understanding of the phenomenon. We shall show an exact analogue of the Penrose process with the superradiance of wave packets and a slightly different behaviour for fields "emerging" inside the ergoregion. We shall also explore the related question of black hole bombs and present some recent observations. 

It is well-known that only a limited number of the fluid flow problems can be solved (or approximated) by the solutions in the explicit form. To derive such solutions, we usually need to start with (over)simplified mathematical models and consider ideal geometries on the flow domains with no distortions introduced. However, in practice, the boundary of the fluid domain can contain various small irregularities (rugosities, dents, etc.) being far from the ideal one. Such problems are challenging from the mathematical point of view and, in most cases, can be treated only numerically. The analytical treatments are rare because introducing the small parameter as the perturbation quantity in the domain boundary forces us to perform tedious change of variables. Having this in mind, our goal is to present recent analytical results on the effects of a slightly perturbed boundary on the fluid flow through a channel filled with a porous medium. We start from a rectangular domain and then perturb the upper part of its boundary by the product of the small parameter $\varepsilon$ and arbitrary smooth function. The porous medium flow is described by the Darcy-Brinkman model which can handle the presence of a boundary on which the no-slip condition for the velocity is imposed. Using asymptotic analysis with respect to $\varepsilon$, we formally derive the effective model in the form of the explicit formulae for the velocity and pressure. The obtained asymptotic approximation clearly shows the nonlocal effects of the small boundary perturbation. The error analysis is also conducted providing the order of accuracy of the asymptotic solution. We will also address the problem of the solute transport through a semi-infinite channel filled with a fluid saturated sparsely packed porous medium. A small perturbation of magnitude $\varepsilon$ is applied on the channel's walls on which the solute particles undergo a first-order chemical reaction. The effective model for solute concentration in the small-Péclet-number-regime is derived using asymptotic analysis with respect to $\varepsilon$. The obtained mathematical model clearly indicates the influence of the porous medium, chemical reaction and boundary distortion on the effective flow.

This is a joint work with Eduard Marušić-Paloka (University of Zagreb).

Fix a loop group LG, a level k∈ℕ, and let Repᵏ(LG) be corresponding category of positive energy representations.
For any pair of pants Σ (with complex structure in the interior and parametrized boundary), there is an associated functor Repᵏ(LG) × Repᵏ(LG) → Repᵏ(LG): (H,K) ↦ H⊠K, called the fusion product.

It had been widely expected (but never proven) that this operation should be unitary. Namely, that the choice of LG-invariant inner products on H and on K should induce an LG-invariant inner product on H⊠K. We show that this is not the case: there is an anomaly.
More precisely, there is an ℝ₊-torsor canonically associated to Σ. It is only after trivialising of this ℝ₊-torsor that the fusion product acquires an LG-invariant inner product. (The same statement applies when Σ is an arbitrary Riemann surface with boundary.)
Joint work with James Tener.

Having made sense of differential equations driven by rough paths, we now have a new set of models available but when it comes to calibrating them to data, the tools are still underdeveloped. I will present some results and discuss some challenges related to building these tools.

Aharoni and Berger conjectured that in any bipartite multigraph that is properly edge-coloured by n colours with at least n+1 edges of each colour there must be a matching that uses each colour exactly once (such a matching is called rainbow). This conjecture recently have been proved asymptotically by Pokrovskiy. In this talk, I will consider the same question without the bipartiteness assumption. It turns out that in any multigraph with bounded edge multiplicities, that is properly edge-coloured by n colours with at least n+o(n) edges of each colour, there must be a matching of size n-O(1) that uses each colour at most once. This is joint work with Peter Keevash.

Consider the Loewner equation associated to the upper-half plane. This is an equation originated from an extremal problem in complex analysis. Nowadays, it attracts a lot of attention due to its connection to probability. Normally this equation is driven by a real-valued function. In this talk, we will show that the equation still makes sense when being driven by a complex-valued function. We will relate this situation to the classical situation and also to complex dynamics. 

