• Erlangen Hub Seminar: Riemannian Neural Optimal Transport, Alessandro Micheli

    Computational optimal transport (OT) provides a principled framework for generative modelling. Neural OT methods learn transport maps from data using neural networks and can be evaluated out of sample after training; however, existing approaches are largely restricted to Euclidean settings. Extending neural OT to high-dimensional Riemannian manifolds presents significant theoretical and computational challenges. In this …

  • Erlangen AI Hub Seminar: Estimating Intrinsic Dimensionality with L2N2

    Estimating the intrinsic dimensionality (ID) of data is a fundamental problem in machine learning and computer vision, as it reveals the true degrees of freedom underlying high-dimensional observations. In this talk, Eng-Jon Ong (QMUL) introduces L2N2, a simple yet powerful ID estimator based on nearest-neighbour distance ratios that achieves state-of-the-art performance with minimal computational overhead. …

  • Erlangen AI Hub Seminar: Geometry, Complexity, and Generalization in Learning Systems

    Compression-based complexity measures have been used to construct non-vacuous generalization bounds for deep neural networks. In this talk, Branton DeMoss (University of Oxford) will discuss the relationship between compression, complexity, and the geometry of the loss landscape. Using a geometric complexity measure to track memorization and generalization in some pathological deep learning phenomena like grokking …

  • Erlangen AI Hub Seminar: Numerically verified proofs aided by machine learning, Daniel Platt

    Many methods are available to approximately solve all sorts of equations: ODEs, PDEs, polynomial systems, algebraic equations. Through recent advances it has become possible to turn these approximate solutions into rigorous existence proofs in some cases. I will explain how numerically verified proofs work and present several examples of them: (1) to get started, the …

  • Erlangen AI Hub Seminar: A Topological Lens on Neural Networks, Qiquan Wang

    Neural networks exhibit rich geometric and topological structure across their weights, activations, and representation spaces. This talk surveys a number of methods for probing these structures, providing complementary perspectives on how representations evolve across depth, encode structure, and respond to perturbations. I will highlight a recent piece of work studying the effect of adversarial influence …

  • Erlangen Hub Seminar: Beyond Second-Order Methods: Sum-of-Squares, Tensor Methods, and AI for Polynomial Optimization, Kate Wenqi Zhu

    Hilbert’s seventeenth problem famously showed that not every nonnegative polynomial admits a sum-of-squares (SoS) representation. Hilbert also identified several special classes for which nonnegativity and SoS are equivalent, including univariate polynomials, quadratic polynomials, and bivariate quartics. In this talk, we will extend this connection to new classes of multivariate quartically regularized polynomials arising as subproblems …

  • Mathematical Foundations of AI:
    The Erlangen Hub Conference 2026

    Mathematical Institute University of Oxford

    Artificial intelligence has achieved remarkable success, yet we still lack a rigorous mathematical understanding of why modern AI systems work, and when they fail. Closing this gap is one of the defining scientific challenges at the intersection of mathematics and computer science. The Mathematical Foundations of AI: Erlangen Hub Conference 2026 will bring together leading …

  • Seminar: Patnaik-Pearson intrinsic dimension for internal representations of neural networks

    Thursday 15th October 14:00-15:00 Speaker: Dr. Tom Hadfield, Imperial College London Title: Patnaik-Pearson intrinsic dimension for internal representations of neural networks Abstract: We define a new measure of intrinsic dimension of a data manifold, which we call the Patnaik-Pearson dimension, and apply this to internal representations of neural networks, in particular transformers. The inspiration for …

  • Seminar: Best-Effort Synthesis for Temporally Extended Goals in Nondeterministic Planning Domains

    Tuesday 20th October 12:00-13:00 Speaker: Gianmarco Parretti, University of Oxford Title: Best-Effort Synthesis for Temporally Extended Goals in Nondeterministic Planning Domains Abstract: In recent years, (Reactive) Synthesis has emerged as a powerful framework for the automated development of autonomous AI agents: given a specification of its goal and of the environment in which it operates, …

  • Singular Learning Theory: A Mathematical Introduction

    This talk is an introduction to the mathematics of singular learning theory: why algebraic geometry enters, how resolution of singularities turns an analytic question into a combinatorial one, and what can be computed exactly.