Seminari del Dipartimento di Matematica

nel ciclo di seminari: DYNAMICS AND LANDSCAPES IN HIGH DIMENSION: FROM CONSTRAINT SATISFACTION PROBLEMS TO OVERPARAMETRIZED AND RECURRENT NEURAL NETWORKS

ore 11:00

presso Aula Enriques

analisi numerica

fisica matematica

interdisciplinare

probabilità

This series of lectures aims to review the research landscape on high-dimensional dynamical systems, a topic that arises in multiple disciplines. In physics, understanding the many-body dynamics of complex systems is crucial for characterizing their equilibration properties—or the lack thereof. Beyond physics, the study of dynamics is equally vital in optimization and computer science, particularly when dealing with high-dimensional, non-convex problems—such as those encountered in the training dynamics of artificial neural networks. In neuroscience, these systems also model recurrent neural networks, where fixed points, chaos, and control mechanisms shape their behavior. The course has a twofold purpose: first, to survey current knowledge of high-dimensional dynamical systems across various contexts; and second, to explore dynamical mean field theory, the primary toolkit for tackling these challenges. Outline: 1. Stochastic differential equations, Langevin dynamics, Fokker-Planck equation and stationary measures. 2. Equilibrium dynamics and Fluctuation-Dissipation relations. 3. Dynamical mean field theory: Dynamical cavity method 4. Dynamics of mean field spin glasses. High temperature phase and relaxation to equilibrium; Low temperature phase, aging. Landscape interpretation. 5. High-dimensional inference: Langevin/Gradient Descent algorithms and their suboptimality with respect to Approximate Message Passing. 6. Training dynamics of artificial neural networks. Neural tangent kernel and feature learning theory. Separation of timescales between feature learning and overfitting. 7. Recurrent neural networks. Transition to and chaos in high dimension. Maximal Lypaunov exponent. 8. Learning algorithms and optimal control of high-d chaotic dynamics. 9. From Neural ODEs to LLMs and generative models 10. Open problems and perspectives

Can small intersections determine global components?

ore 11:15

presso Aula Vitali

algebra e geometria

In any fixed dimension, consider a finite collection of sets. The ordinary nerve records which intersections are nonempty, but forgets their connected components. Can we recover this missing information by looking at only a few sets at a time? I will discuss this question through a Helly-type theorem for sets with two convex components: under suitable local assumptions, the full intersection turns out to have exactly two convex components as well.

Scalable Multi-View Clustering: Graph Learning and Anchor Selection

nell'ambito della serie: SCUBE

ore 16:00

presso Seminario II

seminario on line • collegamento al meeting

https://sites.google.com/view/na-seminars-unibo/home?authuser=1

analisi numerica

Multi-view clustering aims to integrate complementary information from multiple data sources to identify a shared clustering structure. This talk presents my recent research on scalable multi-view clustering, with a focus on consensus bipartite graph learning and anchor selection. I will introduce two clustering formulations and discuss the nonlinear matrix equations arising in their optimization. The current methods use linearized or surrogate updates to reduce computational cost, while directly solving the original nonlinear equations remains an open question. I will also discuss column subset selection as a numerical linear algebra problem motivated by the need to select representative anchors. The talk will highlight the connections between clustering models, matrix equation solvers, and matrix approximation, together with unresolved questions arising from this work.

Malliavin Calculus - Lecture 5

nel ciclo di seminari: TOPICS IN STOCHASTIC ANALYSIS: MALLIAVIN CALCULUS

ore 11:00

presso Seminario II

seminario on line • collegamento al meeting

probabilità

Lecture 5

Lecture 7: Dynamics and landscapes in high dimension

nel ciclo di seminari: DYNAMICS AND LANDSCAPES IN HIGH DIMENSION: FROM CONSTRAINT SATISFACTION PROBLEMS TO OVERPARAMETRIZED AND RECURRENT NEURAL NETWORKS

