Courses & Seminars


Students can select courses, under the guidance of supervisors, from those organized by the PhD program in Smart Computing as listed below as well as from those proposed by partner Universities:
In addition, courses can be also taken from the other PhD programs of the School of Engineering of the University of Florence: For all the above courses the authorization of the PhD commitee is not needed. It is instead required for any other courses. Rules for accounting credits (CFU) for courses and seminars can be found under the Procedures page. The program for the academic year 2025/2026 of complemantary and soft skills is also available here

Courses
  • Instructor: Dr. Alessandro Betti (IMT Scuola Alti Studi, Lucca)
    Period: 14 Sept. 2026 (h.9-13) room TBD, 16 Sept. 2026 (h.9-13) room TBD, 21 Sept. 2026 (h.9-13) room TBD, 23 Sept. 2026, 28 Sept. 2026 (h.9-13) room TBD
    Location: University of Florence
    Duration: 20 hours
    Abstract: This course explores the structural parallels between physical laws and learning dynamics through the unifying lens of variational principles. Starting from the calculus of variations and the principle of least action in classical mechanics, the course develops a rigorous framework in which learning algorithms are interpreted as dynamical systems evolving in an energy landscape. The course introduces novel perspectives on optimization, regularization, and stability in machine learning. A central goal is to formulate online and recurrent learning as an optimal control problem, uncovering connections with Pontryagin’s principle and Hamiltonian dynamics. Students will gain both theoretical insight and practical tools to design learning algorithms grounded in physical and variational reasoning.
    Additional information: 5 classes, 4h per class     link
  • Instructor: Dr. Pietro Bongini, Dr. Niccolò Pancino (UniSI)
    Period: September 2026
    Location: University of Siena
    Duration: 12 hours
    Abstract:
    Additional information: 4 classes, 3h per class link
  • Instructor: Prof. Jiannong Cao, Chair Professor Hong Kong Polytechnic University, IEEE Fellow, and ACM Distinguished Member
    Period: 6 July 6 (h.14-18), 7 July (h.9-13), 8 July (h.9-13), 9 July (h.9-13), 10 July (h.9-13)
    Location: Meeting Room, Dept. of Information Engineering, Largo Lazzarino 1, Pisa
    Duration: 20 hours
    Abstract: Advanced doctoral-level research-oriented course, centered on collaborative edge AI and its applications in AIoT. It systematically covers the various aspects of edge-AI research and applications, including edge computing paradigms, collaborative framework, task scheduling, edge AI model training/inference, and future research directions. This course focuses on academic challenges, state-of-the-art technologies, key research issues, and major application cases, aiming to cultivate students’ independent research ability in collaborative edge computing and intelligence. Course Objectives. Introduce AIoT ecosystem, core enabling technologies, and application areas; Overview edge computing paradigms, edge computing and AI principles, and key research challenges; Understand collaborative edge computing architecture. Framework and its components; Study resource virtualization and management mechanisms; Study the theory and techniques of collaborative edge task scheduling; Study the collaborative edge AI model training and inference algorithms and techniques; Investigate collaborate edge AI application scenarios and develop solutions; Discuss research methods and explore future research topics. Course Contents in brief. AIoT and Its Key Enabling Technologies; Edge Computing and Edge AI; Collaborative Edge AI; Collaborative Task Scheduling; AI Model Training, Inference and Future Research Topics. Final Exam. Class participation and in-class presentation (20%); After-class assignments (30%); Doctoral-level research innovation proposal (50%).
    Additional information: 5 classes, 4h per class. course registration link
  • Instructor: Dr. Claudio Ferrari (UniFI)
    Period: 29 June 2026 (h.14-17) room 226, 30 June 2026 (h.14-17) room 226, 2 July 2026 (h.14-17) room 217, 3 July 2026 (h.14-17) room 221
    Location: University of Florence
    Duration: 12 hours
    Abstract: This course will provide an overview of the landscape of generative AI, starting from early architectures such as Variational Auto-Encoders (VAEs) and Generative Adversarial Networks (GANs) to the more recent Diffusion Models. The course will also explore the challenge of fairly evaluating such models and the extension to multi-modal solutions where different data modalities interact and are used to condition the generation of data samples. Specific applications in the domain of human generation and understanding will be presented.
