Luca Romeo
@docenti.unimc.it
Tenure Track Assistant Professor Computer Science, at Department Economics and Law
University of Macerata
LUCA ROMEO received a Ph.D. degree in computer science from the Department of Information Engineering (DII), Università Politecnica delle Marche, in 2018. His Ph.D. thesis was on "applied machine learning for human motion analysis and affective computing". He is currently a Tenure Track Assistant Professor of Computer Science with University of Macerata | UniMC Department Economics and Law. He is also Adjunct Professor of Customer Intelligence & Big Data, at Luiss, Roma and he is affiliated with the Unit of Computational Statistics and Machine Learning, Fondazione Istituto Italiano di Tecnologia Genova. His research topics include the design of novel Machine learning algorithms for solving relevant challenges in different real-world domains.
RESEARCH, TEACHING, or OTHER INTERESTS
Artificial Intelligence
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Massimo Martini, Riccardo Rosati, Luca Romeo, and Adriano Mancini
Elsevier BV
Denis Bernovschi, Alex Giacomini, Riccardo Rosati, and Luca Romeo
Elsevier BV
Jonathan Montomoli, Maria Maddalena Bitondo, Marco Cascella, Emanuele Rezoagli, Luca Romeo, Valentina Bellini, Federico Semeraro, Emiliano Gamberini, Emanuele Frontoni, Vanni Agnoletti,et al.
Springer Science and Business Media LLC
AbstractThe integration of Clinical Decision Support Systems (CDSS) based on artificial intelligence (AI) in healthcare is groundbreaking evolution with enormous potential, but its development and ethical implementation, presents unique challenges, particularly in critical care, where physicians often deal with life-threating conditions requiring rapid actions and patients unable to participate in the decisional process. Moreover, development of AI-based CDSS is complex and should address different sources of bias, including data acquisition, health disparities, domain shifts during clinical use, and cognitive biases in decision-making. In this scenario algor-ethics is mandatory and emphasizes the integration of ‘Human-in-the-Loop’ and ‘Algorithmic Stewardship’ principles, and the benefits of advanced data engineering. The establishment of Clinical AI Departments (CAID) is necessary to lead AI innovation in healthcare, ensuring ethical integrity and human-centered development in this rapidly evolving field.
Riccardo Rosati, Luca Romeo, Víctor Manuel Vargas, Pedro Antonio Gutiérrez, Emanuele Frontoni, and César Hervás-Martínez
Institute of Electrical and Electronics Engineers (IEEE)
Real-world classification problems may disclose different hierarchical levels where the categories are displayed in an ordinal structure. However, no specific deep learning (DL) models simultaneously learn hierarchical and ordinal constraints while improving generalization performance. To fill this gap, we propose the introduction of two novel ordinal-hierarchical DL methodologies, namely, the hierarchical cumulative link model (HCLM) and hierarchical-ordinal binary decomposition (HOBD), which are able to model the ordinal structure within different hierarchical levels of the labels. In particular, we decompose the hierarchical-ordinal problem into local and global graph paths that may encode an ordinal constraint for each hierarchical level. Thus, we frame this problem as simultaneously minimizing global and local losses. Furthermore, the ordinal constraints are set by two approaches ordinal binary decomposition (OBD) and cumulative link model (CLM) within each global and local function. The effectiveness of the proposed approach is measured on four real-use case datasets concerning industrial, biomedical, computer vision, and financial domains. The extracted results demonstrate a statistically significant improvement to state-of-the-art nominal, ordinal, and hierarchical approaches.
Oleksandr Kuznetsov, Paolo Sernani, Luca Romeo, Emanuele Frontoni, and Adriano Mancini
Institute of Electrical and Electronics Engineers (IEEE)
As reliance on disruptive applications based on Artificial Intelligence (AI) and Blockchain grows, the need for secure and trustworthy solutions becomes ever more critical. Whereas much research has been conducted on AI and Blockchain, there is a shortage of comprehensive studies examining their integration from a security perspective. Hence, this survey addresses such a gap and provides insights for policymakers, researchers, and practitioners exploiting AI and Blockchain’s evolving integration. Specifically, this paper analyzes the potential benefits of the integration of AI and Blockchain as well as the related security concerns, identifying possible mitigation strategies, suggesting regulatory measures, and describing the impact it has on public trust.