ore 11:00

presso Seminario II

Link Team

analisi numerica

fisica matematica

interdisciplinare

probabilità

This series of lectures aims to review the research landscape on high-dimensional dynamical systems, a topic that arises in multiple disciplines. In physics, understanding the many-body dynamics of complex systems is crucial for characterizing their equilibration properties—or the lack thereof. Beyond physics, the study of dynamics is equally vital in optimization and computer science, particularly when dealing with high-dimensional, non-convex problems—such as those encountered in the training dynamics of artificial neural networks. In neuroscience, these systems also model recurrent neural networks, where fixed points, chaos, and control mechanisms shape their behavior. The course has a twofold purpose: first, to survey current knowledge of high-dimensional dynamical systems across various contexts; and second, to explore dynamical mean field theory, the primary toolkit for tackling these challenges. Outline: 1. Stochastic differential equations, Langevin dynamics, Fokker-Planck equation and stationary measures. 2. Equilibrium dynamics and Fluctuation-Dissipation relations. 3. Dynamical mean field theory: Dynamical cavity method 4. Dynamics of mean field spin glasses. High temperature phase and relaxation to equilibrium; Low temperature phase, aging. Landscape interpretation. 5. High-dimensional inference: Langevin/Gradient Descent algorithms and their suboptimality with respect to Approximate Message Passing. 6. Training dynamics of artificial neural networks. Neural tangent kernel and feature learning theory. Separation of timescales between feature learning and overfitting. 7. Recurrent neural networks. Transition to and chaos in high dimension. Maximal Lypaunov exponent. 8. Learning algorithms and optimal control of high-d chaotic dynamics. 9. From Neural ODEs to LLMs and generative models 10. Open problems and perspectives

Lecture 8: Dynamics and landscapes in high dimension

nel ciclo di seminari: DYNAMICS AND LANDSCAPES IN HIGH DIMENSION: FROM CONSTRAINT SATISFACTION PROBLEMS TO OVERPARAMETRIZED AND RECURRENT NEURAL NETWORKS

ore 09:00

presso Aula Seminario VIII piano

Link Team

analisi numerica

fisica matematica

interdisciplinare

probabilità

This series of lectures aims to review the research landscape on high-dimensional dynamical systems, a topic that arises in multiple disciplines. In physics, understanding the many-body dynamics of complex systems is crucial for characterizing their equilibration properties—or the lack thereof. Beyond physics, the study of dynamics is equally vital in optimization and computer science, particularly when dealing with high-dimensional, non-convex problems—such as those encountered in the training dynamics of artificial neural networks. In neuroscience, these systems also model recurrent neural networks, where fixed points, chaos, and control mechanisms shape their behavior. The course has a twofold purpose: first, to survey current knowledge of high-dimensional dynamical systems across various contexts; and second, to explore dynamical mean field theory, the primary toolkit for tackling these challenges. Outline: 1. Stochastic differential equations, Langevin dynamics, Fokker-Planck equation and stationary measures. 2. Equilibrium dynamics and Fluctuation-Dissipation relations. 3. Dynamical mean field theory: Dynamical cavity method 4. Dynamics of mean field spin glasses. High temperature phase and relaxation to equilibrium; Low temperature phase, aging. Landscape interpretation. 5. High-dimensional inference: Langevin/Gradient Descent algorithms and their suboptimality with respect to Approximate Message Passing. 6. Training dynamics of artificial neural networks. Neural tangent kernel and feature learning theory. Separation of timescales between feature learning and overfitting. 7. Recurrent neural networks. Transition to and chaos in high dimension. Maximal Lypaunov exponent. 8. Learning algorithms and optimal control of high-d chaotic dynamics. 9. From Neural ODEs to LLMs and generative models 10. Open problems and perspectives

Variational Geometric Evolution Model for Image Segmentation

ore 11:00

presso Seminario I

analisi numerica

Ottobre

28

Mercoledì

Gerardo Barrera Vargas

Cutoff phenomenon: an overview for SDEs

nell'ambito della serie: STOCHASTICS AND APPLICATIONS

ore 11:00

presso Seminario I

fisica matematica

probabilità

sistemi dinamici

The cutoff phenomenon describes an abrupt transition in the convergence of a stochastic process toward its equilibrium, occurring within a time window that is negligible compared to the mixing time. In this talk, we provide an overview of the cutoff phenomenon for stochastic differential equations (SDEs), with particular emphasis on small-noise diffusions. We discuss the main ideas behind cutoff, the role of deterministic dynamics and stochastic fluctuations, and the emergence of cutoff times and profiles.

