Fabio Petroni

This week, Andrea Mannocci, Data Scientist working on the OpenAIRE Graph at National Research Council of Italy (CNR-ISTI), will be in Tokyo to present the OpenAIRE Graph at the 140th GIST Seminar organised by the GRIPS Innovation, Science and Technology Policy Program. In this seminar, Andrea will introduce how the OpenAIRE Graph functions as a large-scale open scholarly knowledge graph that connects publications, datasets, software, researchers, institutions, projects, and funding information through rich semantic relationships. The 140th GIST Seminar 📅 12 March 2026 | 10:30–12:00 CET 📍 GRIPS, Tokyo (hybrid) We look forward to sharing how open scholarly data infrastructures like the #OpenAIREGraph enable new insights for science policy, research assessment, and Open Science monitoring. More information and registration: https://lnkd.in/drFQTG5q #OpenScience #OpenAIRE #ResearchData #ScholarlyCommunication

9

🆕 A new article by Laura Meneghetti, Edoardo Bianchi, Nicola Demo, and Gianluigi Rozza has been published on the Journal of Systems Architecture by Elsevier: “Plug-and-play neural compression: A knowledge distillation framework with flexible dimensionality reduction”. 🧠 The paper proposes a plug-and-play framework to compress neural networks: it relies on knowledge distillation, a technique in which a smaller model learns to replicate the behavior of a larger one. This approach manages to reduce model size and improve efficiency while maintaining strong predictive performance. 🔗 Further information is available here: https://lnkd.in/dxuwk7_k #SISSA #mathLab #ScientificComputing #ScientificMachineLearning #NeuralNetworks

18

🚀 New Journal Paper Out! We’re back with fresh insights — this time focusing on how CNF encodings affect SAT and SMT enumeration! 📄 Our new paper, "On CNF Conversion for SAT and SMT Enumeration", just appeared in the Journal of Artificial Intelligence Research (JAIR)! Gabriele Masina Giuseppe Spallitta Roberto Sebastiani 🔍 What’s new? ✅ We show that classical CNF transformations like Tseitin and Plaisted-Greenbaum—designed for satisfiability—are often harmful for enumeration. ✅ We prove and demonstrate that preprocessing the formula into NNF before CNF conversion can dramatically reduce the number of partial models and runtime. ✅ Extensive experiments confirm these findings, with up to orders of magnitude improvement. ✅ Our analysis applies to both SAT and SMT, and supports disjoint and non-disjoint enumeration. 📘 Curious? Read the full paper here 👉 https://lnkd.in/d4DawQnq #SAT #SMT #AllSAT #AllSMT #FormalMethods

76

1 commento

Alessia Borghini is a first-year PhD student in the National Doctorate Programme in Artificial Intelligence under the supervision of Fabrizio Silvestri. Her research focuses on the application of Mechanistic Interpretability techniques to improve #AISafety. Her work aims to better understand the internal behaviour of Large Language Models, contributing to their alignment, transparency, and reliability. Previously, she obtained a Bachelor’s degree in Computer Engineering and a Master’s degree in Artificial Intelligence and Robotics from Sapienza University of Rome. She is always open to discussing research ideas and potential collaborations, feel free to get in touch! #ArtificialIntelligence #MechanisticInterpretability #LLMs #PhD

31

1 commento

Happy to share that our tutorial has been accepted at International Conference on Software Architecture ICSA #ICSA2026 🎉 “Mining Architectural Decisions and Trade-Offs from ADRs and Pull Requests with #LLMs at Scale” Ruoyu Su Matteo Esposito Davide Taibi In this tutorial, we address a key challenge in software architecture: how to systematically extract architectural decisions and their rationale from fragmented artifacts such as ADRs, pull requests, and issue discussions. We will present: -How to leverage LLMs for architectural decision mining at scale -A reproducible pipeline using HPC (SLURM) and efficient inference (vLLM) -Practical strategies to ensure rigor, trustworthiness, and validation when using LLMs -Lessons learned from running large-scale experiments across software repositories The tutorial is designed for both researchers and practitioners interested in: ➡️ Architectural knowledge and evolution ➡️ AI-driven software architecture analysis ➡️ Scalable and reproducible empirical methods Department of Mathematics and Computer Science (IMADA), University of Southern Denmark (SDU) University of Oulu #ICSA2026 #SoftwareArchitecture #LLMs #AI #EmpiricalSE #ArchitectureDecisions