Oleksandr Kuznetsov, Emanuele Frontoni, Luca Romeo, and Riccardo Rosati
Springer Science and Business Media LLC
Matilde Rocca, Lucia Maria Sacheli, Luca Romeo, and Andrea Cavallo
Springer Science and Business Media LLC
AbstractExtensive evidence shows that action observation can influence action execution, a phenomenon often referred to as visuo-motor interference. Little is known about whether this effect can be modulated by the type of interaction agents are involved in, as different studies show conflicting results. In the present study, we aimed at shedding light on this question by recording and analyzing the kinematic unfolding of reach-to-grasp movements performed in interactive and noninteractive settings. Using a machine learning approach, we investigated whether the extent of visuo-motor interference would be enhanced or reduced in two different joint action settings compared with a noninteractive one. Our results reveal that the detrimental effect of visuo-motor interference is reduced when the action performed by the partner is relevant to achieve a common goal, regardless of whether this goal requires to produce a concrete sensory outcome in the environment (joint outcome condition) or only a joint movement configuration (joint movement condition). These findings support the idea that during joint actions we form dyadic motor plans, in which both our own and our partner’s actions are represented in predictive terms and in light of the common goal to be achieved. The formation of a dyadic motor plan might allow agents to shift from the automatic simulation of an observed action to the active prediction of the consequences of a partner’s action. Overall, our results demonstrate the unavoidable impact of others’ action on our motor behavior in social contexts, and how strongly this effect can be modulated by task interactivity.
Michele Bernardini, Anastasiia Doinychko, Luca Romeo, Emanuele Frontoni, and Massih-Reza Amini
Elsevier BV
Víctor Manuel Vargas, Riccardo Rosati, César Hervás-Martínez, Adriano Mancini, Luca Romeo, and Pedro Antonio Gutiérrez
Elsevier BV
Alexandr Kuznetsov, Emanuele Frontoni, Luca Romeo, Nikolay Poluyanenko, Sergey Kandiy, Kateryna Kuznetsova, and Eleonóra Beňová
MDPI AG
Nonlinear substitutions or S-boxes are important cryptographic primitives of modern symmetric ciphers. They are designed to complicate the plaintext-ciphertext dependency. According to modern ideas, the S-box should be bijective, have high nonlinearity and algebraic immunity, low delta uniformity, and linear redundancy. These criteria directly affect the cryptographic strength of ciphers, providing resistance to statistical, linear, algebraic, differential, and other cryptanalysis techniques. Many researchers have used various heuristic search algorithms to generate random S-boxes with high nonlinearity; however, the complexity of this task is still high. For example, the best-known algorithm to generate a random 8-bit bijective S-box with nonlinearity 104 requires high computational effort—more than 65,000 intermediate estimates or search iterations. In this article, we explore a hill-climbing algorithm and optimize the heuristic search parameters. We show that the complexity of generating S-boxes can be significantly reduced. To search for a random bijective S-box with nonlinearity 104, only about 50,000 intermediate search iterations are required. In addition, we generate cryptographically strong S-Boxes for which additional criteria are provided. We present estimates of the complexity of the search and estimates of the probabilities of generating substitutions with various cryptographic indicators. The extracted results demonstrate a significant improvement in our approach compared to the state of the art in terms of providing linear non-redundancy, nonlinearity, algebraic immunity, and delta uniformity.