Symmetrized Fractional Variation for signal denoising

ore 16:00

presso Aula Vitali

analisi matematica

analisi numerica

We define, study and implement the model SFV (Symmetrized Fractional Variation): a variational approach to signal analysis exploiting the Riemann-Liouville fractional derivatives of every positive real order higher than zero. The model exploites an L_1 fitting data term together with both right and left fractional derivatives as regularizing terms: this approach aims to achieve an orientation-independent protocol. We express the model as an energy minimization and introduce a functional framework where existence of minimizers is proved: namely the space BVs∗. We show that the embedding in BVs∗ of the Sobolev space of the same fractional order is strict, and we exhibit some nontrivial borderline examples of admissible or non admissible functions in the space BVs∗. To provide evidence of effectiveness for the proposed model, a discretisation based on a second-order consistent Gr¨unwald-Letnikov scheme is introduced. A multi-parameter whiteness criterion is proposed which provides an unsupervised, automatic and simultaneous selection of the two free parameters in the model, namely the fractional order of differentiation and the regularization parameter. Numerical experiments on 1-d and 2-d signals are performed which show how the proposed model holds the potential to achieve good quality results for denoising signals corrupted by additive Laplace noise. This is a joint research with Alessandro Lanza(1), Antonio Leaci(2) and Serena Morigi(1). (1) Dipartimento di Matematica, Universit´a di Bologna (2) Universit´a del Salento, Dipartimento di Matematica e Fisica “Ennio De Giorgi”, Lecce

La regina delle matematiche medievali a Bisanzio

ore 16:30

presso Accademia delle Scienze dell'Istituto di Bologna

Pagina istituzionale dello speaker

fisica matematica

storia della matematica

La regina delle matematiche medievali è la logistica, cioè la disciplina che tratta le tecniche di calcolo elementare e le loro applicazioni. Con l’aiuto di un ampio corredo iconografico, la presentazione descrive le grandi linee delle tre diramazioni principali della logistica a Bisanzio: i Computi pasquali; le esposizioni delle tecniche di calcolo elementare nei sistemi decimale (posizionale e non) e sessagesimale; le raccolte di problemi aritmetici note come Rechenbücher (“libri di conti”), il cui esempio più noto in Occidente è il Liber Abbaci di Fibonacci.

Un protocollo di trattamento automatico del lessico di testi scientifici antichi e delle loro traduzioni

ore 11:30

presso Seminario II

Pagina istituzionale dello speaker

fisica matematica

storia della matematica

I testi scientifici scritti in linguaggio naturale impiegano un lessico limitato e una sintassi strettamente regimentata, talvolta al punto di configurare un proto-formalismo. Prendendo come esempio gli Elementi di Euclide e tre sue traduzioni latine, la presentazione descrive un protocollo di trattamento automatico del lessico e della sintassi di questi testi. Il protocollo applica in maniera sistematica nozioni elementari di teoria dell’informazione e di termodinamica statistica.

Aspherical 4-manifolds with positive Euler characteristic

nell'ambito della serie: SEMINARIO DI ALGEBRA E GEOMETRIA

ore 11:30

presso Aula Vitali

algebra e geometria

I will present an explicit construction of closed oriented aspherical smooth 4-manifolds realizing every positive Euler characteristic. This family of manifolds saturates the Hopf-Thurston conjecture on the Euler characteristic of aspherical manifolds, as well as the Gromov-Lück inequality on their geography.

Insiemi algebrici slice regolari e irriducibilità

ore 11:30

presso Aula Vitali

algebra e geometria

A quaternionic version of the classical Hilbert Nullstellensatz has recently been established in its strong form within the framework of slice regular polynomials. This result naturally led to the introduction of the notion of a quaternionic slice algebraic set, endowing (\mathbb{H}^n) with a Zariski-type topology. In this seminar, after a brief overview of the subject, we will discuss a characterization of the irreducibility of slice algebraic sets in terms of algebraic properties of the associated ideals of slice regular polynomials. The results presented are based on joint works with Giulia Sarfatti and Fabio Vlacci.

L’EREDITÀ DI VITO VOLTERRA

ore 20:30

presso Cinema Perla - via san Donato 38

storia della matematica

The multiple cover formula conjecture for Gromov-Witten invariants of abelian surfaces

nell'ambito della serie: SEMINARIO DI ALGEBRA E GEOMETRIA

ore 11:30

presso Aula Vitali

algebra e geometria

Abelian surfaces are complex tori whose enumerative invariants satisfy remarkable regularity properties. The computation of their (reduced) Gromov-Witten invariants for the so called primitive classes is fairly well understood and many complete computations are available. A few years ago, G. Oberdieck conjectured a multiple cover formula expressing in a very simple way the invariants for the non-primitive classes in terms of the primitive one. In this talk, I'll sketch the proof of multiple cover formula conjecture obtained in a joint work with T. Blomme . This effectively solves completely the GW theory of abelian surfaces. The argument relies on a reduced degeneration formula and on the computation of correlated Gromov-Witten invariants for trivial bundles on elliptic curves.