44

CFP: QUANCOM 2026 – International Conference on Quantum Computing Infrastructure & Technologies I am pleased to share that the Call for Papers for QUANCOM 2026 is now open. 📍 Trento, Italy | 1–4 September 2026 📣 Submission site: Hosted by Università di Trento, with the strong support of Prof. Fausto Giunchiglia and Prof. Marco Roveri, QUANCOM 2026 aims to bring together researchers, engineers, industry leaders, and policymakers to advance the foundations and infrastructure of quantum computing. As quantum technologies transition from laboratory prototypes to scalable systems, infrastructure, integration, and reliability become central challenges. QUANCOM 2026 focuses on the system-level questions that will shape the next decade of quantum computing. We welcome contributions in areas including: 🔹 Quantum hardware architectures & scalable qubit technologies 🔹 Quantum communication networks & distributed systems 🔹 Quantum software stacks, compilers & programming frameworks 🔹 Hybrid quantum–classical systems & HPC integration 🔹 Quantum cloud platforms & orchestration 🔹 Error correction, fault tolerance & system reliability 🔹 Security, cryptography & post-quantum readiness 🔹 Applications, benchmarking & industrial use cases 📚 Proceedings: Springer CCIS Indexed in DBLP, EI Compendex, INSPEC, Scopus, SCImago ⏳ Key deadlines • Submission: 25 May 2026 • Notification: 15 July 2026 • Camera-ready: 10 August 2026 I look forward to welcoming the global quantum computing community to Trento in 2026.

217

2 commenti

Human motion analysis software is widely used across domains, such as healthcare and Industry 5.0, but often suffers from noisy or incomplete pose data. Filtering techniques, ranging from simple low-pass filters to advanced state observers, are essential for denoising and data completion. This survey reviews state-of-the-art filtering methods for 3D positional data from marker-based and marker-less motion capture, offering a clear taxonomy and summarizing key concepts and real-world applications.

15

🗺️ Explainable Convolutional Neural Networks for Brain Cancer Detection and Localisation 🧑 Francesco Mercaldo*, Luca Brunese, Fabio Martinelli, Antonella Santone and Mario Cesarelli 🏫 University of Molise, National Research Council of Italy, University of Sannio 🔎 #Brain cancer is widely recognised as one of the most aggressive types of tumors. In fact, approximately 70% of patients diagnosed with this malignant cancer do not survive. In this paper, we propose a method aimed to detect and localise brain cancer, starting from the analysis of magnetic resonance images. The proposed method exploits deep learning, in particular convolutional neural networks and class activation mapping, in order to provide explainability by highlighting the areas of the medical image related to brain cancer (from the model point of view). We evaluate the proposed method with 3000 magnetic resonances using a free available dataset. The results we obtained are encouraging. We reach an accuracy ranging from 97.83% to 99.67% in brain cancer detection by exploiting four different models: VGG16, ResNet50, Alex_Net, and MobileNet, thus showing the effectiveness of the proposed method. https://lnkd.in/gZzaViNu

1

We’re happy to share our new paper, recently published in SN Computer Science (Springer)! Gabriele Gianini Ernesto Damiani Corrado Mio Ghalia Tello, PhD In this work, we look at how to predict violations in business processes using deep learning but with a different starting point than most existing research. Most approaches assume that event data is already prepared at the “right” level of abstraction. In reality, that’s rarely the case. Organizations often deal with raw, low-level events — noisy, detailed, and not neatly structured for analysis. Our approach works directly with these low-level events and automatically adapts to the appropriate abstraction level. This means less manual preprocessing and a more realistic way of applying predictive monitoring in real-world settings. You can read the full paper here: https://lnkd.in/d_n-FTAe