    Additional information: 4 classes, 3h per class link
  • Instructor: Dr. Cosimo Rulli (ISTI-CNR, Pisa), Dr. Franco Maria Nardini (ISTI-CNR, Pisa), Prof. Rossano Venturini (UniPI), Dr. Salvatore Trani (ISTI-CNR, Pisa)
    Period: 22 June 2026 (h.9:00-13:00) room 006, 23 June 2026 (h.9:00-13:00) room 101, 25 June 2026 (h.9:00-13:00) room 119, 26 June 2026 (h.9:00-13:00) room 118
    Location: Centro Didattico Morgagni, UniFI, viale Morgagni 40-44, Firenze
    Duration: 16 hours
    Abstract: This PhD course focuses on Information Retrieval and discusses the state-of-the-art and the challenges in the two main areas of Web search. i) indexing and ii) query processing. The course introduces each area by discussing the state of the art in the field and by presenting the open research questions. The course emphasizes query processing, a research line where machine learning is important to advance the state of the art. After introducing the different query processing techniques, the course introduces supervised techniques explicitly focused on targeting the ranking problem and discusses several time and space efficiency/effectiveness trade-offs in query processing. The course will also provide an in-depth analysis of query processing techniques employing transformer-based large language models. Four hands-on sessions will cover indexing and query processing of public Web collections.
    Additional information: Remote participation is possible for motivated reasons. registration form
  • Instructor: Dr. Simone Magistri (University of Florence)
    Period: 28 April 2026 (h.16-19) room 57, 29 April 2026 (h.16:00-19:00) room S12, 7 May 2026 (h.16:00-19:00) room 171, 8 May 2026 (h.14:00-17:00) room 029
    Location: via Santa Marta 3, School of Engineering
    Duration: 12h
    Abstract: The increasing size and cost of modern foundation models call for efficient strategies to update, adapt, and reuse existing deep neural networks. This course focuses on two closely related approaches that address this challenge; Continual Learning and Model Merging. Continual Learning studies how a single model can be updated incrementally as new data becomes available, without performing full retraining or storing past data. Model Merging investigates how multiple available pretrained models can be combined into a single one without direct access to the original training data. These two paradigms are strongly connected; model merging techniques can support continual learning in non-stationary data scenarios, while continual learning principles provide insights for effective model fusion. Both aim at efficient model reuse under resource, data, and privacy constraints, offering practical alternatives to conventional expensive retraining pipelines that require access to all data at once. The course will explore the standard settings of Continual Learning, including task-incremental, class-incremental, and domain-incremental scenarios. It will cover approaches ranging from classical methods such as Elastic Weight Consolidation (EWC) and Learning Without Forgetting (LwF) to more recent advances on foundation models. On the Model Merging side, the course will review techniques such as Weight Averaging and Task Arithmetic, as well as more recent methods, and discuss how recent advances in model merging have been applied to continual learning, and vice versa, highlighting their connection..
    Additional information: link
  • Instructor: Prof. Pericle Perazzo (UniPI)
    Period: 5-6-7-8 May 2026 (h.14:30-17:30)
    Location: Dept. of Information Engineering (DII), Largo Lazzarino 1, Pisa
    Duration: 12 hours
    Abstract: This PhD course provides an in-depth exploration of Bitcoin and Blockchain technologies, offering participants a comprehensive understanding of their foundational principles and advanced concepts. The course will examine the cryptographic primitives and the decentralized paradigm that underpins Bitcoin, the structure of transactions, their validation and propagation within the network, the Proof-of-Work Consensus, the concept of Lightweight Clients. Building on this, we will explore Bitcoin’s scripting capabilities and the evolution toward programmable money. Through theoretical insights and practical examples, participants will gain a solid foundation for research and innovation in blockchain systems.