Víctor Manuel Vargas, Pedro Antonio Gutiérrez, Riccardo Rosati, Luca Romeo, Emanuele Frontoni, and César Hervás-Martínez
Elsevier BV
Alexandr Kuznetsov, Nicolas Luhanko, Emanuele Frontoni, Luca Romeo, and Riccardo Rosati
Springer Science and Business Media LLC
Luca Romeo, Temitayo Olugbade, Massimiliano Pontil, and Nadia Bianchi-Berthouze
Institute of Electrical and Electronics Engineers (IEEE)
Dominique Lepore, Emanuele Frontoni, Alessandra Micozzi, Sara Moccia, Luca Romeo, and Francesca Spigarelli
Elsevier BV
Riccardo Rosati, Luca Romeo, Víctor Manuel Vargas, Pedro Antonio Gutiérrez, César Hervás-Martínez, Lorenzo Bianchini, Alessandra Capriotti, Rosario Capparuccia, and Emanuele Frontoni
Springer Nature Switzerland
Víctor Manuel Vargas, Pedro Antonio Gutiérrez, Riccardo Rosati, Luca Romeo, Emanuele Frontoni, and César Hervás-Martínez
Elsevier BV
Riccardo Rosati, Luca Romeo, Gianalberto Cecchini, Flavio Tonetto, Paolo Viti, Adriano Mancini, and Emanuele Frontoni
Springer Science and Business Media LLC
AbstractThe Internet of Things (IoT), Big Data and Machine Learning (ML) may represent the foundations for implementing the concept of intelligent production, smart products, services, and predictive maintenance (PdM). The majority of the state-of-the-art ML approaches for PdM use different condition monitoring data (e.g. vibrations, currents, temperature, etc.) and run to failure data for predicting the Remaining Useful Lifetime of components. However, the annotation of the component wear is not always easily identifiable, thus leading to the open issue of obtaining quality labeled data and interpreting it. This paper aims to introduce and test a Decision Support System (DSS) for solving a PdM task by overcoming the above-mentioned challenge while focusing on a real industrial use case, which includes advanced processing and measuring machines. In particular, the proposed DSS is comprised of the following cornerstones: data collection, feature extraction, predictive model, cloud storage, and data analysis. Differently from the related literature, our novel approach is based on a feature extraction strategy and ML prediction model powered by specific topics collected on the lower and upper levels of the production system. Compared with respect to other state-of-the-art ML models, the experimental results demonstrated how our approach is the best trade-off between predictive performance (MAE: 0.089, MSE: 0.018, $$R^{2}: 0.868$$ R 2 : 0.868 ), computation effort (average latency of 2.353 s for learning from 400 new samples), and interpretability for the prediction of processing quality. These peculiarities, together with the integration of our ML approach into the proposed cloud-based architecture, allow the optimization of the machining quality processes by directly supporting the maintainer/operator. These advantages may impact to the optimization of maintenance schedules and to get real-time warnings about operational risks by enabling manufacturers to reduce service costs by maximizing uptime and improving productivity.
Giacomo Turri, Andrea Cavallo, Luca Romeo, Massimiliano Pontil, Alan Sanfey, Stefano Panzeri, and Cristina Becchio
Elsevier BV
Antonio Nicolucci, Luca Romeo, Michele Bernardini, Marco Vespasiani, Maria Chiara Rossi, Massimiliano Petrelli, Antonio Ceriello, Paolo Di Bartolo, Emanuele Frontoni, and Giacomo Vespasiani
Elsevier BV
Riccardo Rosati, Luca Romeo, Víctor Manuel Vargas, Pedro Antonio Gutiérrez, César Hervás-Martínez, and Emanuele Frontoni
Springer Science and Business Media LLC
AbstractNowadays, decision support systems (DSSs) are widely used in several application domains, from industrial to healthcare and medicine fields. Concerning the industrial scenario, we propose a DSS oriented to the aesthetic quality control (AQC) task, which has quickly established itself as one of the most crucial challenges of Industry 4.0. Taking into account the increasing amount of data in this domain, the application of machine learning (ML) and deep learning (DL) techniques offers great opportunities to automatize the overall AQC process. State-of-the-art is mainly oriented to approach this problem with a nominal DL classification method which does not exploit the ordinal structure of the AQC task, thus not penalizing the error among distant AQC classes (which is a relevant aspect for the real use case). The paper introduces a DL ordinal methodology for the AQC classification. Differently from other deep ordinal methods, we combined the standard categorical cross-entropy with the cumulative link model and we imposed the ordinal constraint via the thresholds and slope parameters. Experimental results were performed for solving an AQC task on a novel image dataset originated from a specific company’s demand (i.e., aesthetic assessment of wooden stocks). We demonstrated how the proposed methodology is able to reduce misclassification errors (up to 0.937 quadratic weight kappa loss) among distant classes while overcoming other state-of-the-art deep ordinal models and reducing the bias factor related to the item geometry. The proposed DL approach was integrated as the main core of a DSS supported by Internet of Things (IoT) architecture that can support the human operator by reducing up to 90% the time needed for the qualitative analysis carried out manually in this specific domain.
