Past

The (Local) Lifting Property ((L)LP) is introduced by Kirchberg and deals with lifting completely positive maps. We will discuss various examples, characterizations, and closure properties of the (L)LP and, if time permits, connections with some other lifting properties of C*-algebras.  Joint work with Dominic Enders.

Large Language Models (LLMs) have demonstrated impressive achievements. However, recent research has shown that their deeper layers often contribute minimally, with effectiveness diminishing as layer depth increases. This pattern presents significant opportunities for model compression. 

In the first part of this seminar, we will explore how this phenomenon can be harnessed to improve the efficiency of LLM compression and parameter-efficient fine-tuning. Despite these opportunities, the underutilization of deeper layers leads to inefficiencies, wasting resources that could be better used to enhance model performance. 

The second part of the talk will address the root cause of this ineffectiveness in deeper layers and propose a solution. We identify the issue as stemming from the prevalent use of Pre-Layer Normalization (Pre-LN) and introduce Mix-Layer Normalization (Mix-LN) with combined Pre-LN and Post-LN as a new approach to mitigate this training deficiency.

The Brunn-Minkowski inequality gives a lower bound on the volume of the set of midpoints of line segments joining two sets. On a Riemannian manifold, line segments are replaced by geodesic segments, and the Brunn-Minkowski inequality characterizes manifolds with nonnegative Ricci curvature. I will present a generalization of the Riemannian Brunn-Minkowski inequality where geodesics are replaced by magnetic geodesics, which are minimizers of a functional given by length minus the integral of a fixed one-form on the manifold. The Brunn-Minkowski inequality is then equivalent to nonnegativity of a suitably defined magnetic Ricci curvature. More generally, I will present a magnetic version of the Borell-Brascamp-Lieb inequality of Cordero-Erausquin, McCann and Schmuckenschläger. The proof uses the needle decomposition technique.

Multicomponent flows, i.e. mixtures, can be modeled effectively using the Onsager-Stefan-Maxwell (OSM) equations. The OSM equations can account for a wide variety of phenomena such as diffusive, convective, non-ideal mixing, thermal, pressure and electrochemical effects for steady and transient multicomponent flows. I will first introduce the general OSM framework and a finite element discretisation for multicomponent diffusion of ideal gasses. Then I will show two ways of preconditioning the multicomponent diffusion problem for various boundary conditions. Time permitting, I will also discuss how this can be extended to the non-ideal, thermal, and nonisobaric settings.

How should we interpret past mathematicians who may use the same vocabulary as us but with different meanings, or whose philosophical outlooks differ from ours? Errors aside, it is often assumed that past mathematicians largely made true claims—but what exactly justifies that assumption?

In this talk, we will explore these questions through general philosophical considerations and three case studies: 19th-century analysis, 18th-century geometry, and 19th-century matricial algebra.  In each case, we encounter a significant challenge to supposing that the mathematicians in question made true claims. We will show how these challenges can be addressed and overcome.

In 1911, Otto Toeplitz posed the intriguing "Square Peg Problem," asking whether every Jordan curve admits an inscribed square. Despite over a century of study, the problem remains unsolved in its full generality. However, significant progress has been made over the years. In this talk, we explore recent advancements by Andrew Lobb and Joshua Greene, who approach the problem through the lens of Lagrangian Floer homology. Specifically, we outline a proof of their result: every smooth Jordan curve inscribes every rectangle up to similarity.

I will present TeXmacs (www.texmacs.org), a document preparation system with structured editing and high quality typography, designed to create technical documents. 

For more details see https://mgubi.github.io/docs/programming.html

Motivated by a recent experiment on a superconducting quantum
information processor, I will discuss the Floquet quantum Ising model in
the presence of integrability- and symmetry-breaking random fields. The
talk will focus on the relation between boundary spin correlations,
spectral pairings, and effects of the random fields. If time permits, I
will also touch upon self-similarity in the dynamic phase diagram of
Fibonacci-driven quantum Ising models.
 

I will give a light introduction to the concept of a quantum expander, which is an analogue of an expander graph that arises in quantum information theory.  Most examples of quantum expanders that appear in the quantum information literature are obtained by random matrix techniques.  I will explain another, more algebraic approach to constructing quantum expanders, which is based on using actions and representations of discrete quantum groups with Kazhdan's property (T).  This is joint work with Eric Culf (U Waterloo) and Matthijs Vernooij (TU Delft).   

In a recent breakthrough Campos, Griffiths, Morris and Sahasrabudhe obtained the first exponential improvement of the upper bound on the classical Ramsey numbers since 1935. I will outline a reinterpretation of their proof, replacing the underlying book algorithm with a simple inductive statement. In particular, I will present a complete proof of an improved upper bound on the off-diagonal Ramsey numbers and describe the main steps involved in improving their upper bound for the diagonal Ramsey numbers to $R(k,k)\le(3.8)^k$ for sufficiently large $k$.