40

National Repository Infrastructure and Open Access Challenges: The Croatian Perspective This paper presents the Croatian national repository infrastructure as a case study, offering insights for comparison with similar national infrastructures. The study employs a quantitative research approach, divided into two parts to provide a comprehensive overview of the current state and future development of repositories in Croatia. The first part analyses quantitative data and repository statistics. The DABAR infrastructure currently comprises 182 repositories and hosts over 249,000 digital objects, yet only slightly more than 50% of them are openly accessible. To investigate the reasons behind the high percentage of restricted or closed-access objects, a survey was conducted among institutions that primarily deposit such items. https://lnkd.in/e_AZQQtB

7

I’m thrilled to announce that our paper “Simple and Effective Specialized Representations for Fair Classifiers” (with Alberto Sinigaglia*, Davide Sartor*, Marina Ceccon) — from the AMCO Lab - UniPD — has been accepted at NeurIPS 2025! 🎉 (* = equal contribution) In this work, we introduce a fairness-aware representation learning approach that avoids adversarial training by leveraging Characteristic Function Distance (CFD) to remove sensitive information from learned representations. Across standard fairness benchmarks, our methods achieve strong accuracy while minimizing sensitive information leakage — all with a stable, efficient, and easy-to-integrate training procedure that requires no sensitive attributes at inference. Check out the project page for paper, code, and details: 🔗 https://lnkd.in/di_DFe96 Excited to present this at NeurIPS and discuss fairness in ML with the community!

76

3 commenti

Picuslab is thrilled to share their last publication entitled “Translating code with Large Language Models and human-in-the-loop feedback” – whose authors are Gabriele Dario De Siano, Anna Rita Fasolino, Giancarlo Sperlì, and Andrea Vignali – has been accepted in Elsevier Information and Software Technology journal. This paper investigates, from the human perspective, the use of three Generative AI tools (ChatGPT, Google Bard, and GitHub Copilot) in the context of translation tasks from code written in query languages to code written in framework-specific code languages, specifically focused on SQL dialects and PySpark. This translation is especially crucial during the migration from centralized architectures to cloud-based architectures. https://lnkd.in/dhaciiQq #LargeLanguageModel #HumancenteredAI #Codetranslation #Softwareengineering

17

Francesco L. De Faveri introducing our work on “DP-COMET: A Differential Privacy Contextual Obfuscation MEchanism for Texts in NLP”. Co-authored with Guglielmo Faggioli and Nicola Ferro, this research explores how to protect user privacy using contextual obfuscation methods in NLP and IR applications. Abstract: Protecting sensitive information within textual data strongly depends on the context in which the data is presented. However, current privacy-preserving obfuscation mechanisms based on epsilon-Differential Privacy (DP) produce an obfuscated private text, changing the original phrase term-by-term without considering the context in which such a term is placed. This paper introduces DP-COMET, an epsilon-DP obfuscation mechanism that evaluates a text's context before producing its private version. The mechanism defines a representation of the original text that considers the entire context within the text, producing an obfuscated version after adding noise to this representation and depending on the privacy parameter epsilon. We test DP-COMET on different Natural Language Processing (NLP) and Information Retrieval (IR) downstream tasks, and our findings show that our obfuscation mechanism not only achieves comparable performance results to traditional term-by-term mechanisms but also produces obfuscated texts less similar to the originals. To promote the reproducibility of DP-COMET, we make the code publicly available at https://lnkd.in/dcsAWeqx. https://lnkd.in/dvcYP5jp https://lnkd.in/dcsAWeqx