We show that a closed almost Kähler 4-manifold of globally constant holomorphic sectional curvature k<=0 with respect to the canonical Hermitian connection is automatically Kähler. The same result holds for k < 0 if we require in addition that the Ricci curvature is J-invariant. The proofs are based on the observation that such manifolds are self-dual, so that Chern–Weil theory implies useful integral formulas, which are then combined with results from Seiberg–Witten theory.

I will talk about a generalisation of the Seiberg-Witten equations introduced by Taubes and Pidstrygach, in dimension 3 and 4 respectively, where the spinor representation is replaced by a hyperKahler manifold admitting certain symmetries. I will discuss the 4-dimensional equations and their relation with the almost-Kahler geometry of the underlying 4-manifold. In particular, I will show that the equations can be interpreted in terms of a PDE for an almost-complex structure on 4-manifold. This generalises a result of Donaldson. 

In recognition of a lifetime's contribution across the mathematical sciences, we are initiating a series of annual Public Lectures in honour of Roger Penrose. The first lecture will be given by his long-time collaborator and friend Stephen Hawking.

Unfortunately the lecture is now sold out and we have a full waiting list. However, we will be podcasting the lecture live (and also via the University of Oxford Facebook page).

In certain business environments, it is essential to the success of the business that workers stick closely to their plans and are not distracted, diverted or stopped. A warehouse is a great example of this for businesses where customers order goods online and the merchants commit to delivery dates.  In a warehouse, somewhere, a team of workers are scheduled to pick the items which will make up those orders and get them shipped on time.  If the workers do not deliver to plan, then orders will not be shipped on time, reputations will be damaged, customer will be lost and companies will go out of business.

StayLinked builds software which measures what these warehouse workers do and measures the factors which cause them to be distracted, diverted or stopped.  We measure whenever they start or end a task or process (e.g. start an order, pick an item in an order, complete an order). Some of the influencing factors we measure include the way the worker interacts with the device (using keyboard, scanner, gesture), navigates through the application (screens 1-3-4-2 instead of 1-2-3-4), the performance of the battery (dead battery stops work), the performance of the network (connected to access point or not, high or low latency), the device types being used, device form factor, physical location (warehouse 1, warehouse 2), profile of worker, etc.

We are seeking to build a configurable real-time mathematical model which will allow us to take all these factors into account and confidently demonstrate a measure of their impact (positive or negative) on the business process and therefore on the worker’s productivity. We also want to alert operational staff as soon as we can identify that important events have happened.  These alerts can then be quickly acted upon and problems resolved at the earliest possible opportunity.

In this project, we would like to collaborate with the maths faculty to understand the appropriate mathematical techniques and tools to use to build this functionality.  This product is being used right now by our customers so it would also be a great opportunity for a student to quickly see the results of their work in action in a real-world environment.

Quiver varieties are an attractive research topic of many branches of contemporary mathematics - (geometric) representation theory, (hyper)Kähler differential geometry, (symplectic) algebraic geometry and quantum algebra.

In the talk, I will define different types of quiver varieties, along with some interesting examples. Afterwards, I will focus on Nakajima quiver varieties (hyperkähler moduli spaces obtained from framed-double-quiver representations), stating main results on their topology and geometry. If the time permits, I will say a bit about the symplectic topology of them.

Let k be a field of characteristic zero and K=k((t)). Semi-algebraic sets over K are boolean combinations of algebraic sets and sets defined by valuative inequalities. The associated Grothendieck ring has been studied by Hrushovski and Kazhdan who link it via motivic integration to the Grothendieck ring of varieties over k. I will present a morphism from the former to the Grothendieck ring of motives of rigid analytic varieties over K in the sense of Ayoub. This allows to refine the comparison by Ayoub, Ivorra and Sebag between motivic Milnor fibre and motivic nearby cycle functor.
 

Let k be a field of characteristic zero and K=k((t)). Semi-algebraic sets over K are boolean combinations of algebraic sets and sets defined by valuative inequalities. The associated Grothendieck ring has been studied by Hrushovski and Kazhdan who link it via motivic integration to the Grothendieck ring of varieties over k. I will present a morphism from the former to the Grothendieck ring of motives of rigid analytic varieties over K in the sense of Ayoub. This allows to refine the comparison by Ayoub, Ivorra and Sebag between motivic Milnor fibre and motivic nearby cycle functor.
 