Based on joint work with Parth Gupta, Ndiame Ndiaye, and Louis Wei.

I will present a joint work with Thang Nguyen where we show that there exists a quasi-isometric embedding of the product of n copies of the hyperbolic plane into any symmetric space of non-compact type of rank n, and there exists a bi-Lipschitz embedding of the product of n copies of the 3-regular tree into any thick Euclidean building of rank n. This extends a previous result of Fisher--Whyte. The proof is purely geometrical, and the result also applies to the non Bruhat--Tits buildings. I will start by describing the objects and the embeddings, and then give a detailed sketch of the proof in rank 2.

Networks provide a powerful language to model and analyse interconnected systems. Their building blocks are  edges, which can  then be combined to form walks and paths, and thus define indirect relations between distant nodes and model flows across the system. In a traditional setting, network models are first-order, in the sense that flow across nodes is made of independent sequences of transitions. However, real-world systems often exhibit higher-order dependencies, requiring more sophisticated models. Here, we propose a variable-order network model that captures memory effects by interpolating between first- and second-order representations. Our method identifies latent modes that explain second-order behaviors, avoiding overfitting through a Bayesian prior. We introduce an interpretable measure to balance model size and description quality, allowing for efficient, scalable processing of large sequence data. We demonstrate that our model captures key memory effects with minimal state nodes, providing new insights beyond traditional first-order models and avoiding the computational costs of existing higher-order models.

Let $(V(u))$ be a branching random walk and $(\eta(u))$ be i.i.d marks on the vertices. To each path $\xi$ of the tree, we associate the discounted sum $D(\xi)$ which is the sum of the $\exp(V(u))\eta_u$ along the path. We study conditions under which $\sup_\xi D(\xi)$ is finite, answering an open question of Aldous and Bandyopadhyay. We will see that this problem is related to the study of the local time process of the branching random walk along a path. It is a joint work with Yueyun Hu and Zhan Shi.

Suppose a finite group satisfies the following property: If you take two random elements, then with probability bigger than 5/8 they commute. Then this group is commutative. 

Starting from this well-known result, it is natural to ask: Do similar results hold for other laws (p-groups, nilpotent groups...)? Are there analogous results for infinite groups? Are there phenomena specific to the infinite setup? 

We will survey known and new results in this area. New results are joint with Gideon Amir, Maria Gerasimova and Gady Kozma.

In this talk I will explain how the size of the Einstein-Rosen (ER) bridge dual to the Double Scaled SYK (DSSYK) model saturates at late times because of finiteness of the underlying quantum Hilbert space.  I will extend recent work implying that the ER bridge size equals the spread complexity of the dual DSSYK theory with an appropriate initial state.  This work shows that the auxiliary "chord basis'' used to solve the DSSYK theory is the physical Krylov basis of the spreading quantum state.  The ER bridge saturation follows from the vanishing of the Lanczos spectrum, derived by methods from Random Matrix Theory (RMT).

The Dyson Brownian motion (DMB) is a system of interacting Brownian motions with logarithmic interaction potential, which was introduced by Freeman Dyson '62 in relation to the random matrix theory. In this talk, we discuss the case where the number of particles is infinite and show that the DBM induces a diffusion structure on the configuration space having the Bakry-Émery lower Ricci curvature bound. As an application, we show that the DBM can be realised as the unique Benamou-Brenier-type gradient flow of the Boltzmann-Shannon entropy associated with the sine_beta point process. 

Topological Quantum Field Theories (TQFTs) have been studied as mathematical toy models for quantum field theories in physics and are described by a functor out of some bordism category. In dimension 2, TQFTs are fully classified by Frobenius algebras. Homotopy Quantum Field Theories (HQFTs), introduced by Turaev, consider additional homotopy data to some target space X on the bordism categories. For homotopy 1-types Turaev also gives a classification via crossed G-Frobenius algebras, where G denotes the fundamental group of X.
In this talk we will introduce a multi-object generalization of Frobenius algebras called Frobenius categories and give a version of this classification theorem involving the fundamental groupoid. Further, we will give a classification theorem for HQFTs with target homotopy 2-types by considering crossed modules (joint work with Alexis Virelizier).
 

