BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Erlangen AI Hub - ECPv6.18.4.1//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Erlangen AI Hub
X-ORIGINAL-URL:https://erlangenhub.ox.ac.uk
X-WR-CALDESC:Events for Erlangen AI Hub
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:20250330T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:20251026T010000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:20260329T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:20261025T010000
END:STANDARD
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:20270328T010000
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:20271031T010000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20261117T130000
DTEND;TZID=Europe/London:20261117T140000
DTSTAMP:20261112T143226Z
CREATED:20261023T160127Z
LAST-MODIFIED:20261112T143226Z
UID:1345-1794920400-1794924000@erlangenhub.ox.ac.uk
SUMMARY:Singular Learning Theory: A Mathematical Introduction
DESCRIPTION:Speaker: Kamillo Ferry\, Imperial College London \nTitle: Singular Learning Theory: A Mathematical Introduction \nAbstract: Classical statistics rests on an assumption that modern machine learning almost never satisfies: that a model’s parameters can be identified from data\, so that near the truth the model looks locally like a Euclidean space of known dimension. In neural networks\, mixture models and matrix factorisations\, many different parameters describe the same model. The set of optimal parameters is then a singular real algebraic variety\, and familiar tools such as the Fisher information and the BIC break down. Watanabe’s singular learning theory shows that the asymptotics of Bayesian learning are nevertheless governed by a single invariant of this singularity\, the real log canonical threshold\, which can be read off from a resolution of singularities. This talk is an introduction to the mathematics of the theory: why algebraic geometry enters\, how resolution of singularities turns an analytic question into a combinatorial one\, and what can be computed exactly. I will work through examples\, from toy models to reduced-rank regression. No background in algebraic geometry is assumed. \nJoin: https://teams.microsoft.com/meet/3395756456794587?p=KXola0KUUnCaaSx1QX \nWe ask that you please familiarise yourself with the Hub’s code of conduct prior to attendance. Please note\, this talk will be recorded.
URL:https://erlangenhub.ox.ac.uk/event/singular-learning-theory-a-mathematical-introduction/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/10/Hub-seminar-series-templates-4.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20261020T120000
DTEND;TZID=Europe/London:20261020T130000
DTSTAMP:20261023T130410Z
CREATED:20261007T153123Z
LAST-MODIFIED:20261023T130410Z
UID:1309-1792497600-1792501200@erlangenhub.ox.ac.uk
SUMMARY:Seminar: Best-Effort Synthesis for Temporally Extended Goals in Nondeterministic Planning Domains
DESCRIPTION:Tuesday 20th October 12:00-13:00 \nSpeaker: Gianmarco Parretti\, University of Oxford \nTitle: Best-Effort Synthesis for Temporally Extended Goals in Nondeterministic Planning Domains \nAbstract: 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\, the agent must compute a strategy that ensures the satisfaction of the goal regardless of how the external environment unfolds its nondeterminism – called a winning strategy. While it is highly desirable that the agent uses a winning strategy\, the problem is that such strategies do not always exist. In this talk\, I will present best-effort strategies\, a form of strategy ensuring that – when a winning strategy does not exist – the agent does its best to achieve the goal. I will consider the setting of an autonomous agent that operates in a Fully Observable Nondeterministic (FOND) Domain and whose goal is expressed in Linear Temporal Logic on finite traces (LTLf). For this setting\, I will show that best-effort strategies have some notable properties: (i) they always exist; (ii) if a winning strategy exists\, the best-effort strategies are exactly the winning strategies; and (iii) they can be computed in worst-case 2EXPTIME (though often much less in practice)\, i.e.\, the same as winning strategies. In fact\, I will show that these properties generalize to the setting where the agent reasons simultaneously over multiple increasingly nondeterministic planning domains (aka multi-tier planning domains). Finally\, I will detail on how best-effort strategies can be used to monitor the behaviour of an autonomous agent operating in a FOND domain\, but whose strategy is not readily available for inspection. \nJoin: https://teams.microsoft.com/meet/3542095257933196?p=3PGx1hha2nCmpU8mRJ \nWe ask that you please familiarise yourself with the Hub’s code of conduct prior to attendance.\nPlease note\, this talk will be recorded.