Brain convolutions are specificity of mammals. Varying in intensity according to the animal species, it is measured by an index called the gyrification index, ratio between the effective surface of the cortex compared to its apparent surface. Its value is closed to 1 for rodents (smooth brain), 2.6 for new-borns and 5 for dolphins.  For humans, any significant deviation is a signature of a pathology occurring in fetal life, which can be detected now by magnetic resonance imaging (MRI). We propose a simple model of growth for a bilayer made of the grey and white matter, the grey matter being in cortical position. We analytically solved the Neo-Hookean approximation in the short and large wavelength limits. When the upper layer is softer than the bottom layer (possibly, the case of the human brain), the selection mechanism is dominated by the physical properties of the upper layer. When the anisotropy favours the growth tangentially as for the human brain, it decreases the threshold value for gyri formation. The gyrification index is predicted by a post-buckling analysis and compared with experimental data. We also discuss some pathologies in the model framework.

Several optimization problems combine nonlinear constraints with the integrality of a subset of variables. For an important class of problems  called Mixed Integer Second-Order Cone Optimization (MISOCO), with applications in facility location, robust optimization, and finance, among others, these nonlinear constraints are second-order (or Lorentz) cones.

For such problems, as for many discrete optimization problems, it is crucial to understand the properties of the union of two disjoint sets of feasible solutions. To this end, we apply the disjunctive programming paradigm to MISOCO and present conditions under which the convex hull of two disjoint sets can be obtained by intersecting the feasible set with a specially constructed second-order cone. Computational results show that such cone has a positive impact on the solution of MISOCO problems.

The Landau-Lifshitz-Gilbert equation (LLG) is a continuum model describing the dynamics for the spin in ferromagnetic materials. In the first part of this talk we describe our work concerning the properties and dynamical behaviour of the family of self-similar solutions under the one-dimensional LLG-equation.  Motivated by the properties of this family of self-similar solutions, in the second part of this talk we consider the Cauchy problem for the LLG-equation with Gilbert damping and provide a global well-posedness result provided that the BMO norm of the initial data is small.  Several consequences of this result will be also given.

If $G$ is an irreducible lattice in a semisimple Lie group, every action of $G$ on a tree has a global fixed point. I will give an elementary discussion of Y. Shalom's proof of this result, focussing on the case of $SL_2(\mathbb{R}) \times SL_2(\mathbb{R})$. Emphasis will be placed on the geometric aspects of the proof and on the importance of reduced cohomology, while other representation theoretic/functional analytic tools will be relegated to a couple of black boxes.

Modular forms holomorphic functions on the upper half of the complex plane, H, invariant under certain matrix transformations of H. The have a very rich structure - they form a graded algebra over C and come equipped with a family of linear operators called Hecke operators. We can also view them as functions on a Riemann surface, which we refer to as a modular curve. It transpires that the integral homology of this curve is equipped with such a rich structure that we can use it to compute modular forms in an algorithmic way. The modular symbols are a finite presentation for this homology, and we will explore this a little and their connection to modular symbols.

Hilbert's Nullstellensatz asserts the existence of a complex point satisfying lying on a given variety, provided there is no (ideal-theoretic) proof to the contrary.
I will describe an analogue for curves (of unbounded degree), with respect to conditions specifying that they lie on a given smooth variety, and have homology class
near a specified ray.   In particular, an analogue of the Lefschetz principle (relating large positive characteristic to characteristic zero) becomes available for such questions.
The proof is very close to a theorem of  Boucksom-Demailly-Pau-Peternell on moveable curves, but requires a certain sharpening.   This is part of a joint project with Itai Ben Yaacov, investigating the logic of the product formula; the algebro-geometric statement is needed for proving the existential closure of $\Cc(t)^{alg}$ in this language.  
 

