Abstract:
Topological and discrete data such as trees and networks constitute strongly non-Euclidean structures, often beyond the reach of traditional statistical tools. Their analysis is often challenged by computational tractability, as natural constraints such as permutation invariance are computationally hard to implement. Still, such data can be often transposed into more familiar contexts, whether by transformation, asymptotic arguments, or replicated observations. This workshop will explore both modelling and processing of discrete structures, also drawing in areas such as signal processing on graphs and nonparametrics.
08:00 • EPF Lausanne, Room GA 3 21
Peter K. Friz (Technische Universität Berlin)
Rough Analysis and Applications (ONLINE) abstract
Abstract:
We plan to discuss: the algebra of iterated integrals, signatures and expected signatures, rough paths and models; Kolmogorov type criteria; Schauder estimates; p-variation, abstract Riemann-Stieltjes integration (Sewing); rough integration and reconstruction; a paracontrolled view; rough differential equations and flows; rough transport equations; second order rough partial differential equations; rough volatility; a regularity structure view; large deviations and precise asymptotics for rough volatility.A good part of the lecture will follow F-Hairer (2014,2020).
10:15 • ETH Zentrum, Rämistrasse 101, Zürich, Building HG, Room
Wei-kuo Chen (University of Minnesota)
Universality of Approximate Message Passing Algorithms abstract
Abstract:
Approximate Message Passing (AMP) algorithms are non-linear power iteration methods originally arising from the context of compressed sensing. In this talk, I will introduce a Lipschitizian functional iteration, as a generalization of the AMP algorithms, and discuss its universality in disorder. In addition, I will explain how our results imply universality in a number of AMPs popularly adapted in Bayesian inferences and optimizations in spin glasses.
11:00 • Universität Basel, Zoom Seminar
Dr. Oren Becker (Cambridge University)
Stability of approximate group actions abstract
Abstract:
An approximate unitary representation of a group G is a function f from G to U(n) such that f(g_1g_2) is close to f(g_1)f(g_2) for all g_1,g_2.Is every approximate unitary representation just a slight deformation of a unitary representation? The answer depends on G and on the norm on U(n).If G is amenable, the answer is positive for the operator norm on U(n) (Kazhdan \'82). The answer remains positive if we use the normalized Hilbert-Schmidt norm and allow a slight change in the dimension n (Gowers-Hatami \'15, De Chiffre-Ozawa-Thom \'17). For both norms, the answer is negative if G is a nonabelian free group (or a nonelementary word-hyperbolic group).In this talk we shall discuss a similar notion where U(n) is replaced by Sym(n) with the normalized Hamming metric. We study the cases where G is either free, amenable or equal to SL_r(Z), r>=3.When G is finite, a slight variation of our main theorem provides an efficient probabilistic algorithm to determine whether a function f:G->Sym(n) is close to a homomorphism when |G| and n are both large.Based on a joint work with Michael Chapman.
Onlline Seminar
Prof. Dr. Gretchen Mathews (Virginia Tech)
eSeminar: Coding for local recovery abstract
Abstract:
(**This eSeminar will take place on Zoom, using the same meeting details as previous seminars. If you do not have meeting details, please contact karan.khathuria@math.uzh.ch **)
Codes with locality are designed to recover erased symbols by accessing only a few others. Such codes are useful in settings such as distributed storage where servers may routinely go offline for maintenance but users still need to access their information. This situation suggests that it is also desirable to have a variety of ways in which to recover a single symbol, leading to the concept of availability. In this talk, we share recent work on codes from curves which provide locality and availability.
16:00 • Onlline Seminar
Prof. Dr. Lenka Zdeborova (CNRS, CEA in Saclay, France.)
Understanding machine learning via exactly solvable statistical physics models abstract
Abstract:
The affinity between statistical physics and machine learning has a long history, this is reflected even in the machine learning terminology that is in part adopted from physics. I will describe the main lines of this long-lasting friendship in the context of current theoretical challenges and open questions about deep learning. Theoretical physics often proceeds in terms of solvable synthetic models, I will describe the related line of work on solvable models of simple feed-forward neural networks. I will highlight a path forward to capture the subtle interplay between the structure of the data, the architecture of the network, and the learning algorithm.
Prof. Dr. Martin Oberlack (TU Darmstadt, Germany)
High-moment turbulent scaling laws and its roots in statistical symmetries abstract
Abstract:
Using the symmetry-based turbulence theory we derive turbulent scaling laws for arbitrarily high moments of the stream-wise velocity $U_1$. In the region of the log-law, the theory predicts an algebraic law with the exponent $t_2 (n-1)$ for moments $n > 1$. The exponent $s_2$ of the $2^{nd}$ moment determines the exponent of all higher moments. Moments here always refer to the instantaneous quantities and not to the fluctuations. For the core region of a Poiseuille flow we find a deficit law for arbitrary moments $n$ of algebraic type with a scaling exponent $n(s_2-s_1)+2s_1-s_2$. Hence, the moments of order one and two with its scaling exponents $s_1$ and $s_2$ determine all higher exponents. To validate the new theoretical results we have conducted a Poiseuille flow DNS at $Re_\\tau=10^4$. All of the latter theoretical findings could be verified with high accuracy using DNS data.
Yves Le Jan (Université Paris-Sud)
CANCELLED: Title T.B.A.
17:15 • UZH Irchel, Winterthurerstrasse 190, Zürich, Building Y27, Room H 12