URL:https://erlangenhub.ox.ac.uk/event/seminar-best-effort-synthesis-for-temporally-extended-goals-in-nondeterministic-planning-domains/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/10/Hub-seminar-series-templates-2.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20261015T140000
DTEND;TZID=Europe/London:20261015T150000
DTSTAMP:20261007T121404Z
CREATED:20261007T121404Z
LAST-MODIFIED:20261007T121404Z
UID:1299-1792072800-1792076400@erlangenhub.ox.ac.uk
SUMMARY:Seminar: Patnaik-Pearson intrinsic dimension for internal representations of neural networks
DESCRIPTION:Thursday 15th October 14:00-15:00 \nSpeaker: Dr. Tom Hadfield\, Imperial College London \nTitle: Patnaik-Pearson intrinsic dimension for internal representations of neural networks \nAbstract: 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 this comes from the HTSR and SETOL work of Martin\, Mahoney and Hinrichs\, combined with the TwoNN intrinsic dimension estimator of Facco et al. We prove various properties of this intrinsic dimension estimator. Treating weight matrices of neural networks as data manifolds\, for weight matrices whose empirical spectral density follows a Pareto (power law) distribution\, we relate the Patnaik-Pearson dimension to the HTSR and SETOL analysis\, and show that critical values of the tail exponent coincide for the two approaches. Using a combination of theoretical and numerical techniques\, including Voiculescu’s free probability\, we study the behaviour of the Patnaik-Pearson dimension of a data manifold under the transformations typical to neural networks. https://arxiv.org/abs/2606.19268 \nhttps://teams.microsoft.com/l/meetup-join/19%3ameeting_MzY2ZDkwMTAtNGE5OC00Yzc1LWFjNzItODU4YmYyMjMwZjc3%40thread.v2/0?context=%7b%22Tid%22%3a%222b897507-ee8c-4575-830b-4f8267c3d307%22%2c%22Oid%22%3a%22311f0e52-bbf7-47a9-8cc6-752fcc16adcc%22%7d \nWe ask that you please familiarise yourself with the Hub’s code of conduct prior to attendance.\nPlease note\, this talk will be recorded.
URL:https://erlangenhub.ox.ac.uk/event/seminar-patnaik-pearson-intrinsic-dimension-for-internal-representations-of-neural-networks/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/10/Hub-seminar-series-templates-1.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260901T100000
DTEND;TZID=Europe/London:20260903T153000
DTSTAMP:20261026T135413Z
CREATED:20260211T124940Z
LAST-MODIFIED:20261026T135413Z
UID:811-1788256800-1788449400@erlangenhub.ox.ac.uk
SUMMARY:Mathematical Foundations of AI: The Erlangen Hub Conference 2026
DESCRIPTION: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 researchers in geometry\, algebra\, topology\, probability\, dynamical systems\, machine learning and related fields for an interdisciplinary exploration of the mathematical principles underlying modern AI. The conference aims to foster new connections across disciplines and advance the mathematical foundations needed to build more reliable\, robust\, and trustworthy AI systems. \nConfirmed speakers: \n\nErik Bekkers\, Associate Professor\, University of Amsterdam (AMLab\, UvA)\nMichael Bronstein\, DeepMind Professor of AI and Erlangen AI Hub Director\, University of Oxford\nRebekka Burkholz\, Faculty in Machine Learning\, CISPA Helmholtz Center for Information Security\nKevin Buzzard\, Professor of Mathematics\, Imperial College London\nCoralia Cartis\, Professor of Numerical Optimization\, Mathematical Institute\, University of Oxford\nOliver Clarke\, Erlangen AI Hub PDRA\, Durham University\nHolly Francois\, AI Principal Online Safety Technology\, Ofcom\nPeter Grindrod\, Professor of Mathematics\, Mathematical