Seminari - Dipartimento Informatica validi dal 18.03.2026 al 18.03.2027. https://www.di.univr.it/?ent=seminario&lang=it&rss=0 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6930 Relatore: Ihsen Yengui (Visiting Professor INdAM 2025-26); Provenienza: University of Sfax, Tunisia; Data inizio: 2026-03-23; Ora inizio: 14.30; Referente interno: Giulio Fellin; Riassunto: Abstract: After a reminder of the key concepts and methods of constructive/computational algebra, this course will aim at generalised Buchberger and Schreyer algorithms for coherent rings. Chapter 1 focuses on the concept of Grouml;bner bases over a field as a powerful tool for problem solving in algebraic geometry. In contrast to typical treatments of Grouml;bner bases, we give in Chapter 2 a theory of Grouml;bner bases in the broader framework of coherent arithmetical rings (possibly with nonzero zero-divisors) and, more generally in Chapter 3, over coherent strongly discrete rings. Notably, for any finitely generated submodule M of a free module over a multivariate polynomial ring with coefficients in a discrete coherent ring, we prove that its module LT(M) of leading terms is countably generated, and provide an algorithm for computing explicitly a generating set. This result is also useful when LT(M) is not finitely generated. We further give a general constructive version of Hilbertrsquo;s syzygy theorem over strongly discrete coherent rings, following Schreyer#39;s method. Format: The course is offered in a hybrid format, both in person and online.All participants will be provided with the necessary links. Assessment method: student project with essay, presentation and discussion Period: from 23rd March 2026 to 24th April 2026 Number of academic hours: 20 Tentative timetable: - Moday, March 23, 14:30-17:30, Aula L, - Wednesday, March 25, 15:30-17:30, Aula M, - Moday, March 30, 14:30-17:30, Aula L, - Wednesday, April 1, 15:30-17:30, Aula M, - Wednesday, April 8, 15:30-17:30, Aula M, - Moday, April 13, 14:30-17:30, Aula L, - Wednesday, April 15, 15:30-17:30, Aula M, - Wednesday, April 22, 15:30- 18:30 , Aula M. Venue: Department of Computer Science University of Verona Strada le Grazie 15 37134 Verona Enroll: LINK . Mon, 23 Mar 2026 14:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6930 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6950 Relatore: Minh Ha Quang; Provenienza: RIKEN Center for Advanced Intelligence Project (AIP), Tokyo, JAPAN ; Data inizio: 2026-03-24; Ora inizio: 16.30; Note orario: CV3 T.06 (presenza ed on line); Referente interno: Alessandra Di Pierro; Riassunto: ABSTRACT : Scaling laws describe how learning performance improves with data, compute, or training time, and have become a central theme in modern deep learning. We study this phenomenon in a canonical nonlinear model: phase retrieval with anisotropic Gaussian inputs whose covariance spectrum follows a power law. Unlike the isotropic case, where dynamics collapse to a two-dimensional system, anisotropy yields a qualitatively new regime in which an infinite hierarchy of coupled equations governs the evolution of the summary statistics. We develop a tractable reduction that reveals a three-phase trajectory: (i) fast escape from low alignment, (ii) slow convergence of the summary statistics, and (iii) spectral-tail learning in low-variance directions. From this decomposition, we derive explicit scaling laws for the mean-squared error, showing how spectral decay dictates convergence times and error curves. Experiments confirm the predicted phases and exponents. These results provide the first rigorous characterization of scaling laws in nonlinear regression with anisotropic data, highlighting how anisotropy reshapes learning dynamics. SHORT BIO: Prof. Minh Ha Quang is the Leader of the Functional Analytic Learning Team at RIKEN. His current research interests focus on machine learning and statistical methodologies using theories and techniques from Functional Analysis and related mathematical fields. In particular, he has been working on theories and methods involving reproducing kernel Hilbert spaces (RKHS), Riemannian geometry, Matrix and Operator Theory, Information Geometry, and Optimal Transport, especially in the Infinite-Dimensional setting. Prof. Minh Ha Quang received his PhD in mathematics from Brown University (Providence, RI, USA) and wrote his dissertation under the supervision of Stephen Smale. Before joining RIKEN, he was a researcher at the Pattern Analysis and Computer Vision group at the Italian Institute of Technology (Istituto Italiano di Tecnologia) in Genoa (Genova), Italy. Prior to Italy, he was a postdoctoral researcher at the University of Vienna, Austria, and the Humboldt University of Berlin, Germany. Join Zoom Meeting: https://univr.zoom.us/j/87279109947 Meeting ID: 872 7910 9947 . Tue, 24 Mar 2026 16:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6950 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6912 Relatore: Francesco Logozzo; Provenienza: Meta - Menlo Park, CA, United States; Data inizio: 2026-03-25; Ora inizio: 12.00; Note orario: Sala Verde (solo presenza); Referente interno: Mila Dalla Preda; Riassunto: Abstract : Privacy and security share the same engineering goal: prevent harm by enforcing constraints. The difference is that privacy failures are often not immediately visible, and are therefore judged against explicit regulatory requirementsmdash;so ldquo;being compliantrdquo; must be engineered, continuously, and demonstrated with evidence.In this talk, drawing on how we build and operate technical safeguards at Meta, Irsquo;ll describe a practical lifecycle for privacy compliance at scale: first defining whatrsquo;s in scope, then preventing new leaks in new code (ldquo;stop the bleedrdquo;), remediating legacy issues, and finally preventing regressions through continuous monitoring. Irsquo;ll then cover how compliance is proven in practice, with automated reminders and workflows, or with targeted human reviews, or