    Additional information: link
  • Instructor: Prof. Tiberio Uricchio (UniPI)
    Period: April-May 2026
    Location: University of Pisa
    Duration: 12h
    Abstract: The rapid rise of agentic coding, where large language model (LLM)-based agents autonomously plan, execute, and verify software development tasks, is reshaping how computational research is conducted across all disciplines. Recent empirical studies show that 15–22% of GitHub projects already incorporate coding agents (Rasheed et al., 2026), while industry surveys report 84% of developers using AI tools (Stack Overflow, 2025). Yet a randomized controlled trial by METR (2025) found that experienced developers were 19% slower with AI tools, revealing a significant gap between tool availability and effective adoption – a gap this course aims to close. This course equips PhD researchers with both the theoretical foundations and practical skills to leverage agentic programming effectively in their research. The course covers the emerging academic landscape through state of the art surveys (Wang et al., arXiv:2508.11126), context engineering principles (Anthropic, 2025), the Model Context Protocol ecosystem (now under the Linux Foundation’s Agentic AI Foundation), and multi-agent architectures for parallel development. Students gain hands-on experience with Claude Code as the primary tool while surveying the broader landscape including Gemini CLI, open-source agents, and local models. Research specific applications include AI-assisted literature review with citation verification, data analysis pipelines with reproducibility guarantees, and scientific writing with integrity protocols. The course critically examines both capabilities and limitations, including hallucination risks, sycophancy patterns, and research integrity challenges.
    Additional information: 4 classes, 3h each. One class per week
  • Instructor: Prof. Giovanni L.C. Masala and Dr. Ioanna Giorgi (University of Kent, School of Computing, UK)
    Period: 24 March 2026 (h. 11-13 and h. 14-18), 27 March (h. 8-10, final multiple choice test)
    Location: Meeting Room DII – Polo A, Largo L. Lazzarino 1, Pisa
    Duration: 8 hours
    Abstract: Cognitive models play a central role in advancing autonomous and adaptive behavior in robotics by providing computational accounts of perception, memory, language, and decision-making. Rather than relying solely on task-specific control architectures, cognitively inspired systems aim to reproduce general mechanisms underlying human cognition, enabling learning, reasoning, and flexible interaction with the environment. This course introduces key principles of cognitive modeling in robotics, with particular emphasis on neural and symbolic–subsymbolic approaches. As a representative example, we consider the ANNABELL (Artificial Neural Network with Adaptive Behavior Exploited for Language Learning) model, which implements language acquisition and use through biologically inspired neural mechanisms and working-memory structures. ANNABELL illustrates how complex cognitive functions can emerge from the interaction of simple neural components, offering insights into scalable, brain-inspired architectures for cognitive robotics.
    Additional information: registration link
  • Instructor: Prof. Luigi Chisci (UniFI)
    Period: 2-3-5-9-10 February 2026 (h. 9:00-13:00)
    Location: online via Google Meet (mandatory registration using the following link)
    Duration: 20h
    Abstract: The course will provide an overview of advanced research in estimation, specifically concerning the two topics of multi-agent and multi-object estimation. Multi-agent estimation deals with a network of agents with sensing, processing and communication capabilities that aim to cooperatively monitor a given system of interest. Multi-object estimation aims to detect an unknown number of objects present in a given area and estimate their states. Special attention will be devoted to the Kullback-Leibler paradigm for fusion of possibly correlated information from multiple agents and on the random-finite-set paradigm for the statistical representation of multiple objects. Applications to distributed cooperative surveillance, monitoring and navigation tasks will be discussed.
