Seminari - Dipartimento Informatica validi dal 14.10.2025 al 14.10.2026. https://www.di.univr.it/?ent=seminario&lang=it&rss=0 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6720 Relatore: Francesco Leofante; Provenienza: Imperial College London (UK); Data inizio: 2025-10-16; Ora inizio: 14.30; Referente interno: Daniele Meli; Riassunto: The course will focus on the introduction of counterfactual explanations for black-box machine learning models, and their application to reinforcement learning problems. Schedule: 16/10/2025 2:30-4:30 pm, room t02.a 21/10/2025 8:30-10:30 am, room t02.b 23/10/2025 2:30-4:30 pm, room t.02a. Thu, 16 Oct 2025 14:30:00 +0200 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6720 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6718 Relatore: Francesco Leofante; Provenienza: Imperial College (Londra - UK); Data inizio: 2025-10-21; Ora inizio: 14.00; Note orario: Sala Verde (presenza ed on line); Referente interno: Alessandro Farinelli; Riassunto: Abstract : Counterfactual Explanations (CEs) are a leading paradigm for enhancing AI explainability, valued for their intelligibility and alignment with human reasoning. Defined as minimally altered inputs that yield a more desirable AI decision, CEs enable systematic quot;what-ifquot; analysis and automated recourse. However, mainstream CE methodologies focus primarily on single-shot decisions (e.g., in classification), often proving ill-suited for tasks involving sequences of reasoning steps.In this talk, I will present three recent proposals we put forward to generate CEs for sequential decision-making problems. I will consider three different domains ndash; automated planning, neuro-symbolic multi-agent systems, and Markov decision processes ndash; discuss different types of counterfactual edits and present some preliminary solutions to compute CEs for these domains. Short bio : Francesco is an Assistant Professor in the Department of Computing at Imperial College London. His research focuses on AI safety, explainability and alignment, with special emphasis on counterfactual explanations. Since 2022, he leads the project ldquo;ConTrust: Robust Contrastive Explanations for Deep Neural Networksrdquo;, a four-year effort devoted to the formal study of robustness issues arising in XAI. More details about Francesco and his research can be found at https://fraleo.github.io/. https://univr.zoom.us/j/89036897160 ID riunione: 890 3689 7160 . Tue, 21 Oct 2025 14:00:00 +0200 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6718 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6731 Relatore: Mauro Dalla Mura; Provenienza: Grenoble INP; Data inizio: 2025-10-27; Ora inizio: 9.30; Referente interno: Giacomo Albi; Riassunto: TBA Contact: Giandomenico Orlandi. Mon, 27 Oct 2025 09:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6731 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6730 Relatore: Thorsten Altenkirch; Provenienza: University of Nottingham; Data inizio: 2025-10-27; Ora inizio: 16.30; Note orario: (see after abstract for timetable and venue); Referente interno: Peter Michael Schuster; Riassunto: Type Theory is at the same time an alternative to set theory as a mathematical foundation and a programming language originally conceived by the Swedish philosopher and mathematician Per Martin-Louml;f. The course is an introduction to Type Theory which covers the following topics: lambda;-calculus and combinatory logic Propositions as types Classical vs intuitionistic reasoning Dependent types, Pi;- and Sigma;-types Reasoning in Type Theory, Equality Inductive and coinductive types Universes and paradoxes Homotopy Type Theory The Agda system will be used for examples and exercises. Timetable and venue: Monday 27th: 16:30-18:30 - Aula G (2 hrs core lectures) Tuesday 28th: 08:30-10:30 - Aula T.04 (2 hrs core lectures) Wednesday 29th: 08:30-11:30 - Aula T.04 (2 hrs core lectures + 1 hsupplementary lecture) Friday 31st: 16:30-19:30 - Aula T.04 (3 hrs supplementary lectures) All lectures will be held in Ca#39; Vignal 2 and 3, strada Le Grazie 15, Verona. Mon, 27 Oct 2025 16:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6730 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6741 Relatore: Christine Gaßner; Provenienza: University of Greifswald - GER; Data inizio: 2025-11-05; Referente interno: Peter Michael Schuster; Riassunto: Abstract: The BSS-RAM model can be used for characterizing uniform algorithms within a mathematical framework. BSS RAMs over first-order structures are the result of a generalization of various types of abstract machines such as BSS machines and Turing machines. On the one hand, universal machines are not necessary for executing arbitrary programs since any machine has its own program. On the other hand, universal BSS RAMs and, in particular, universal non-deterministic BSS RAMs are helpful in defining complete problems for some classes in different hierarchies of decision problems. Here, we will discuss sufficient conditions for their existence. We will consider first-order structures that contain only a finite number of operations and relations, including or excluding the identity relation, and with or without constants. Christine