Prof. Dr. Pierre Colmez (CNRS)
Dr. Andrew Graham (University of Oxford)
Prof. Dr. Andreas Langer (University of Exeter)
Prof. Dr. Wieslawa Niziol (CNRS)
Abstract:
This is the fifth meeting of the Arithmetica Transalpina, a joint Number Theory seminar between ETH Zürich, Unidistance Suisse, the Universities of Milan, Padova, Genova and Vienna.To register, please send an email to arithmeticatransalpina@gmail.comThe deadline for registration is September 30.
Hugo Duminil-Copin (University of Geneva)
Towards a rigorous Coulomb Gas Formalism abstract
Abstract:
In this talk, we will explore the rich interplay between two-dimensional critical percolation models and the six-vertex model, a classical integrable random height model. By leveraging the remarkable symmetries and emergent structures that arise in the large-scale behavior of these systems, we will discuss how the so-called Coulomb Gas Formalism may be placed on rigorous mathematical foundations in this context. This perspective opens new pathways toward a deeper mathematical understanding of the phase transition of these models. The presentation is intended to be accessible to a broad mathematical audience.
11:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room D 16.2
Marcella Bonazzoli (Inria, Institut Polytechnique de Paris)
Abstract:
In this talk we are interested in reconstructing the interface between the concrete structure of a hydroelectric gravity dam and the underlying rock, using Full Waveform Inversion. Indeed, it appears that the roughness of the dam-rock interface has an effect on the sliding stability of gravity dams. We minimize a regularized misfit cost functional by computing its shape derivative and iteratively updating the interface shape by the gradient descent method. At each iteration, we simulate time-harmonic elasto-acoustic wave propagation models, coupling linear elasticity in the solid medium with acoustics in the reservoir. Numerical results using realistic noisy synthetic data demonstrate the method ability to accurately reconstruct the dam-rock interface, even with a limited number of measurements. This is joint work with Mohamed Aziz Boukraa, Lorenzo Audibert, Houssem Haddar and Denis Vautrin. For further information about the seminar, please visit this webpage .
11:00 • Universität Basel, Room 05.001, Spiegelgasse 5, 4051 Basel
Prof. Dr. Michael Novack (Louisiana State University,)
Special Analysis Seminar: Plateau\'s problem for soap films with positive volume: new directions abstract
Abstract:
We discuss a Plateau problem based on capillarity theory in which soap films are described as sets with small volume v that satisfy a spanning condition. Existence and interior regularity are understood for fixed v>0 and for the limiting Plateau problem, and so several questions arise regarding boundary regularity and the nature of the convergence to minimal surfaces as v approaches zero. We present ongoing joint work in these directions with Francesco Maggi (UT Austin) and Daniel Restrepo (Johns Hopkins).
12:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Ittai Rubinstein (MIT, US)
Data Attribution in High-Dimensions and without Strong Convexity abstract
Abstract:
Data attribution methods aim to quantify how training examples shape model predictions, supporting applications in interpretability, unlearning, and robustness. The dominant tools in practice are influence functions (IF) and Newton step (NS) approximations, yet their theoretical guarantees and practical accuracy have remained poorly understood. In this talk, I will present new analytic techniques that uncover the scaling laws of the approximation error of IF and NS. Our results improve on prior analyses both by establishing asymptotically sharper bounds and by avoiding dependence on the global strong convexity parameter, which is often prohibitively small in practice. These insights not only explain long-standing empirical observations—such as why and when NS is more accurate than IF—but also guide the design of new methods. As an application, I will present rescaled influence functions (RIF), a simple, drop-in replacement for IF that matches the efficiency of IF while achieving the accuracy of NS. I will discuss both theoretical advances and empirical results on real-world datasets. Together, these contributions provide a first principled understanding of data attribution methods and demonstrate how to turn this understanding into more reliable tools.
13:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 26.3
Istvan Kadar (ETH Zürich)
Smooth singularity formation for the energy critical wave equation without quantization I: Approximate solutions abstract
Abstract:
In this talk, I present ongoing work on singularity formation for the energy-critical nonlinear wave equation (NW) in three dimensions. In their foundational work, Krieger-Schlag-Tataru constructed non-smooth solutions to (NW) that develop singularities with arbitrarily fast, polynomial blow-up rates exceeding the self-similar rate. For comparison, their analogous results for the wave map equation stand in sharp contrast to the quantized blow-up rates for smooth solutions obtained by Raphaël-Rodnianski. In this talk, I discuss a new result establishing the existence of smooth approximate solutions to (NW) that develop singularities at a continuum of polynomial rates, achieved through the collapse of multiple solitons.
14:15 • EPF Lausanne, MA B1 11
Dr. Hyeonjun Park (Korea Institute for Advanced Study)
Lagrangian classes and GLSM CohFT abstract
Abstract:
In this talk, I will explain the construction of Lagrangian classes for perverse sheaves in cohomological Donaldson-Thomas theory, whose existence was conjectured by Joyce. The two key ingredients are a relative version of the DT perverse sheaves and a hyperbolic version of the dimensional reduction theorem. As a special case, we recover Borisov-Joyce/Oh-Thomas virtual classes in DT4 theory. As an application, I will explain how to construct cohomological field theories for gauged linear sigma models. This is joint work in progress with Adeel Khan, Tasuki Kinjo, and Pavel Safronov.
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Prof. Dr. Bastian Rieck (Universität Fribourg)
Abstract:
A large driver contributing to the undeniable success of deep-learning models is their ability to synthesize task-specific features from data. For a long time, the predominant belief was that \'given enough data, all features can be learned.\' However, as large language models are hitting diminishing returns in output quality while requiring an ever-increasing amount of training data and compute, new approaches are required. One promising avenue involves focusing more on aspects of modeling, which involves the development of novel inductive biases such as invariances that cannot be readily gleaned from the data.This approach is particularly useful for data sets that model real-world phenomena, as well as applications where data availability is scarce.Given their dual nature, geometry and topology provide a rich source of potential inductive biases. In this talk, I will present novel advances in harnessing multi-scale geometrical-topological characteristics of data. A special focus will be given to show how geometry and topology can improve representation learning tasks. Underscoring the generality of a hybrid geometrical-topological perspective, I will furthermore showcase applications from a diverse set of data domains, including point clouds, graphs, and higher-order combinatorial complexes.
16:15 • Universität Bern, Sidlerstrasse 5, 3012 Bern, Room B77