Hilbert's Nullstellensatz asserts the existence of a complex point satisfying lying on a given variety, provided there is no (ideal-theoretic) proof to the contrary.
I will describe an analogue for curves (of unbounded degree), with respect to conditions specifying that they lie on a given smooth variety, and have homology class
near a specified ray.   In particular, an analogue of the Lefschetz principle (relating large positive characteristic to characteristic zero) becomes available for such questions.
The proof is very close to a theorem of  Boucksom-Demailly-Pau-Peternell on moveable curves, but requires a certain sharpening.   This is part of a joint project with Itai Ben Yaacov, investigating the logic of the product formula; the algebro-geometric statement is needed for proving the existential closure of $\Cc(t)^{alg}$ in this language. 

In this talk we introduce a novel dynamic programming (DP) approximation that exploits the inherent network structure present in revenue management problems. In particular, our approximation provides a new lower bound on the value function for the DP, which enables conservative revenue forecasts to be made. Existing state of the art approximations of the revenue management DP neglect the network structure, apportioning the prices of each product, whereas our proposed method does not: we partition the network of products into clusters by apportioning the capacities of resources. Our proposed approach allows, in principle, for better approximations of the DP to be made than the decomposition methods currently implemented in industry and we see it as an important stepping stone towards better approximate DP methods in practice.

A subset S of initially infected vertices of a graph G is called forcing if we can infect the entire graph by iteratively applying the following process. At each step, any infected vertex which has a unique uninfected neighbour, infects this neighbour. The forcing number of G is the minimum cardinality of a forcing set in G. It was introduced independently as a bound for the minimum rank of a graph, and as a tool in quantum information theory.

The focus of this talk is on the forcing number of the random graph. Furthermore, we will state our bounds on the forcing number of pseudorandom graphs and related problems. The results are joint work with Thomas Kalinowski and Benny Sudakov.

Dimer models with boundary were introduced in joint work with King and Marsh as a natural
generalisation of dimers. We use these to derive certain infinite dimensional algebras and
consider idempotent subalgebras w.r.t. the boundary.
The dimer models can be embedded in a surface with boundary. In the disk case, the
maximal CM modules over the boundary algebra are a Frobenius category which
categorifies the cluster structure of the Grassmannian.

 

One of the key challenges in revenue management is unconstraining demand data. Existing state of the art single-class unconstraining methods make restrictive assumptions about the form of the underlying demand and can perform poorly when applied to data which breaks these assumptions. In this talk, we propose a novel unconstraining method that uses Gaussian process (GP) regression. We develop a novel GP model by constructing and implementing a new non-stationary covariance function for the GP which enables it to learn and extrapolate the underlying demand trend. We show that this method can cope with important features of realistic demand data, including nonlinear demand trends, variations in total demand, lengthy periods of constraining, non-exponential inter-arrival times, and discontinuities/changepoints in demand data. In all such circumstances, our results indicate that GPs outperform existing single-class unconstraining methods.

Finding implementable descriptions of the possible configurations of a knotted DNA molecule has remarkable importance from a biological point of view, and it is a hard and well studied problem in mathematics.
Here we present two newly developed mathematical tools that describe the configuration space of knots and model the action of Type I and II Topoisomerases on a covalently closed circular DNA molecule: the Reidemeister graphs.
We determine some local and global properties of these graphs and prove that in one case the graph-isomorphism type is a complete knot invariant up to mirroring.
Finally, we indicate how the Reidemeister graphs can be used to infer information about the proteins' action.

In this talk I will present recent results obtained within the
framework of perturbative algebraic quantum field theory. This novel
approach to mathematical foundations of quantum field theory allows to
combine the axiomatic framework of algebraic QFT by Haag and Kastler with
perturbative methods. Recently also non-perturbative results have been
obtained within this approach. I will report on these results and present
new perspectives that they open for better understanding of foundations of
QFT.

We consider a couple of problems belonging to Random Geometry, and describe some new analytical challenges they pose for planar PDE's via Beltrami equations. The talk is based on joint work with various people including K. Astala, P. Jones, A. Kupiainen, Steffen Rohde and T. Tao.

