Dr. Gareth Peters (University College London)
Statistical Machine Learning Methods for Risk and Insurance abstract
Abstract:
In this lecture series, we will explore topics of machine learning, from a statistical perspective. The topics selected will have direct relevance to modern data analytic methods in risk management and insurance. Core topics to be discussed will be covered in three parts:
Part 1. Working with Complex Data Types: topics will include:
- feature extraction methods from non-standard data types
- dealing with missingness, outliers, robust probabilistic feature extraction,
- topological feature extraction,
- dimension reduction methods (PCA, PPCA, ICA and functional versions).
Part 2. Regression models and Time series: topics will include- State Space models (reduced rank regressions and cointegration)
- GAMLSS and time series extensions;
- Covariance Regressions
- Generalized count processes with attributes such as long-memory, self-excitation, stochastic intensity and dependence, and
- Quantile regression and time series
- Gaussian processes and warped Gaussian processes + Tukey processes.
Part 3. Estimation: Topics will include
- Monte carlo methods: Gibbs, Metropolis Hastings, Hamiltonian Monte Carlo, Reimann-Manifold HMC, Geodesic HMC.
- Sequential Monte Carlo
- Particle Markov chain, Monte Carlo Applications will come from insurance and risk: mortality modelling, non-life claims reserving analysis, temperature and weather modelling, commodity futures, Green finance and Green bonds, Interest rate models.
10:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Dr. David González (Université de Fribourg)
Manifolds with positive sectional curvature abstract
Abstract:
In this talk we will review some classical results concerning manifolds with positive sectional curvature, and we will discuss some related openproblems. In particular we will study potential obstructions to the existence of Riemannian submersions between positively curved manifolds, based on joint work with Luis Guijarro and Marco Radeschi.
10:20 • Université de Fribourg, Chemin du Musée 23, Lonza (Math II) seminar room
Jakob Oesinghaus (Universität Zürich)
Chow group of expanded pairs and quasi-symmetric functions abstract
Abstract:
We study the stack of expanded pairs originally defined by Jun Li, and in greater generality by Abramovich, Cadman, Fantechi, and Wise, which, roughly speaking, parametrises certain standard ways to expand a pair $(X,D)$ of a scheme $X$ and a Cartier divisor $D$. We use an alternative description of this stack as the moduli stack of aligned log structures to compute its Chow group, which turns out to be the ring of quasi-symmetric functions, a variant of the usual ring of symmetric functions. We discuss applications of this result to Chow groups of some moduli stacks of genus 0 semi-stable curves.
13:30 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Prof. Dr. Sihem Mesnager (University of Paris VIII)
Mathematical contributions on the construction of good codes for secret sharing and storage
15:00 • UZH Irchel, Winterthurerstrasse 190, Zürich, Building Y27, Room H 12
Timo Weidl (Universität Stuttgart)
Semiclassical Spectral Estimates with Remainder Terms abstract
Abstract:
Starting from the classical Berezin - and Li-Yau - bounds on the eigenvalues of the Laplace operator with Dirichlet boundary conditions I give a survey on various improvements of these inequalities by remainder terms. Beside the Melas inequality we deal with modifications thereof for operators with and without magnetic field and bounds with (almost) classical remainders. Finally we extend these results to the Heisenberg sub-Laplacian and the Stark operator in domains.
15:00 • EPF Lausanne, Salle CM 1 113
Cinzia Casagrande (Politecnico Torino)
On the Fano variety of linear spaces contained in two odd-dimensional quadrics abstract
Abstract:
I will describe the geometry of the 2m-dimensional Fano manifold G parametrizing (m-1)-planes in a smooth complete intersection Z of two quadric hypersurfaces in the complex projective space of dimension 2m+2. We will see that there are exactly 2^{2m+2} distinct isomorphisms in codimension one between G and the blow-up of P^{2m} at 2m + 3 general points, parametrized by the 2^{2m+2} distinct m-planes contained in Z. This birational maps allow to determine the cones of nef, movable and effective divisors of G, and the automorphism group of G. This generalizes to arbitrary even dimension the classical description of quartic del Pezzo surfaces (m = 1). This is a joint work with Carolina Araujo (IMPA).
15:15 • EPF Lausanne, Salle MA A1 10
Alain Valette (Université de Neuchâtel)
Diameters in box spaces abstract
Abstract:
If $G$ is a finitely generated residually finite group, and $(N_i)_{i>0}$ is a decreasing sequence of finite index normal subgroups with trivial intersection, we study the diameter of $G/N_i$ as a function of the order $|G/N_i|$. For $0<\\alpha\\geq 1$, we say (after Breuillard and Tointon) that the box space $\\square_(N_i)G$ has property $D_\\alpha$, if the diameter of $G/N_i$ grows at least like $|G/N_i|^\\alpha$. We show that a box space has property $D_1$ if and only if G is virtually cyclic; and that a group G admits some box space with property $D_\\alpha$ (for some $\\alpha>0$) if and only if G virtually maps onto $\\mathbb{Z}$. For the lamplighter group and for a lattice in SOL, we provide explicit examples of box spaces with and without property $D_\\alpha$.
15:45 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Tovohery Hajatiana Randrianarisoa (Universität Zürich)
A Decoding Algorithm for Rank Metric Codes
16:00 • UZH Irchel, Winterthurerstrasse 190, Zürich, Building Y27, Room H 12
Prof. Dr. A. Nouy (Ecole Centrale de Nantes)
Learning high-dimensional functions with tree-based tensor formats. abstract
Abstract:
Tensor methods are among the most prominent tools for the approximation of high-dimensional functions. Such approximation problems naturallyarise in statistical learning, stochastic analysis and uncertainty quantification. In many practical situations, the approximation of high-dimensional functions is made computationally tractable by using rank-structured approximations. In this talk, we give an introduction to tree-based (hierarchical) tensor formats and then present adaptive algorithms for the approximation in these formats using statistical methods.
16:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room E 1.2
Prof. Dr. David-Olivier Jaquet-Chiffelle (Université de Lausanne)
Cryptanalyse du Code de Vigenère
16:15 • Université de Neuchâtel, Institut de Mathématiques, Rue Emile - Argand 11, ALG, F 200