Jacob Bedrossian (University of Maryland)
New and classical perspectives on hydrodynamic stability abstract
Abstract:
Exponential sensitivity to initial conditions is a well-known hallmark of chaotic systems referred to as the \'butterfly effect\' in popular media. It is ubiquitous in many physical systems to varying degrees, however we have few tools for providing a mathematically rigorous proof of this sensitivity for realistic systems. These lectures will introduce students to the study of exponential sensitivity (quantified by Lyapunov exponents, which measure the asymptotic, exponential growth rate of the Jacobian with respect to the initial condition) in stochastic differential equations (SDEs). The course will introduce the basic theory of SDEs, Markov semigroups, random dynamical systems, the multiplicative ergodic theorem for getting well-defined Lyapunov exponents, and then classical and recent advances on obtaining estimates on them. This course will assume an introductory graduate level understanding of analysis, probability, and partial differential equations. Prior exposure to ergodic theorems or stochastic differential equations would be helpful but is not required.
10:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 43
Richard Nickl (University of Cambridge)
Bayesian non-linear inverse problems: statistical and computational guarantees abstract
Abstract:
Common examples for non-linear inverse problems range from parameter identification in PDEs to tomography and data assimilation problems. They naturally involve high- or infinite dimensional parameter spaces and appropriate statistical noise models lead to a class of non-convex inference problems that present substantial challenges in contemporary data science. In influential work, Andrew Stuart (2010) has proposed a unified Bayesian approach to solve such problems. It is computationally feasible via Gaussian process priors and high-dimensional MCMC algorithms and provides important uncertainty quantification methodology (\'error bars\' or confidence regions) based on posterior distributions. Despite evident empirical success, the theoretical understanding of the performance of such methods has been limited until recently. Specifically in non-linear settings Bayesian methods are distinct from optimisation based algorithms and their analysis requires a very different set of mathematical ideas.In these lectures we will summarise recent developments that allow to give rigorous statistical and computational guarantees for the use of these algorithms in high-dimensional and non-convex settings. The general theory will be illustrated in two non-linear model examples arising with elliptic partial differential equations.A standard background in probability and measure, statistics and real analysis will be sufficient to follow this course.
10:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room E 3
Dmitry Kleinbock (Brandeis University)
Dynamics on homogeneous spaces and new applications to number theory abstract
Abstract:
Dynamical systems on homogeneous spaces have been the subject of much attention recently, due to applications to number theory, geometric group theory, hyperbolic geometry etc. Interest in this area rose significantly after the seminal work of Margulis (the proof of the Oppenheim conjecture) and Ratner (Raghunathan’s conjectures) involving unipotent flows on homogeneous spaces. Later developments were considerably stimulated by applications to Diophantine approximation. I will survey basics of ergodic theory, then specialize to homogeneous dynamics, then highlight connections to number theory. The latter allow one to utilize certain properties of homogeneous flows, such as quantitative non-divergence or effective equidistribution of unstable leaves, to prove new results in number theory.
13:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room G 19.2
Stephan Stadler (MPI Bonn)
CAT(0) spaces of higher rank abstract
Abstract:
A Hadamard manifold -- or more generally a CAT(0) space -- is said to have higher rank if every geodesic line lies in a flat plane. If a higher rank Hadamard manifold admits finite volume quotients, then it has to be a symmetric space or split as a direct product. This is the content of Ballmann\'s celebrated Rank Rigidity Theorem, proved in the 80s. It has been conjectured by Ballmann that his theorem generalizes to the synthetic setting of CAT(0) spaces. In the talk I will discuss Ballmann\'s conjecture as well as the related Regular Point Conjecture by Lytchak.
15:45 • Location T.B.A.
Prof. Dr. Michael Feischl (TU Wien)
Towards optimal adaptivity for time-dependent problems abstract
Abstract:
We prove new optimality results for adaptive mesh refinement algorithms for non-symmetric, indefinite, and time-dependent problems by proposing a generalization of quasi-orthogonality which follows directly from the inf-sup stability of the underlying problem. This completely removes a central technical difficulty in modern proofs of optimal convergence of adaptive mesh refinement algorithms. The main technical tools are new stability bounds for the LU-factorization of matrices together with a recently established connection between quasi-orthogonality and matrix factorization.
16:30 • UZH Irchel, Winterthurerstrasse 190, Zürich, Building Y27, Room H 35/36 +
Zoom Meeting Dr. Giuseppe Genovese (University of Zurich)
Reconstruction of a random pattern with restricted Boltzmann machines abstract
Abstract:
A Restricted Boltzmann machine (RBM) is a Gibbs probability distribution of great theoretical and methodological relevance in machine learning. It is defined on a bipartite graph, with one layer (so-called visible) usually made of binary variables encoding the data, and a second ancillary layer (so-called hidden). One says that the RBM retrieves a pattern, if any algorithmic search initialised in proximity of it will not end up so far (or alternatively, the patterns and the local minima of the energy are close enough). I will present some recent results showing that the ability of a RBM to retrieve a random pattern depends on the choice of the distribution of the hidden layer. Indeed efficient retrieval is possible for distributions with a strict sub-Gaussian decay, while strict super-Gaussian tails give poor performance. The case of Gaussian tail (of which the Hopfield model is a special case) is critical and separates these two regimes.
17:15 • UZH Irchel, Winterthurerstrasse 190, Zürich, Building Y27, Room H12
Maryna Viazovska (EPF Lausanne)
Fourier interpolation pairs and their applications abstract
Abstract:
This lecture is about Fourier uniqueness and Fourier interpolation pairs, some explicit constructions of such pairs and their applications. Suppose that we have two subsets X and Y of the Euclidean space. Can we reconstruct a function f from its restriction to the set X and the restriction of its Fourier transform to the set Y? We are interested in the pairs (X,Y) such that the answer to the question above is affirmative. In this talk I will give an overview of recent progress on explicit constructions and existence results for Fourier interpolation pairs and corresponding interpolation formulas. Also I will try to convince you that explicit Fourier interpolation is a useful gadget in solving optimization problem and analysing differential equations.
17:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG , Room F 30