Institute\, University of Oxford\nMaz Hardey\, Professor\, Durham University Business School\nKathryn Hess\, Professor of Mathematics and Life Sciences\, EPFL\nStefanie Jegelka\, Visiting Professor at MIT EECS\, and a Humboldt Professor at TU Munich\nRoland Kwitt\, Full Professor of Machine Learning at University of Salzburg\nRichard Lane\, Principal Data Scientist\, QinetiQ\nEng-Jon Ong\, Erlangen AI Hub PDRA\, Queen Mary University of London\nYue Ren\, Associate Professor\, Durham University \nTatiana Shavrina\, Research Scientist Manager\, Meta\nSuvrit Sra\, Associate Professor of EECS and Professor for Resource Aware Machine Learning at TU Munich\nMarika Taylor\, Head of College of Engineering and Physical Sciences and Pro Vice Chancellor\nQiquan Wang\, Erlangen AI Hub PDRA\, Queen Mary University of London\nGeorge Williamson\, CEO\, The Alan Turing Institute\nArne Wolf\, PhD student\, Imperial College London\nKa Man (Ambrose) Yim\, Erlangen AI Hub PDRA\, University of Oxford\n\nFor further details of speakers click here.
URL:https://erlangenhub.ox.ac.uk/event/the-erlangen-hub-conference-2026/
LOCATION:Mathematical Institute\, University of Oxford
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/02/Erlangen-Hub-conference-artwork-MI-building-1.png
ORGANIZER;CN="Erlangen AI Hub":MAILTO:erlangenhub@cs.ox.ac.uk
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260701T130000
DTEND;TZID=Europe/London:20260701T140000
DTSTAMP:20261023T130428Z
CREATED:20260619T133341Z
LAST-MODIFIED:20261023T130428Z
UID:958-1782910800-1782914400@erlangenhub.ox.ac.uk
SUMMARY:Erlangen Hub Seminar: Beyond Second-Order Methods: Sum-of-Squares\, Tensor Methods\, and AI for Polynomial Optimization\, Kate Wenqi Zhu
DESCRIPTION: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. \nIn this talk\, we will extend this connection to new classes of multivariate quartically regularized polynomials arising as subproblems in high-order tensor methods for nonconvex optimization. In particular\,we will show that\, under mild conditions and for sufficiently large regularization parameters\, globally nonnegative quartically regularized cubic polynomials admit exact SoS certificates. We will also discuss structured subclasses for which this equivalence holds for every positive regularization parameter\, together with counterexamples demonstrating the crucial role played by Euclidean regularization. These results lead naturally to tractable high-order optimization frameworks based on SoS and SoS-convex Taylor models\, in which each regularized polynomial subproblem can be solved in polynomial time. Finally\, we will briefly discuss an AI-integrated approach to polynomial nonnegativity and SoS reasoning\, including a specialised dataset and fine-tuned language model designed to tackle these computationally challenging mathematical problems. \nRegister for this seminar here.
URL:https://erlangenhub.ox.ac.uk/event/erlangen-hub-seminar-beyond-second-order-methods-sum-of-squares-tensor-methods-and-ai-for-polynomial-optimization-kate-wenqi-zhu/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/06/Hub-seminar-series-1-7.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260617T130000
DTEND;TZID=Europe/London:20260617T140000
DTSTAMP:20260618T132124Z
CREATED:20260528T152642Z
LAST-MODIFIED:20260618T132124Z
UID:928-1781701200-1781704800@erlangenhub.ox.ac.uk
SUMMARY:Erlangen AI Hub Seminar: A Topological Lens on Neural Networks\, Qiquan Wang
DESCRIPTION: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 on large language model latent spaces using persistent homology. \nRegistration for this event has now closed.