The construction of the measure of the $\Phi^4_3$ model in the 1970s has been one of the major achievements of constructive quantum field theory. In the 1980s Parisi and Wu suggested an alternative way of constructing quantum field theory measures by viewing them as invariant measures of certain stochastic PDEs. However, the highly singular nature of these equations prevented their application in rigorous constructions until the breakthroughs in the area of singular stochastic PDEs in the past decade. After explaining the basic idea behind stochastic quantization proposed by Parisi and Wu I will show how to apply this technique to construct the measure of a certain quantum field theory model generalizing the $\Phi^4_3$ model called the fractional $\Phi^4$ model. The measure of this model is obtained as a perturbation of the Gaussian measure with covariance given by the inverse of a fractional Laplacian. Since the Gaussian measure is supported in the space of Schwartz distributions and the quartic interaction potential of the model involves pointwise products, to construct the measure it is necessary to solve the so-called renormalization problem. Based on joint work with M. Gubinelli and P. Rinaldi.

A well-known problem in algebraic geometry is to construct smooth projective Calabi--Yau varieties $Y$. In the smoothing approach, we construct first a degenerate (reducible) Calabi--Yau scheme $V$ by gluing pieces. Then we aim to find a family $f\colon X \to C$ with special fiber $X_0 = f^{-1}(0) \cong V$ and smooth general fiber $X_t = f^{-1}(t)$. In this talk, we see how infinitesimal logarithmic deformation theory solves the second step of this approach: the construction of a family out of a degenerate fiber $V$. This is achieved via the logarithmic Bogomolov--Tian--Todorov theorem as well as its variant for pairs of a log Calabi--Yau space $f_0\colon X_0 \to S_0$ and a line bundle $\mathcal{L}_0$ on $X_0$.
 

We consider minimizing the sum of a large number of smooth and possibly non-convex functions, which is the typical problem encountered in the training of deep neural networks on large-size datasets. 

Improving the Controlled Minibatch Algorithm (CMA) scheme proposed by Liuzzi et al. (2022), we propose CMALight, an ease-controlled incremental gradient (IG)-like method. The control of the IG iteration is performed by means of a costless watchdog rule and a derivative-free line search that activates only sporadically to guarantee convergence. The schemes also allow controlling the updating of the learning rate used in the main IG iteration, avoiding the use of preset rules, thus overcoming another tricky aspect in implementing online methods.

Convergence to a stationary point holds under the lonely assumption of Lipschitz continuity of the gradients of the component functions without knowing the Lipschitz constant or imposing any growth assumptions on the norm of the gradients.

We present two sets of computational tests. First, we compare CMALight against state-of-the-art mini-batch algorithms for training standard deep networks on large-size datasets, and deep convolutional neural networks and residual networks on standard image classification tasks on CIFAR10 and CIFAR100. 

Results shows that CMALight easily scales up to problem with order of millions  variables and has an advantage over its state-of-the-art competitors.

Finally, we present computational results on generative tasks, testing CMALight scaling capabilities on image generation with diffusion models (U-Net architecture). CMA Light achieves better test performances and is more efficient than standard SGD with weight decay, thus reducing the computational burden (and the carbon footprint of the training process).

Laura Palagi, @email

Department of Computer, Control and Management Engineering,

Sapienza University of Rome, Italy

Joint work with 

Corrado Coppola, @email

Giampaolo Liuzzi, @email

Lorenzo Ciarpaglini, @email

Since Hawking first discovered that black holes radiate, the evaporation of black holes has been a subject of great interest. In this talk, based on [2411.03447], we review some recent results about the evaporation of charged (Reissner-Nordström) black holes. We consider in particular the difference between neutral and charged particle emission, and explain how this drives the black hole near extremality, as well as how evaporation is then changed in that limit.

In the fast-paced world of trading, where exabytes of data and advanced mathematical models offer powerful insights, how do you harness these to anticipate market shifts and evolving prices? Numbers alone only tell part of the story. Beneath the surface lies the unpredictable force of human behaviour – the delicate balance of buyers and sellers shaping the market’s course. 

In this talk, we’ll uncover how these forces intertwine, revealing insights that not only harness data but challenge conventional thinking about the future of trading.

Speaker: Chris Horrobin (Head of European and US people development for Optiver)

Speaker bio

Chris Horrobin is Head of European and US people development for Optiver. Chris started his career trading US and German bond options, adding currency and European index options into the mix before moving to focus primarily on index options. Chris spent his first three years in Amsterdam before transferring to Sydney. 

During these years, Chris traded some of the biggest events of his career including Brexit and Trump (first time around) and before moving back to Europe led the positional team in his last year. Chris then moved out of trading and into our training team running our trading education space for four years, owning both the design and execution of our renowned internship and grad programs. 

The Education Team at Optiver is central to the Optiver culture and focus on growth – both of employees and the company. Chris has now extended his remit to cover the professional development of hires throughout the business.