In this talk I will construct a reduced tensor product of braided fusion categories containing a symmetric fusion category $\mathcal{A}$. This tensor product takes into account the relative braiding with respect to objects of $\mathcal{A}$ in these braided fusion categories. The resulting category is again a braided fusion category containing $\mathcal{A}$. This tensor product is inspired by the tensor product of $G$-equivariant once-extended three-dimensional quantum field theories, for a finite group $G$.
To define this reduced tensor product, we equip the Drinfeld centre $\mathcal{Z}(\mathcal{A})$ of the symmetric fusion category $\mathcal{A}$ with an unusual tensor product, making $\mathcal{Z}(\mathcal{A})$ into a 2-fold monoidal category. Using this 2-fold structure, we introduce a new type of category enriched over the Drinfeld centre to capture the braiding behaviour with respect to $\mathcal{A}$ in the braided fusion categories, and use this encoding to define the reduced tensor product.
 

In the talk, we discuss how to combine the recurrent neural network with the signature feature set to tackle the supervised learning problem where the input is a data stream. We will apply this method to different datasets, including the synthetic datasets( learning the solution to SDEs ) and empirical datasets(action recognition) and demonstrate the effectiveness of this method.

In this talk, we will revisit the proof of the large deviations principle of Wiener chaoses partially given by Borell, and then by Ledoux in its full form. We show that some heavy-tail phenomena observed in large deviations can be explained by the same mechanism as for the Wiener chaoses, meaning that the deviations are created, in a sense, by translations. More precisely, we prove a general large deviations principle for a certain class of functionals $f_n : \mathbb{R}^n \to \mathcal{X}$, where $\mathcal{X}$ is some metric space, under the probability measure $\nu_{\alpha}^n$, where $\nu_{\alpha} =Z_{\alpha}^{-1}e^{-|x|^{\alpha}}dx$, $\alpha \in (0,2]$, for which the large deviations are due to translations. We retrieve, as an application, the large deviations principles known for the so-called Wigner matrices without Gaussian tails of the empirical spectral measure, the largest eigenvalue, and traces of polynomials. We also apply our large deviations result to the last-passage time which yields a large deviations principle when the weight matrix has law $\mu_{\alpha}^{n^2}$, where $\mu_{\alpha}$ is the probability measure on $\mathbb{R}^+$ with density $2Z_{\alpha}^{-1}e^{-x^{\alpha}}$ when $\alpha \in (0,1)$.

While many cubic fourfolds are known to be rational, it is expected that the very general cubic fourfold is irrational (although none have been
proven to be so). There is a conjecture for precisely which cubics are rational, which can be expressed in Hodge-theoretic terms (by work of Hassett)
or in terms of derived categories (by work of Kuznetsov). The conjecture can be phrased as saying that one can associate a `noncommutative K3 surface' to any cubic fourfold, and the rational ones are precisely those for which this noncommutative K3 is `geometric', i.e., equivalent to an honest K3 surface. It turns out that the noncommutative K3 associated to a cubic fourfold has a conjectural symplectic mirror (due to  Batyrev-Borisov). In contrast to the algebraic side of the story, the mirror is always `geometric': i.e., it is always just an honest K3 surface equipped with an appropriate Kähler form. After explaining this background, I will state a theorem: homological mirror symmetry holds in this context (joint work with Ivan Smith).

I will talk about three-dimensional N=2 supersymmetric gauge theories on a class of Seifert manifold. More precisely, I will compute the supersymmetric partition functions and correlation functions of BPS loop operators on M_{g,p}, which is defined by a circle bundle of degree p over a genus g Riemann surface. I will also talk about four-dimensional uplift of this construction, which computes the generalized index of N=1 gauge theories defined on elliptic fiberation over genus g Riemann surface. We will find that the partition function or the index can be written as a sum over "Bethe vacua” of two-dimensional A-twisted theory obtained by a circle compactification. With this framework, I will show how the partition functions on manifolds with different topologies are related to each other. We will also find that these observables are very useful to study the action of Seiberg-like dualities on co-dimension two BPS operators.