Francesco Ferracuti, Alessandro Freddi, Andrea Monteriu, and Luca Romeo
Institute of Electrical and Electronics Engineers (IEEE)
This article presents a fault diagnosis algorithm for rotating machinery based on the Wasserstein distance. Recently, the Wasserstein distance has been proposed as a new research direction to find better distribution mapping when compared with other popular statistical distances and divergences. In this work, first, frequency- and time-based features are extracted by vibration signals, and second, the Wasserstein distance is considered for the learning phase to discriminate the different machine operating conditions. Specifically, the 1-D Wasserstein distance is considered due to its low computational burden because it can be evaluated directly by the order statistics of the extracted features. Furthermore, a distance weighting stage based on neighborhood component features selection (NCFS) is exploited to achieve robust fault diagnosis at low signal-to-noise ratio (SNR) conditions and with high-dimensional features. In detail, the NCFS framework is here adapted to weight 1-D Wasserstein distances evaluated from time/frequency features. Experiments are conducted on two benchmark data sets to verify the effectiveness of the proposed fault diagnosis method at different SNR conditions. The comparison with state-of-the-art fault diagnosis algorithms shows promising results. Note to Practitioners—This article was motivated by the problem of fault diagnosis of rotating machinery under low SNR and different machine operating conditions. The algorithm employs a statistical distance-based fault diagnosis technique, which permits to obtain an estimation of the fault signature without the need for training a classifier. The algorithm is computationally efficient during the training and testing stages, and thus, it can be used in embedded hardware. Finally, the proposed methodology can be applied to other application domains such as system monitoring and prognostics, which can help to schedule the maintenance of rotating machinery.
Aleksandr Kuznetsov, Davyd Kvaratskheliia, Andrea Maranesi, Luca Romeo, Alessandro Muscatello, and Riccardo Rosati
IEEE
Modern systems of face recognition (FRS) are used in a wide range of computer applications: for user authentication; in email marketing; in social networks; for personal identification and more. However, such technologies are often vulnerable to spoofing attacks. Facial image can be faked in various ways: print a photo; record a video; create a high-quality silicone mask, etc. By presenting a fake to the FRS the attacker has an intention of passing himself off as another person, for instance, trying to get an access to a secure computer system. Face Liveliness Detection solves this problem by detecting whether the person in front of the camera is real or fake. In this article, we explore the possibilities of using deep learning technology for face liveliness detection. We consider several models and setting numerous experiments. As datasets for experiments we use various fake images and for each such dataset we obtained evaluation of effectiveness. In our research, our main goal is to improve basic deep learning architectures using latest technologies to get more accurate model.
Alexandr Kuznetsov, Nicolas Luhanko, Emanuele Frontoni, Luca Romeo, and Riccardo Rosati
IEEE
Various cryptographic and steganographic techniques are used to hide digital information during its processing, storage, and transmission. While cryptography hides the information content of digital data (by converting them into a meaningless set of noise-like sequences), steganography hides the very existence of information messages. In other words, steganographic techniques hide digital messages by embedding them in so-called. containers. Containers are other digital data or physical objects. To do this, containers (covers, media) must be highly redundant data. Revealing the fact of steganographic hiding and detecting an embedded message is usually extremely difficult. In fact, hidden messages are some noise added to the container, and we must, based on the study of this noise, decide on the presence or absence of an embedded message. In this article, we consider deep learning methods for steganoanalysis of digital cover images. We have considered several deep learning models and conduct numerous tests on various datasets. Our experiments show that deep learning does indeed make it possible to design effective stego-detectors, but this requires fine-tuning of model hyperparameters and optimization of the neural network architecture.
Alexandr Kuznetsov, Dmytro Zakharov, Emanuele Frontoni, Luca Romeo, and Riccardo Rosati
IEEE
Biometric techniques have traditionally been used in various cybersecurity applications. For instance, some user authentication systems use biometric facial images, fingerprints, iris images, vein patterns, and much more. Most of these applications store biometric features copies (or data derived from these features). Authentication is carried out based on the results of comparing the presented biometric images with the reference ones. However, if the storage is compromised, biometric personal data will be lost and this significantly limits the scope of biometric techniques. Fuzzy extractors solve this problem. Instead of reference biometric data, fuzzy extractors extract cryptographically strong keys (secret bit strings, passwords) that are used to authenticate users. In addition, the extracted keys can be used as a source of entropy for various cryptographic mechanisms (encryption, electronic signature, etc.). In this paper, we propose a fuzzy extractor for generating cryptographically strong keys from biometric images of a human face. Our extractor uses biometric image preprocessing using deep learning methods, as well as code-based cryptosystems that provide a post-quantum (quantum-resistant) level of security.