URL:https://erlangenhub.ox.ac.uk/event/erlangen-ai-hub-seminar-a-topological-lens-on-neural-networks-qiquan-wang/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/05/Seminar-slide-3.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260529T120000
DTEND;TZID=Europe/London:20260529T130000
DTSTAMP:20260529T103643Z
CREATED:20260515T100028Z
LAST-MODIFIED:20260529T103643Z
UID:909-1780056000-1780059600@erlangenhub.ox.ac.uk
SUMMARY:Erlangen AI Hub Seminar: Numerically verified proofs aided by machine learning\, Daniel Platt
DESCRIPTION: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 classical problem of root enclosures for polynomials in one variable; (2) the main example is the Nirenberg problem\, where in arXiv:2602.12368 a PDE was solved using a PINN\, and in arXiv:2603.29544 I proved that there exists a genuine solution to the PDE; (3) another small application that’s to be decided\, possibly related to topological data analysis. \nRegister for this event has now closed.
URL:https://erlangenhub.ox.ac.uk/event/erlangen-ai-hub-seminar-numerically-verified-proofs-aided-by-machine-learning-daniel-platt/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/05/Seminar-slide-1-1.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260430T130000
DTEND;TZID=Europe/London:20260430T140000
DTSTAMP:20260417T164508Z
CREATED:20260413T182326Z
LAST-MODIFIED:20260417T164508Z
UID:876-1777554000-1777557600@erlangenhub.ox.ac.uk
SUMMARY:Erlangen AI Hub Seminar: Geometry\, Complexity\, and Generalization in Learning Systems
DESCRIPTION: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 and double descent\, and refuting a common criticism of sharpness-based generalization measures based on their lack of parameterization-invariance. \nRegister here: Erlangen AI Hub Seminar: Geometry\, Complexity\, and Generalization in Learning Systems
URL:https://erlangenhub.ox.ac.uk/event/erlangen-ai-hub-seminar-geometry-complexity-and-generalization-in-learning-systems/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260416T130000
DTEND;TZID=Europe/London:20260416T140000
DTSTAMP:20260413T182732Z
CREATED:20260413T182649Z
LAST-MODIFIED:20260413T182732Z
UID:877-1776344400-1776348000@erlangenhub.ox.ac.uk
SUMMARY:Erlangen AI Hub Seminar: Estimating Intrinsic Dimensionality with L2N2
DESCRIPTION: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.\n  \nWe present a theoretical framework showing that L2N2 is universal\, in the sense that it provably converges to the true intrinsic dimensionality independently of the underlying data distribution. We complement this analysis with extensive experiments on synthetic benchmark manifolds\, demonstrating strong empirical performance.\nBeyond controlled settings\, we show how L2N2 can be applied to real-world data\, including estimating the intrinsic dimensionality of datasets such as MNIST via autoencoder representations. Finally\, we illustrate how L2N2 provides new insights into deep learning by tracking how the intrinsic dimensionality of neural network layer representations evolves during training\, particularly on the phenomenon of grokking. \nRegister here: Erlangen AI Hub Seminar: Estimating Intrinsic Dimensionality with L2N2 (Eng-Jon Ong)
URL:https://erlangenhub.ox.ac.uk/event/erlangen-ai-hub-seminar-estimating-intrinsic-dimensionality-with-l2n2/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260323T110000
DTEND;TZID=Europe/London:20260323T120000
DTSTAMP:20260323T083553Z
CREATED:20260212T144224Z
LAST-MODIFIED:20260323T083553Z
UID:820-1774263600-1774267200@erlangenhub.ox.ac.uk
SUMMARY:Erlangen Hub Seminar: Riemannian Neural Optimal Transport\, Alessandro Micheli
DESCRIPTION: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. \nIn this talk\, Alessandro Micheli will show that discretisation-based OT methods on manifolds inherently face severe dimensionality scaling limitations. To address this\, he introduces Riemannian Neural Optimal Transport (RNOT)\, a continuous neural parameterisation of OT maps that avoids discretisation and incorporates geometric structure directly. Under mild regularity assumptions\, RNOT achieves sub-exponential complexity in the manifold dimension. Empirical results on synthetic and real datasets demonstrate improved scalability and competitive performance relative to existing approaches. \nRegister your interest now: \nRegistration for this event has now closed.