Coding theory revolves around the question of what can be accomplished with only memory and redundancy. When we ask what enables the things that transmit and store information, we discover codes at work, connecting the world of geometry to the world of algorithms.

This talk will focus on those connections that link the real world of Euclidean geometry to the world of binary geometry that we associate with Hamming.

A cubic polynomial equation in four or more variables tends to have many integer solutions, while one in two variables has a limited number of such solutions. There is a body of work establishing results along these lines. On the other hand very little is known in the critical case of three variables. For special such cubics, which we call Markoff surfaces, a theory can be developed. We will review some of the tools used to deal with these and related problems.

Joint works with Bourgain/Gamburd and with Ghosh
 

Following several decades of development by applied mathematicians, models of ocean wave interactions with sea ice floes are now in high demand due to the rapid recent changes in the world’s sea ice cover. From a mathematical perspective, the models are of interest due to the thinness of the floes, leading to elastic responses of the floes to waves, and the vast number of floes that waves encounter. Existing models are typically based on linear theories, but the thinness of the floes leads to the unique and highly nonlinear phenomenon of overwash, where waves run over the floes, in doing so dissipating wave energy and impacting the floes thermodynamically. I will give an overview of methods developed for the wave-floe problem, and present a new, bespoke overwash model, along with supporting laboratory experiments and numerical CFD simulations.

Current medical practice is driven by the experience of clinicians, by the difficulties of integrating enormous amounts of complex and heterogeneous static and dynamic data and by clinical guidelines designed for the “average” patient. In this talk, I will describe some of my research on developing novel, specially-crafted machine learning theories, methods and systems aimed at extracting actionable intelligence from the wide variety of information that is becoming available (in electronic health records and elsewhere) and enabling every aspect of medical care to be personalized to the patient at hand. Because of the unique and complex characteristics of medical data and medical questions, many familiar machine-learning methods are inadequate.  My work therefore develops and applies novel machine learning theory and methods to construct risk scores, early warning systems and clinical decision support systems for screening and diagnosis and for prognosis and treatment.  This work achieves enormous improvements over current clinical practice and over existing state-of-the-art machine learning methods.  By design, these systems are easily interpretable and so allow clinicians to extract from data the necessary knowledge and representations to derive data-driven medical epistemology and to permit easy adoption in hospitals and clinical practice. My team has collaborated with researchers and clinicians in oncology, emergency care, cardiology, transplantation, internal medicine, etc. You can find more information about our past research in this area at: http://medianetlab.ee.ucla.edu/MedAdvance.

Christoph Siebenbrunner:

Clearing Algorithms and Network Centrality

I show that the solution of a standard clearing model commonly used in contagion analyses for financial systems can be expressed as a specific form of a generalized Katz centrality measure under conditions that correspond to a system-wide shock. This result provides a formal explanation for earlier empirical results which showed that Katz-type centrality measures are closely related to contagiousness. It also allows assessing the assumptions that one is making when using such centrality measures as systemic risk indicators. I conclude that these assumptions should be considered too strong and that, from a theoretical perspective, clearing models should be given preference over centrality measures in systemic risk analyses.


Andreas Sojmark:

An SPDE Model for Systemic Risk with Default Contagion

In this talk, I will present a structural model for systemic risk, phrased as an interacting particle system for $N$ financial institutions, where each institution is removed upon default and this has a contagious effect on the rest of the system. Moreover, the financial instituions display herding behavior and they are exposed to correlated noise, which turns out to be an important driver of the contagion mechanism. Ultimately, the motivation is to provide a clearer connection between the insights from dynamic mean field models and the detailed study of contagion in the (mostly static) network-based literature. Mathematically, we prove a propagation of chaos type result for the large population limit, where the limiting object is characterized as the unique solution to a nonlinear SPDE on the positive half-line with Dirichlet boundary. This is based on joint work with Ben Hambly and I will also point out some interesting future directions, which are part of ongoing work with Sean Ledger.