Simon's Conjecture Disproved

ore 11:30

presso Aula Vitali

Seminario di Algebra e Geometria

algebra e geometria

In 1994, Simon conjectured that every pure shellable simplicial complex is shelling-completable, i.e. the independence complexes of uniform matroids are extendably shellable. In this seminar, after reviewing the fundamental concepts and providing an overview of the state of the art, I present a technique producing an infinite number of shellable complexes that are not shelling-completable—one for each dimension greater than or equal to 3—thereby disproving Simon's conjecture. This is joint work with Paolo Sentinelli.

Brill-Noether loci of pencils with prescribed ramification

nell'ambito della serie: SEMINARIO DI ALGEBRA E GEOMETRIA

ore 11:30

presso Aula Vitali

algebra e geometria

Classical Brill-Noether theory studies projective algebraic curves through maps to projective space, using the language of linear series. A natural refinement of this framework considers linear series with prescribed ramification. In this talk, I will address the problem of the dimension of Brill–Noether varieties of pencils (i.e., linear series of rank one) with prescribed ramifications. I will compare this refined theory to the classical Brill–Noether problem, and present recent joint work with Andreas Leopold Knutsen and Ciro Ciliberto, which yields new results and improves upon the previous state of the art.

Mean Field Games: From Foundations to Singular Controls and Optimal Stopping

nel ciclo di seminari: SEMINARI DI PROBABILITÀ

ore 14:30

presso Centro Congressi "Alla Sosta dell'Imperatore" Strada dei Margeni, 18 38025 Dimaro Folgarida (TN)

Webpage della 3nd Dolomites Winter School on Mean-field Control, Mean-field Games and Applications

probabilità

Sito web Winter School

This short course provides an introduction to mean field games, with particular emphasis on their probabilistic formulation and on models involving singular control and optimal stopping. We will start from the basic idea of mean field interaction: a representative agent optimizes against the distribution of a large population, while equilibrium requires consistency between the agent’s optimal behavior and the induced population flow. After discussing this fixed point structure and some analytical tools, we will focus on classes of models motivated by economics and finance. Special attention will be devoted to mean field games in which agents make irreversible, monotone, or stopping decisions, both in stationary and non-stationary settings. In these models, a central theme is the interplay between free-boundary methods, reflected or absorbed diffusions, and fixed point arguments in the characterization of equilibria. We will also discuss how structural properties of the interaction, such as monotonicity or submodularity, can be exploited in suitable models to obtain comparison results and to gain insights into uniqueness or multiplicity of equilibria. The course will conclude by highlighting connections between mean field games and mean field control problems, as well as recent perspectives related to learning-based approaches.

Mean-Field Control: Theory, Methods, from Foundations to Learning

ore 15:30

presso Centro Congressi "Alla Sosta dell'Imperatore" Strada dei Margeni, 18 38025 Dimaro Folgarida (TN)

Webpage della 3nd Dolomites Winter School on Mean-field Control, Mean-field Games and Applications

probabilità

These lectures cover the optimal control of McKean-Vlasov equations, a topic that has attracted growing interest in connection with mean-field game theory, large population stochastic control, and more recently machine learning. We start with mean-field Markov decision processes in discrete time, where the key ideas — propagation of chaos, lifting to the Wasserstein space, dynamic programming — can be introduced transparently. We then develop the continuous-time theory of controlled McKean-Vlasov SDEs, building the necessary tools: differentiation with respect to probability measures, Itô's formula along flows of marginal laws, the Master Bellman equation, and the stochastic maximum principle with its associated FBSDE system. Linear-quadratic problems serve as a running illustration throughout. The last part of the course is devoted to numerical methods and applications. We present neural network-based algorithms and reinforcement learning approaches for solving mean-field control problems on the Wasserstein space, and discuss recent extensions to non-exchangeable mean-field control for heterogeneous interacting systems. We conclude with an application to generative modeling, where learning stochastic dynamics from distributional observations is formulated as a McKean-Vlasov control problem, with optimality conditions given by a tractable FBSDE system.

Prossimi seminari del Dipartimento di Matematica