TBA

Over the last decade, adversarial attack algorithms have revealed instabilities in artificial intelligence (AI) tools. These algorithms raise issues regarding safety, reliability and interpretability; especially in high risk settings. Mathematics is at the heart of this landscape, with ideas from  numerical analysis, optimization, and high dimensional stochastic analysis playing key roles. From a practical perspective, there has been a war of escalation between those developing attack and defence strategies. At a more theoretical level, researchers have also studied bigger picture questions concerning the existence and computability of successful attacks. I will present examples of attack algorithms for neural networks in image classification, for transformer models in optical character recognition and for large language models. I will also show how recent generative diffusion models can be used adversarially. From a more theoretical perspective, I will outline recent results on the overarching question of whether, under reasonable assumptions, it is inevitable that AI tools will be vulnerable to attack.

How do you efficiently look for jobs?

How can you make the most of careers fairs?

What makes a CV or cover letter stand out?

Get practical advice and bring your questions!

Transfer learning is a machine learning technique that leverages knowledge acquired in one domain to improve learning in another, related task. It is a foundational method underlying the success of large language models (LLMs) such as GPT and BERT, which were initially trained for specific tasks. In this talk, I will demonstrate how reinforcement learning (RL), particularly continuous time RL, can benefit from incorporating transfer learning techniques, especially with respect to convergence analysis. I will also show how this analysis naturally yields a simple corollary concerning the stability of score-based generative diffusion models.

Based on joint work with Zijiu Lyu of UC Berkeley.

A 2-connected, rational homotopy 7-sphere is classified up to diffeomorphism by three invariants: its (finite) 4th cohomology group, its q-invariant and its Eells-Kuiper invariant.  The q-invariant is a quadratic refinement of the linking form and determines the homeomorphism type, while the Eells-Kuiper invariant then pins down the diffeomorphism type.  In this talk, I will discuss the diffeomorphism classification of a certain family of non-negatively curved, 2-connected, rational homotopy 7-spheres, discovered by Sebastian Goette, Krishnan Shankar and myself, which contains, in particular, all $S^3$-bundles over $S^4$ and all exotic 7-spheres.

In this talk, we use type theory to construct a family of depth $\frac{1}{N}$ minimax supercuspidal representations of $p$-adic $\text{GL}(2N, F)$ which we call \textit{middle supercuspidal representations}. These supercuspidals may be viewed as a natural generalization of simple supercuspidal representations, i.e. those supercuspidals of minimal positive depth. Via explicit computations of twisted gamma factors, we show that middle supercuspidal representations may be uniquely determined through twisting by quasi-characters of $F^{\times}$ and simple supercuspidal representations of $\text{GL}(N, F)$. Furthermore, we pose a conjecture which refines the local converse theorem for general supercuspidal representations of $\text{GL}(n, F)$.

to follow

TBA

Optimal damping consists of identifying a viscosity vector that maximizes the decay rate of a mechanical system's response. This can be rephrased as minimizing the trace of the solution of a Lyapunov equation whose coefficient matrix, representing the system dynamics, depends on the dampers' viscosities. The latter must be nonnegative for a physically meaningful solution, and the system must be asymptotically stable at the solution.

In this talk, we present conditions under which the system is never stable or may not be stable for certain values of the viscosity vector, and, in the latter case, discuss how to modify the constraints so as to guarantee stability. We show that the KKT conditions of our nonlinear optimization problem are equivalent to a viscosity-dependent nonlinear residual function that is equal to zero at an optimal viscosity vector. To minimize this residual function, we propose a Barzilai-Borwein residual minimization algorithm (BBRMA) and a spectral projection gradient algorithm (SPG). The efficiency of both algorithms relies on a fast computation of the gradient for BBRMA, and both the objective function and its gradient for SPG. By fully exploiting the low-rank structure of the problem, we show how to compute these in $O(n^2)$ operations, $n$ being the size of the mechanical system.

This is joint work with Qingna Li (Beijing Institute of Technology).

TBA

Cell and tissue movement during development, immune response, and cancer invasion depends on chemical or mechanical guidance cues. In many systems, this guidance arises not from long-range, pre-patterned cues but from self-generated gradients locally shaped by cells. However, how heterogeneous cell mixtures coordinate their migration by self-generated gradients remains largely unexplored. In this talk, I will first summarize our recent discovery that immune cells steer their long-range migration using self-generated chemotactic cues (Alanko et al., 2023). I will then introduce a multi-component Keller-Segel model that describes migration and patterning strategies of heterogeneous cell populations (Ucar et al., 2025). Our model predicts that the relative chemotactic sensitivities of different cell populations determine the shape and speed of traveling density waves, while boundary conditions such as external cell and attractant reservoirs substantially influence the migration dynamics. We quantitatively corroborate these predictions with in vitro experiments on co-migrating immune cell mixtures. Interestingly, immune cell co-migration occurs near the optimal parameter regime predicted by theory for coupled and colocalized migration. Finally, I will discuss the role of mechanical interactions, revealing a non-trivial interplay between chemotactic and mechanical non-reciprocity in driving collective migration.
 

This session will look at how you can get the most out of your lectures and tutorials. We’ll talk about how to prepare effectively, make lectures more productive, and understand what tutors expect from you during tutorials. You’ll leave with practical tips to help you study more confidently and make your learning time count.

This session is likely to be most relevant for first-year undergraduates, but all are welcome.

Mathematicians often like to think of maths as objective. Science communicator Hana Ayoob joins us to discuss how the fact that humans do maths means that the ways maths is developed, used, and communicated are not neutral.

Modern deep learning methods provide the state-of-the-art in image reconstruction in most areas of computational imaging. However, such techniques are very data hungry and in a number of key imaging problems access to ground truth data is challenging if not impossible. This has led to the emergence of a range of self-supervised learning algorithms for imaging that attempt to learn to image without ground truth data. 

In this talk I will review some of the existing techniques and look at what is and might be possible in self-supervised imaging.

I will relate two notorious open questions in low-dimensional topology.  The first asks whether every hyperbolic group is residually finite. The second, the congruence subgroup property, relates the finite-index subgroups of mapping class groups to the topology of the underlying surface. I will explain why, if every hyperbolic group is residually finite, then mapping class groups enjoy the congruence subgroup property. Time permitting, I may give some further applications to the question of whether hyperbolic 3-manifolds are determined by the finite quotients of their fundamental groups.

Equations of fluctuating hydrodynamics, also called Dean-Kawasaki type equations, are stochastic PDEs describing the evolution of finitely many interacting particles which obey a Langevin equation. First, we give a mathematical derivation for such equations. The focus is on systems of interacting particles described by second order Langevin equations. For such systems,  the equations of fluctuating hydrodynamics are a stochastic variant of Vlasov-Fokker-Planck equations, where the noise is white in space and time, conservative and multiplicative. We show a dichotomy previously known for purely diffusive systems holds here as well: Solutions exist only for suitable atomic initial data, but provably not for any other initial data. The class of systems covered includes several models of active matter. We will also discuss regularisations, where existence results hold under weaker assumptions. 

to follow