    Additional information: Links to online lectures
    link lecture 1, 2 February 2026, h. 9:00-13:00
    link lecture 2, 3 February 2026, h. 9:00-13:00
    link lecture 3, 5 February 2026, h. 9:00-13:00
    link lecture 4, 9 February 2026, h. 9:00-13:00
    link lecture 5, 10 February 2026, h. 9:00-13:00
  • Instructor: Prof. Sajal Das, IEEE Fellow, Missouri University of Science and Technology, USA
    Period: 17-18-19 November, 2025 (h. 10:30-12:30 and 14:00-16:30), 20 November, 2025 (h. 10:30-12:30)
    Location: Aula Riunioni, Sixth Floor, Dept. of Information Engineering (DII), Largo Lazzarino 1, Pisa
    Duration: 16 hours
    Abstract:
    Additional information: link
Seminars
  • Instructor: Prof. Nico Pietroni, School of Computer Science at the University of Technology Sydney (UTS)
    Period: 1 July 2026, h.11
    Location: room 204, Centro Didattico Morgagni, viale Morgagni 40-44, Firenze
    Duration: 1h 30'
    Abstract: Computational design is a rapidly evolving discipline that is transforming fields such as architecture, industrial design, digital fabrication, and art. By integrating algorithmic methods into the design process, it enables designers to focus on creativity and aesthetics while automatically accounting for practical constraints such as manufacturability, structural stability, material usage, cost, assembly, and fabrication processes. In this talk, I will present a series of novel computational design and geometry-processing techniques that bridge the gap between digital design and physical realization. Through examples spanning advanced manufacturing, architectural geometry, and creative applications, I will demonstrate how optimization algorithms can simultaneously address functional requirements, geometric complexity, and fabrication constraints, enabling the creation of innovative designs that can be efficiently and reliably brought into the real world.
    Additional information: link to the speaker's website
  • Instructor: Prof. Bjarne Stroustrup, Columbia University, New York City
    Period: 9 May 2026, h.10
    Location: room 018, building D6, Campus Novoli, Firenze
    Duration: 1h 30m
    Abstract: This talk presents programming techniques to illustrate the facilities and principles of C++ generic programming usin concepts. Concepts are C++’s way to express constraints on generic code. As an initial example, it provides a simple type system that eliminate narrowing conversions and provides range checking. Concepts are used throughout to provide user-defined extensions to the type system. The aim is to show their utility and the fundamental ideas behind them, rather than to provide a detailed or complete explanation of C++’s language support for generic programming or the extensive support provided by the standard library. The final sections briefly present design rationales and origins for key parts of the concept design, including uniform treatment of types, use patterns, the relationship to Object-Oriented Programming, value arguments, syntax, concept type-matching, definition checking and static reflection (a C++26 improvements in the support of general programming).
    Additional information: link
  • Instructor: Dr. Danilo Pau, STMicroelectronics System R&D, IEEE Fellow
    Period: 29 April 2026, h.9:00
    Location: plesso Santa Marta, room 049 CDM, via Santa Marta 3, Florence
    Duration: 4h
    Abstract: Moving inference from the cloud to the edge (TinyML) reduces cloud dependency but creates hardware heterogeneity challenges. ST addresses this issue with its ST EdgeAIunified core software technology, capable of translating models from Google Keras, TensorFlow Lite, and ONNX into C code optimized for microcontrollers and sensors. With the advent of cloud-based Generative AI and its enormous environmental impact, the research community is now challenged to develop revolutionary solutions to bring generative workloads directly to the edge.
    Additional information: link
  • Instructor: Prof. Tiberio Uricchio, University of Pisa
    Period: 26 March 2026, h.15
    Location: Room Pacinotti – Largo L. Lazzarino 1, Pisa
    Duration: 1h
    Abstract: This seminar traces the arc of Generative AI from transformer foundations to autonomous agents, following two parallel tracks that increasingly converge. On the language side, the path runs from large language models through prompting and context-engineering techniques to the emerging paradigm of agentic programming, where AI systems plan, execute, and verify code autonomously. On the vision side, the same transformer revolution has driven diffusion models for image and video generation and, more recently, world models that learn to simulate visual environments. These trajectories meet in today's multimodal agents, systems that reason across text, images, and code. Drawing on recent adoption data, critical case studies, and practical examples, the talk offers a coherent map of the generative AI landscape and its practical implications.