54

The latest special issue of the Italian Journal of Cognitive Sciences (Reti Saperi Linguaggi, published by Il Mulino) is out and is about cybernetics in the age of deep learning. It includes contributions by David Gunkel, Federico Cabitza, Francesco Parisi, and Yours truly - among others. This is my paper: Massimiliano Cappuccio "Cybernetics and the Skilled Mind. Pre-reflective Action Control in High Performance" pp: 271-299 | DOI: 10.12832/119461 Abstract: Is reflective awareness beneficial or rather detrimental to the control of expert action? Cybernetics, the study of self-regulating systems, can help us answer this question in the context of high performance. Each of the three historical waves of cybernetics contributes to explain why experts tend to underperform when they closely monitor their own activity as they perform – for example, when sportspeople, craftspeople, and artists are instructed to single out and track some component processes of their actions. First-wave cybernetics provides the foundations for understanding skilled behaviour as the fine-tuned regulation of perceptual variables through negative feedback and adaptive control. Second-wave cybernetics highlights how habitual patterns emerge from recursive, self-organising interactions between embodied observers and their environments. Third-wave cybernetics examines skilled habits within systems of practices and social forms of regulation, showing how agents maintain local autonomy within broader organisational and normative structures. I argue that all the three strands of cybernetics provide robust arguments in favour of "Radical Habitualism", the thesis that habitual, pre-reflective control is the fundamental enabler and the distinctive hallmark of optimal skilled performance and that reflective control is, in turn, inherently suboptimal. Corroborated by empirical research in sport & performance psychology and supported by the models of expert control developed by cyberneticists, the radically habitualist framework presented in this paper aims to offer a unified anti-representationalist and anti-intellectualist account of skill & expertise, reconciling ecological, enactivist, and pragmatist perspectives on action control. Keywords: Radical Habitualism; Cybernetics; Control Theory; High Performance; Sport and Performance Psychology; Philosophy of Skill and Expertise; Choking Effect; Dispositional Reinvestment Theory https://lnkd.in/g8FQcmQX

5

I’m glad to share that our paper "Multi-domain Validation of LLM-Based Simulators via Interpretable and Latent Representations" has been published in the Discovery Science 2025 proceedings (Springer Nature). This work, co-authored with Luca Coda-Giorgio and Giacomo Fidone, explores how interpretable and latent representations can support the validation of LLM-based simulators. https://lnkd.in/d8zGzvG2

23

> Sharing Resource < Nice reading from Google Scholar alert: "Quantum algorithms for equational reasoning" by Davide Rattacaso, Daniel Jaschke, Marco Ballarin, Ilaria Siloi, Simone Montangero Abstract: We introduce quantum normal form reduction, a quantum computational framework for analyzing abstract symbolic expressions - such as strings, algebraic formulas, or quantum circuits - that are equivalent under a given set of transformation rules. These rules form a term rewriting system, a formal method for deriving equivalences by repeatedly applying substitutions. We construct an efficiently implementable quantum Hamiltonian whose ground state encodes the entire class of equivalent expressions - potentially exponentially many - in a quantum superposition. By preparing and manipulating these ground states, we address fundamental problems in equational reasoning, including the word problem, i.e., determining whether two expressions are equivalent, counting the number of equivalent expressions, and identifying structural properties of equivalence classes. We demonstrate a quantum-inspired version of the algorithm using tensor network simulations by solving instances involving up to 10^28 equivalent expressions, well beyond the reach of standard classical graph exploration techniques. This framework opens the path for quantum symbolic computation in areas ranging from quantum and logical circuit design to data compression, computational group theory, linguistics, polymers and biomolecular modeling, enabling the investigation of problems previously out of reach. Link: https://lnkd.in/ekbRektB #quantumalgorithm #quantummachinelearning #research #paper

22

1 commento

#IJCAI2025 Mario Alviano, Università della Calabria, delivering their talk #4895 Integrating Answer Set Programming and Large Language Models for Enhanced Structured Representation of Complex Knowledge in Natural Language at the Knowledge Representation and Reasoning (4/4)

5

#ResearchSpotlight: Let's get to know the scientists in my FIS2 team for PENSO! Here is Dr. Salvatore Citraro, expert in network science and complex systems. He will work on #interpretable #AI techniques, powered by complex networks, for understanding how LLMs can persuade or convince humans. Watch his interview here: #research #fis2 #FondoItalianoperlaScienza #senior #postdoc #ai

9