with secure-by-default frameworks, or with detection/understanding mechanisms.To ground the discussion, Irsquo;ll close with two concrete examples from Meta: deletion as a secure-by-default safeguard, and data lineage as a detection and enforcement capabilitymdash;illustrated through a sensitive ldquo;religionrdquo; use case. Bio: Francesco Logozzo is a director-level engineer at Meta, where he has played a pivotal role in shaping the companyrsquo;s security and privacy strategy. He joined Meta in 2015, where he has designed, led and grown Zoncolan, the static analysis platform that helps protect billions of people by automatically detecting security and privacy vulnerabilities across Metarsquo;s family of apps, including Facebook and Instagram. Zoncolan is a cornerstone of Metarsquo;s security infrastructure and is responsible for finding more than 50% of the security bugs across Metarsquo;s family of apps.Before Meta, Francesco spent nine years at Microsoft Research, Redmond, WA, contributing foundational work in program analysis and software verification. He studied at the Scuola Normale Superiore of Pisa before earning a PhD in computer science, advised by Radhia Cousot. He has published extensively in top research venues, including POPL, PLDI, OOPSLA, VMCAI, and SAS.Francesco is a frequent keynote speaker at leading academic and industry conferences and is a co-recipient of the IEEE Cybersecurity Award for contributions to scalable security solutions. He is recognized for deep technical expertise, leadership, and sustained impact on the security and privacy landscape. Wed, 25 Mar 2026 12:00:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6912 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6949 Relatore: Esha Gupta; Provenienza: University of Paris-Saclay; Data inizio: 2026-03-25; Ora inizio: 17.00; Note orario: Sala Verde (presenza e remoto); Referente interno: Alessio Cipriani; Riassunto: For a finite-dimensional algebra, it is known from the work of Adachi-Iyama-Reiten that two-term silting complexes are in bijection with functorially finite torsion pairs and support tau-tilting pairs in the module category. Later, more classes were added to these bijections, including complete cotorsion pairs, left-finite semibricks, and left-finite wide subcategories. In this talk, we will generalise the above bijections to arbitrary d-term silting complexes by introducing rsquo;extended module categoriesrsquo; or rsquo;truncated derived categoriesrsquo;. We then provide appropriate generalisations of torsion classes, semibricks, and wide subcategories to extriangulated categories to show that d-term silting complexes are in bijection with functorially finite positive torsion pairs and complete hereditary cotorsion pairs. We will also show them to be in bijection with left-finite semibricks and left-finite wide subcategories in the extended module category. This talk is based on a joint work with Yu Zhou. Link: https://unipd.zoom.us/j/82518660070?pwd=RUpxL1FnZG9yVzFrOCtrM0xYMEZaZz09 Meeting ID: 825 1866 0070 Password: 62542. Wed, 25 Mar 2026 17:00:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6949 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6937 Relatore: Marzia Riso; Provenienza: Centre Inria d’Université Côte d’Azur; Data inizio: 2026-03-30; Ora inizio: 11.00; Note orario: Sala Verde (solo presenza); Referente interno: Umberto Castellani; Riassunto: Abstract: The creation of 2D and 3D assets is a central component of many computer graphics applications, from industrial design to film and video game production. Despite the growing demand for high-quality digital assets, matching a desired target appearance remains a time-consuming and labor-intensive process. This talk presents a series of methods aimed at improving controllability in both procedural and non-procedural asset editing. First, inverse procedural modeling approaches for vector patterns and procedural implicit shapes are introduced, allowing parameter estimation from examples or through in-viewport edits. Finally, a method for structured texture synthesis is presented, expanding a small user-designed sketch into large-scale, high-quality and tileable textures through a diffusion model pipeline. Together, these approaches aim to make asset creation more controllable, efficient, and accessible for both novice and experienced artists. Bio: Marzia Riso is a postdoctoral researcher at the Centre Inria drsquo;Universiteacute; Cocirc;te drsquo;Azur.She received her PhD with honors in Computer Graphics from Sapienza University of Rome, under the supervision of Professor Fabio Pellacini. During her PhD, she also spent six months at Adobe Research in Paris, France. Her research mainly focuses on developing methods to improve controllability in asset editing tasks, including synthesis and inverse procedural modeling techniques for 2D and 3D content. She has also worked on geometry processing applications aimed at extending vector graphics operations to the manifold domain. . Mon, 30 Mar 2026 11:00:00 +0200 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6937 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6915 Relatore: Leonard P. Bos; Provenienza: University of Calgary; Data inizio: 2026-04-20; Ora inizio: 12.30; Note orario: Aula H; Referente interno: Giacomo Albi; Riassunto: Neural Nets generate outputs according to a specific recipe, i.e., they form a certain family of (vector valued) functions, determined by a typically large number of parameters (the weights). Training a Neural Net means to adjust the parameters to produce a desired output, i.e., find a good approximation to a given output function from the family of functions produced by the Net. In this course we will explore, in relation to classical approximation by polynomials and splines, how good an approximation can be so obtained. The course will be completely self contained. Schedule: (TBA). Mon, 20 Apr 2026 12:30:00 +0200 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6915