Department of Mathematics

Search

DACO seminar

×

Modal title

Modal content

Please subscribe here if you would you like to be notified about these events via e-mail. Moreover you can also subscribe to the iCal/ics Calender.

Autumn Semester 2026

Date / Time Speaker Title Location
* 8 October 2026
13:45-15:00 		Yatin Dandi
EPFL
Details

DACO Seminar

Title Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learning
Speaker, Affiliation Yatin Dandi, EPFL
Date, Time 8 October 2026, 13:45-15:00
Location HG D 3.3
Abstract Understanding how deep neural networks learn useful internal representations from data remains a central open problem in the theory of deep learning. We introduce Neural Low-Degree Filtering (Neural LoFi), a stylized limit of gradient-based training in which hierarchical feature learning becomes an explicit iterative spectral procedure. In this limit, the dynamics at each layer decouple: given the current representation, the next layer selects directions with maximal accessible low-degree correlation to the label. This yields a tractable surrogate mechanism for deep learning, together with a natural kernel-space interpretation. Neural LoFi provides a mathematically explicit framework for studying multi-layer feature learning beyond the lazy regime. It predicts how representations are selected layer by layer, explains how emergence of concepts arises with given sample complexity, and gives a concrete mechanism by which depth progressively constructs new features from old ones through low-degree compositionality. We complement the theory with mechanistic experiments on fully connected and convolutional architectures, showing that Neural LoFi improves over lazy random-feature baselines, recovers meaningful structured filters, and predicts representations aligned with early gradient-descent feature discovery with real datasets. The talk is based on joint work with Matteo Vilucchio, Luca Arnaboldi, Hugo Tabanelli, and Florent Krzakala (https://arxiv.org/abs/2605.13612).
Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learningread_more
HG D 3.3
22 October 2026
13:45-14:45 		Enli Chen
TU Delft
Details

DACO Seminar

Title Strongly convergent random matrix models for q-Gaussian variables
Speaker, Affiliation Enli Chen, TU Delft
Date, Time 22 October 2026, 13:45-14:45
Location HG G 19.1
Abstract q-Gaussian variables form a family of noncommutative random variables interpolating between free and classical Gaussian behavior. A natural question is whether they admit finite-dimensional random matrix models that converge not only in moments, but also strongly, meaning that the operator norms of all noncommutative polynomials converge. I will explain our construction of strongly convergent random matrix models for finite families of q-Gaussian variables. The construction proceeds in two stages. First, normalized sums of graph-product (Γ-independent) semicircular variables converge strongly to finite q-Gaussian families. Second, the corresponding graph-product semicircular systems are realized as strong limits of finite-dimensional tensor-GUE random matrices. Both steps admit matrix-coefficient versions, and the resulting random-matrix convergence is uniform for growing coefficient dimensions that may even exceed the dimension of the random matrices themselves. Our current arXiv version treats the range |q| < √2 − 1, and we have recently extended the result to the full range |q|<1. I will conclude with a brief discussion of recent connections between our construction and q-Gaussian limits arising from Kikuchi matrices.
Strongly convergent random matrix models for q-Gaussian variablesread_more
HG G 19.1
* 23 October 2026
14:15-15:15 		Prof. Dr. Jorge Garza-Vargas
MIT, USA
Details

DACO Seminar

Title Strong convergence of random matrices
Speaker, Affiliation Prof. Dr. Jorge Garza-Vargas, MIT, USA
Date, Time 23 October 2026, 14:15-15:15
Location HG D 3.3
Abstract I will review the method of Chen, Tropp, van Handel, and myself for proving strong convergence of random matrices. This, roughly speaking, can be thought of as a method for obtaining a sharp control on the norm of random matrices whose randomness possess certain symmetries. I will then state some results and open problems.
Strong convergence of random matricesread_more
HG D 3.3
5 November 2026
13:15-15:15 		Dr. Tom Szwagier
IRIT, Toulouse, France
Details

DACO Seminar

Title Random Matrix Theory for Symmetric Tensor Unfolding
Speaker, Affiliation Dr. Tom Szwagier, IRIT, Toulouse, France
Date, Time 5 November 2026, 13:15-15:15
Location HG G 19.1
Abstract How can we recover a hidden low-rank signal from a noisy, high-dimensional tensor? A popular approach, known as unfolding, transforms the tensor into a matrix so that standard spectral methods can be applied. However, when the tensor is symmetric, its entries are dependent, making the theoretical analysis substantially more challenging. In this talk, I will present a random matrix analysis of symmetric tensor unfolding that directly accounts for these dependencies. We establish a sharp phase transition for signal recovery, identifying an exact signal-to-noise ratio threshold above which the signal is partially recovered, the corresponding alignment and limiting spectral distribution. Our results confirm a conjecture of Montanari and Richard (2014). They also show that, after appropriate transformation, unfolded symmetric tensors exhibit the same spectral behavior as classical random matrix models, despite their intricate dependence structure. Based on a joint work with Henrique Goulart and Xiaoyi Mai.
Random Matrix Theory for Symmetric Tensor Unfoldingread_more
HG G 19.1
19 November 2026
13:15-15:15 		Prof. Dr. Coja-Oghlan Amin
Universität Bielefeld
Details

DACO Seminar

Title TBD
Speaker, Affiliation Prof. Dr. Coja-Oghlan Amin, Universität Bielefeld
Date, Time 19 November 2026, 13:15-15:15
Location HG G 19.1
TBD
HG G 19.1
26 November 2026
13:45-14:45 		Gaia Carenini
University of Cambridge
Details

DACO Seminar

Title The scaling window of random k-SAT
Speaker, Affiliation Gaia Carenini, University of Cambridge
Date, Time 26 November 2026, 13:45-14:45
Location HG G 19.1
The scaling window of random k-SAT
HG G 19.1

Notes: the highlighted event marks the next occurring event and events marked with an asterisk (*) indicate that the time and/or location are different from the usual time and/or location.

Organisers:

Archive: AS 26 SS 26 AS 25 SS 25 AS 24 SS 24 AS 23 SS 23 AS 22 SS 22 AS 21 SS 21 AS 20 SS 20

JavaScript has been disabled in your browser