Gaszlig;ner (University of Greifswald) Short CV: After graduating in mathematics in 1981, Christine Gaszlig;ner began to study mathematical logic and work in Guuml;nter Asser#39;s research group. In 1985, she received a doctorate degree for her thesis on the axiom of choice in second-order Henkin logic in Greifswald. Since 1993 she has been dealing with abstract computation over various structures. Among other things, she investigated the possibility of constructing structures with P = NP. She completed her habilitation thesis in 2013 and received the venia legendi for mathematics from the University of Greifswald and the facultas docendi for theoretical computer science. From 2015 to 2023, she was the head of the research group Theory of Computability over Algebraic Structures and responsible for the teaching modules in Mathematical Logic and Theoretical Computer Science at the Institute of Mathematics and Computer Science in Greifswald. Wed, 5 Nov 2025 00:00:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6741 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6732 Relatore: David N. Raynolds; Provenienza: University of Warsaw; Data inizio: 2025-11-16; Ora inizio: 9.30; Referente interno: Giacomo Albi; Riassunto: TBA Contact: Giacomo Albi. Sun, 16 Nov 2025 09:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6732 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6735 Relatore: Giulia Cavicchioni; Provenienza: German Aerospace Center; Data inizio: 2025-11-18; Ora inizio: 10.30; Referente interno: Francesca Mantese; Riassunto: Beyond its role in information theory, coding theory plays a crucial role in cryptography, particularly in post-quantum cryptography. Traditional public-key cryptosystems rely on problems such as integer factorization and discrete logarithms on elliptic curves. While these problems are currently intractable for classical computers, Shorrsquo;s algorithm allows quantum computers to solve them in polynomial time. Post-quantum cryptography aims to create secure algorithms based on computationally hard problems that remain resistant to quantum attacks, with a focus on NP-complete problems. Code-based cryptography refers to any cryptographic system whose security relies on hard problems from algebraic coding theory. Classically, this problem consists of decoding a random linear code, which was proven to be NP-complete in 1978. That same year, McEliece introduced the first code-based cryptosystem. The core idea is to select a code with an underlying algebraic structure that enables efficient decoding, and then disguise it as a seemingly random. Encryp- tion works by encoding the message into a codeword and then intentionally adding errors. With the knowledge of the secret code, one can efficiently decode and recover the message, while an attacker is left with the challenge of decoding a random linear code. In this mini-course, we will introduce code-based cryptography, delving into the mathematical foundations of these systems. In particular, we will focus on the main approach to solving the decoding problem, that is, the information set decoding (ISD) algorithms. Lastly, we will discuss how to determine whether a code has a specific algebraic structure or if it is a random code. Shedule: 18/11 10:30-13:30, 19/11 12:30--14:30, 25/11 10:30-13:30, 26/11 12:30-14:30, 27/11 8:30-10:30 . Tue, 18 Nov 2025 10:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6735 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6733 Relatore: Giada Fiandaca; Provenienza: INRIA, Marseille; Data inizio: 2025-12-01; Ora inizio: 10.30; Referente interno: Giacomo Albi; Riassunto: TBA Contact: Giandomenico Orlandi/ Giacomo Albi. Mon, 1 Dec 2025 10:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6733 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6734 Relatore: Marcello Carioni; Provenienza: University of Twente; Data inizio: 2025-12-14; Ora inizio: 8.30; Referente interno: Giacomo Albi; Riassunto: TBA Contact: Giandomenico Orlandi. Sun, 14 Dec 2025 08:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6734 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6726 Relatore: Dante Kalise; Provenienza: Imperial College London; Data inizio: 2025-12-15; Ora inizio: 14.30; Note orario: da definire; Referente interno: Giacomo Albi; Riassunto: This course introduces the fundamental ideas and computational methods behind optimal control and data-driven modelling.In this short course, we will study how to incorporate elements of machine learning into optimal control design. The course will focus on fundamentals on optimal control: dynamic optimization, linear-quadratic control, dynamic programming and Pontryagin#39;s maximum principle. Nonlinear optimal control. Approximation methods in high dimensions are also discussed such as polynomial approximation, deep neural networks. Optimization techniques: LASSO regression, stochastic gradient descent, training neural networks. Finally, combining the first two parts, we will study the construction of data-driven schemes for the approximation of high-dimensional nonlinear control laws. . Mon, 15 Dec 2025 14:30:00 +0100 https://www.di.univr.it/?ent=seminario&lang=it&rss=0&id=6726