Seminari - Dipartimento Computer Science Seminari - Dipartimento Computer Science validi dal 21.05.2024 al 21.05.2025. On the construction of compact high-order semi-implicit DG schemes and applications Relatore: Dott. Maurizio Tavelli; Provenienza: Università di Bolzano; Data inizio: 2024-06-05; Ora inizio: 11.00; Note orario: Sala Verde; Referente interno: Elena Gaburro; Riassunto: Abstract: In this talk we present a class of semi-implicit methods on unstructured staggered meshes in 2D and 3D. The numerical solution is computed using a DG approach, but on staggered meshes. We will consider several PDE systems such as the incompressible/compressible Navier Stokes and linear elasticity. The proper implicit discretization of the terms associated to the fast scale, i.e. the pressure, allows to obtain a weak CFL stability restriction. The resulting maximum time step is then limited by the local velocity and not by the celerity, allowing the algorithm to perform well in the low Mach number regime. The resulting linear system is symmetric and at least semi-positive definite for most of the applications, so an efficient matrix-free implementation of the conjugate gradient method can be used for the solution of the linear part. Thanks to the chosen discretization, the stencil results very compact and involves only the direct neighbors. Besides, high order in time can also be obtained using a genuine space-time DG or a simple IMEX approach. Several extensions to ph-adaptivity and GPU computing will be discussed, as well as a possible application to networks of compliant pipes, useful to simulate physiological flows. . Wed, 5 Jun 2024 11:00:00 +0200 Visual Explainability and Robustness through Language Relatore: Riccardo Volpi; Provenienza: Naver Labs Europe - France; Data inizio: 2024-06-04; Ora inizio: 13.45; Note orario: Aula C (solo presenza); Referente interno: Vittorio Murino; Riassunto: Abstract: In recent years, the vision-and-language paradigm has revolutionized the way we learn and rely on computer vision models. A major drawback of learning visual representation has always been the lack of data: when coupling our vision model with large, pre-trained language models we can partially mitigate these issues by building on large amounts of previously learned information. In this talk, we will discuss how using language can i) broaden the comfort zone of model vision models for tasks such as object detection and classification and ii) improve their interpretability. We will go through the basis of the vision-and-language paradigm, highlight some of its inherent limitations and discuss some innovative solutions, for example to make CLIP-like models robust to arbitrary vocabularies selected by the user. Tue, 4 Jun 2024 13:45:00 +0200 Young Researchers Seminars, Maths Applications & Models (Sfragara, Vardanyan) Relatore: Matteo Sfragara; Viktorya Vardanyan; Provenienza: Stockholm University; Università di Trento; Data inizio: 2024-06-03; Ora inizio: 16.30; Note orario: Sala Verde; Referente interno: Giacomo Canevari; Riassunto: Matteo Sfragara (Stockholm University) Queue-based random-access protocols for wireless networks In this talk, we discuss mathematical models that address fundamental challenges in wireless networks. We first introduce Carrier-Sense Multiple-Access (CSMA) protocols, distributed algorithms that involve randomness to prevent the devices to transmit simultaneously and hencetheir signals to interfere with each other. Thesemodels can be viewed as interacting particle systems on graphs, where the interference is captured by a hard-core interaction model.Wedescribe how they exhibit metastability and how understanding metastability properties iscrucial to design mechanisms to counter starvation effectsand improve the performance of the network. In particular, we focus on random-access models where the transmission rates depend on the queues at the nodes. We discuss three different network topologies: we start with complete bipartite networks, we then generalize our results to arbitrary bipartite networks, and finally we explore dynamic bipartite networks in which the interference graph changes over time, which allows us to capture some effects of user mobility.This talk is based on three joint workswithFrank den Hollander (Leiden University), Sem Borst (Eindhoven University of Technology) and Francesca R.Nardi (University of Florence). Viktorya Vardanyan (Universitagrave; di Trento) Mean field games with terminal state constraints We investigate a mean field game (MFG) characterized by state dynamics modeled through stochastic differential equations influenced by both idiosyncratic and