Research

We have proposed supervised learning algorithms for the induction of the membership function to fuzzy and shadowed sets [Malchiodi and Pedrycz, 2013; Cermenati et al., 2020; Malchiodi and Zanaboni, 2019], which we have applied in the fields of semantic Web [Malchiodi and Tettamanzi, 2018] and bioinformatics [Frasca and Malchiodi, 2017; Frasca and Malchiodi, 2016].

We focused on the analysis of the excessive demand of energy resources by cutting-edge deep neural networks (DNNs) in various application domains [Marinò et al., 2023a], and the subsequent need to shrink them via lossy compression techniques and lossless compact representation formats [Marinò et al., 2023; Marinò et al., 2021]. We applied such techniques in order to reduce the demand of neural models for cancer detection [Gliozzo et al., 2024].

We contributed to the development of the recently introduced field of learned data structures, by investigating the role of classifiers [Malchiodi et al., 2023; Fumagalli et al., 2022] and data complexity in learned Bloom filters [Malchiodi et al., 2024] and proposing specific structures for string indexing [Ferragina et al., 2023].

We apply machine learning methods in engineering contexts, with special focus to predictive maintenance. Specifically, we have used interpretable methods for detecting and localizing quenches in superconductors [Biagiotti et al., 2025] and developed federated learning protocols for anomaly detection using kernel methods [Frasson and Malchiodi, 2024].

We have investigated learned approaches for processing knowledge bases in the fields of semantic web [Malchiodi et al., 2020; Malchiodi and Tettamanzi, 2018; Malchiodi et al., 2018a] and bioinformatics [Cavalleri et al., 2024; Frasca and Malchiodi, 2017].

We developmed parametric Hopfield networks for single- and multi-task classification purposes, with specific attention to scenarios characterized by high label unbalancing [Frasca et al., 2020; Frasca et al., 2013].

Machine learning and data science are powerful tools in the key phases of modern scientific investigations. We provide our expertise in intelligent data analysis to several research groups grounded in the following fields: cultural heritage (classification of archæological remains [Malchiodi et al., 2025; Ruschioni et al., 2023; Zanaboni et al., 2022]), forensics (estimation of fatal fall heights [Blandino et al., 2024] and prediction of the type of vehicle involved in a pedestrian hit [Casali et al., 2021]), veterinary medicine (evaluation of cardiovascular factors [Bagardi et al., 2021; Galizzi et al., 2021]), medicine (support to image-based diagnosis [Gliozzo et al., 2024; Casiraghi et al., 2020; Esposito et al., 2021]) and bioinformatics (LLMs for enzyme prediciton [Nicolini et al., 2025], protein generation [Nicolini et al., 2024; Valentini et al., 2023] and RNA processing [Cavalleri et al., 2024]).

We focus on identifying temporal changes in the distribution of data used to query trained support-vector-based models in the context of federated learning [Frasson and Malchiodi, 2024], in order to promptly react when this distribution significantly deviates from the training data, leading to performance degradation.