    Additional information: registration link
  • Instructor: Prof. Stefano Pietropaoli, University of Florence
    Period: 26 March 2026, h.
    Location: room 327 Centro Didattico Morgagni, viale Morgagni 44
    Duration: 1h
    Abstract: Lo sviluppo dell’intelligenza artificiale sta modificando profondamente le modalità con cui vengono prodotte decisioni, organizzato il lavoro e governati i processi sociali. L’intervento propone una riflessione introduttiva sui principali problemi giuridici ed etici connessi alla crescente diffusione dei sistemi decisionali algoritmici. Dopo una breve ricostruzione del rapporto tra tecnica, potere e diritto, verranno analizzati alcuni ambiti particolarmente significativi, dalla decisione automatizzata alla sorveglianza digitale, fino alle trasformazioni del lavoro e alle implicazioni geopolitiche delle tecnologie di IA. L’obiettivo è offrire una chiave di lettura giuridica e filosofica delle tecnologie informatiche, evidenziando come le scelte tecniche incorporino sempre anche scelte normative e politiche.
    Additional information: link
  • Instructor: Prof. Sajal Das, IEEE Fellow, Missouri University of Science and Technology, USA
    Period: 21 November 2025, h.15:00
    Location: room 211 Centro Didattico Morgagni, viale Morgagni 44
    Duration: 1h
    Abstract: Global crop losses due to pests and diseases currently exceed 40%, resulting in economic losses of over $220 billion annually, which are projected to worsen under climate change. In this talk, we will present a cyber-agriculture framework that enables real-time, data-driven integrated pest management (IPM) by integrating AI/ML, IoT sensors, drones, and edge-cloud computing with a goal to design scalable AgTech solutions for resilient and sustainable crop protection. After outlining agronomic challenges and our vision of smart connected farms, we will present our novel machine learning pipeline for IPM. (i) CNN classifiers for pest detection achieving up to 95% mean average prediction, (ii) RL-motivated dynamic split-learning algorithm that adapts computation across devices and cloud server under resource constraints; (iii) a multimodal scheme that fuses weather, spectral indices, and soil data through tailored network branches and late-fusion regression, achieving robust pest-pressure forecasts; and (iv) personalized agronomic guidance via LLMs. Prototype testbed implementation using Raspberry Pi, camera sensors, and Long Range (LoRa) communication with real datasets will validate predictions and inference for pest detection
    Additional information: link
  • Instructor: Prof. Mohamed Daoudi, University of Lille, France
    Period: 13 November 2025, h.11:00,
    Location: room 204 CDM, viale Morgagni 44
    Duration: 1h
    Abstract: In this talk, I will present some of our recent results on human body shape analysis, learning human interactions, and how computer vision can serve as a tool for understanding mental health disorders.
    Additional information: link
  • Instructor: Dr. Thomas Besnier, University of Lille, France
    Period: 6 November 2025, h.11:15
    Location: room 006 CDM, viale Morgagni 44
    Duration: 1h
    Abstract: We introduce a novel framework to learn mesh deformations and interpolations at the point level, allowing localized control for Partial Non-Rigid Deformations and interpolations of Surfaces (PaNDaS). PaNDaS learns a per-face feature field on the source mesh and fuses it with a global encoding of the target, a deformation generator predicts a Jacobian field and recovers a smooth displacement, enabling precise regional control, pose mixing, and transferable local edits. Unlike previous approaches, our method can restrict the deformations to specific parts of the shape in a versatile way. Across various human body part datasets, PaNDaS achieves state-of-the-art interpolation accuracy and stronger locality than methods based on global shape codes or handles, while remaining robust to remeshing. We also demonstrate several localized shape manipulation tasks and show that our method can generate new shapes by combining different input deformations.
    Additional information: link

Universities of Florence, Pisa and Siena • 2015-2026