Pranav Nuti (Stanford University, Stanford, USA)
Online selection problems and how Ramsey’s theorem solves a card game abstract
Abstract:
In a basic online selection problem, we observe a sequence of numbers one-by-one, and we have to make an irrevocable decision to select one of the numbers. Both the secretary problem and the prophet problem are of this type, but they differ along a few dimensions: the order in which the numbers appear, how much information we have about the numbers before we observe them, and the objective we desire to maximize. Several models that are in between the secretary and prophet problems along these dimensions have previously been investigated. In this talk, I will say a little bit about these different models, and then explain how we can use Ramsey’s theorem (for hypergraphs) to obtain optimal results for one such model. This model has an interpretation as a simple card game.
12:20 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 19.2
Sebastian Goldt (University of Cambridge)
The interplay between data structure and neural networks: going beyond Gaussian models abstract
Abstract:
Neural networks are powerful feature extractors - but what do they actually learn from their data? We discuss two recent works on this question, with a focus on the importance of non-Gaussian statistics for neural networks. We first develop a simple model for images and show that a neural network trained on these images can learn a convolution from scratch. This pattern-formation process is driven by a combination of translation-invariance of the "images" and the non-Gaussian, higher-order statistics of the inputs. Second, we conjecture a "distributional simplicity bias" whereby neural networks learn increasingly complex distributions of their inputs during training. We present analytical and experimental evidence for this conjecture, going from a simple perceptron up to deep ResNets and visual transformers.
13:15 • EPF Lausanne, Salle GA 3 21
Elia Bruè (Scuola Normale Superiore di Pisa)
Instability and non-uniqueness in fluid dynamics abstract
Abstract:
The incompressible Navier-Stokes and Euler equations are fundamental for understanding basic phenomena in fluid dynamics, but many fundamental aspects of these equations still need to be fully understood.In particular, the question of well-posedness and uniqueness within physically relevant classes of solutions is not entirely resolved.Of particular importance are the class of Leray solutions to the Navier-Stokes equations and the class of solutions to the 2d Euler equations with L^p vorticity.
This talk outlines recent progress toward their understanding and discusses newly discovered nonunique solutions.
It is based on joint works with D. Albritton and M. Colombo.
14:15 • EPF Lausanne, Salle CM 1 113
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Equivariant Segre and Verlinde series for quot-schemes II
16:00 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43