URL:https://erlangenhub.ox.ac.uk/event/erlangen-hub-seminar-riemannian-neural-optimal-transport/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2026/02/Hub-seminar-series-1-3.png
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260309T110000
DTEND;TZID=Europe/London:20260309T120000
DTSTAMP:20260308T170842Z
CREATED:20260216T141716Z
LAST-MODIFIED:20260308T170842Z
UID:836-1773054000-1773057600@erlangenhub.ox.ac.uk
SUMMARY:Erlangen Hub Seminar: Computing Diffusion Geometry\, Iolo Jones
DESCRIPTION:Calculus and geometry are ubiquitous in the theoretical modelling of scientific phenomena\, but have historically been very challenging to apply directly to real data as statistics. Diffusion geometry is a new theory that reformulates classical calculus and geometry in terms of a diffusion process\, allowing these theories to generalise beyond manifolds and be computed from data. \nIn this talk\, Iolo Jones will describe a new\, simple computational framework for diffusion geometry that substantially broadens its practical scope and improves its precision\, robustness to noise\, and computational complexity. He and colleagues present a range of new computational methods\, including all the standard objects from vector calculus and Riemannian geometry\, and apply them to solve spatial PDEs and vector field flows\, find geodesic (intrinsic) distances\, curvature\, and several new topological tools like de Rham cohomology\, circular coordinates\, and Morse theory. These methods are data-driven\, scalable\, and can exploit highly optimised numerical tools for linear algebra. \nRegistration for this event has now closed.
URL:https://erlangenhub.ox.ac.uk/event/erlangen-hub-seminar-computing-diffusion-geometry/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Europe/London:20260123T150000
DTEND;TZID=Europe/London:20260123T160000
DTSTAMP:20260107T142313Z
CREATED:20251215T144719Z
LAST-MODIFIED:20260107T142313Z
UID:680-1769180400-1769184000@erlangenhub.ox.ac.uk
SUMMARY:Erlangen Seminar Series: Studying the Geometry of Loops on the Möbius Band
DESCRIPTION:Date: Friday 23 January\, from 3–4pm \nLocation:  Online \nTitle:  Studying the Geometry of Loops on the Mobius Band \nFrom contours of 2D shapes to knotted proteins\, many areas of statistics and physical sciences contend with data in the form of geometric loops — an embedding of S1 into a Euclidean space. Similar to many problems in geometric deep learning\, we desire a representation of the geometry of the loop that is invariant under Euclidean symmetries\, and indifferent to how the loop is parametrised. In this joint work with James Binnie (Cardiff) and Otto Sumray (MPI-Dresden)\, we propose a novel representation of a loop’s geometry by representing the distance matrix of the loop as a Morse function on a Möbius band — the two point configuration space of S1. By considering the persistent homology of this function\, we obtain a representation of the loop’s geometry that has the desired symmetry invariance. We show that this feature map is stable with respect to perturbations\, and give a geometric interpretation of the features obtained. \nTo join us on Teams please fill out the following registration link: https://forms.office.com/e/MPBYstFWun
URL:https://erlangenhub.ox.ac.uk/event/erlangen-seminar-series-studying-the-geometry-of-loops-on-the-mobius-band/
ATTACH;FMTTYPE=image/png:https://erlangenhub.ox.ac.uk/wp-content/uploads/2025/12/Hub-seminar-series-1-1.png
END:VEVENT
END:VCALENDAR