common noise. These dynamics are subject to the constraint that the terminal state variable resides withina nonempty convex closed set. Additionally, the mean field interaction affects both the state and control in the dynamics and the costs. Motivated by the work of Ji and Zhou [Comm. Inf. Syst. 6(4), 2006], we introduce an auxiliary MFG problem and establish the stochastic maximum principle (SMP) for an auxiliary optimization problem with fixed flows. Throughthe formulation of a suitable forward-backward stochastic differential equation (FBSDE) of conditional McKean-Vlasov type, we demonstrate the existence of its solution and confirm itsrole as a MFG equilibrium. Furthermore, we apply these findings within a financial context. The talk is based on a joint work with Luca Di Persio (University of Verona) and Luciano Campi(University of Milan). Further information at this link. Mon, 3 Jun 2024 16:30:00 +0200 How to extract programs from proofs Relatore: Dr. Ingo Blechschmidt; Provenienza: Universität Augsburg (Germania); Data inizio: 2024-06-03; Ora inizio: 14.00; Note orario: - data e ora da confermare - 6 ore in totale; Referente interno: Peter Michael Schuster; Riassunto: Construction, realisability and double negation. Mon, 3 Jun 2024 14:00:00 +0200 Designing Reliable Reinforcement Learning Agents Relatore: Thiago D. Simão; Provenienza: Eindhoven University of Technology; Data inizio: 2024-05-29; Ora inizio: 12.30; Referente interno: Alberto Castellini; Riassunto: Safety is a crucial concern when deploying reinforcement learning (RL) algorithms in real-world scenarios. In this two-part lecture series, we delve into safety considerations from two perspectives: ensuring reasonable performance and adhering to predefined constraints. - PART 1. In the first segment, we investigate the offline setting where the RL agent solely accesses a fixed dataset of prior trajectories, devoid of direct interaction with the environment. Given the availability of the behavior policy responsible for data collection, the primary challenge is crafting a policy that outperforms such behavior policy. We study algorithms that leverage the behavior policy to compute an improved policy with high probability and discuss how to improve their sample efficiency. - PART 2. Transitioning to the latter segment, we confront the limitations inherent in specifying the behavior expected from an agent solely via a reward function. We introduce a model that mitigates this issue using constraints, and we discuss how to compute the corresponding optimal policy when the problem is known. Finally, we study algorithms that can efficiently explore the environment and eventually converge to an optimal policy when the model is unknown. Schedule: May 29, 12.30-14.30 (Room B) May 29, 15.30-17.30 (Room 1.02) May 31, 15.30-17.30 (Room 1.02) June 5, 12.30-14.30 (Room B) June 5, 15.30-17.30 (Room 1.02 June 7, 15.30-17.30 (Room 1.02) The minicourse is related to the quot;Reinforcement Learningquot; course (Master in Artificial Intelligence). Wed, 29 May 2024 12:30:00 +0200 Quantum Machine Learning Relatore: Davide Pastorello; Provenienza: Alma Mater Studiorum, Università di Bologna, Department of Mathematics; Data inizio: 2024-05-27; Ora inizio: 14.00; Note orario: Aula I; Referente interno: Alessandra Di Pierro; Riassunto: Abstract: The application of quantum computing to machine learning offers some interesting solutions characterized by a quantum advantage with respect to the classical counterparts in terms of time and space complexity, expressive power, generalization capability, at least on a theoretical level. In this lecture, we give an overview on some quantum machine learning algorithms including distance-based classifiers and different kinds of quantum neural networks. In particular, we focus on the most promising approaches relating to existing and near-term quantum machines. Short cv: Davide Pastorello received the M.Sc. degrees in Physics and the Ph.D. degree in Mathematics from the University of Trento in 2011 and 2014, respectively. From 2015 to 2019 he was postdoc researcher at Deptartment of Mathematics, University of Trento, also with a 2-year grant from Fondazione Caritro as P.I. of a project on quantum computing. From 2020 to 2023 he was Assistant Professor at the Department of Information Engineering and Computer Science (DISI), University of Trento. Since 2023, he is Assistant Professor at Deptartment of Mathematics, University of Bologna. His main research interests are: mathematical foundations of quantum mechanics, mathematical methods in quantum computing and quantum machine learning. . Mon, 27 May 2024 14:00:00 +0200