Publications
OUR RESEARCH
Scientific Publications
Here you can find the comprehensive list of publications from the members of the Research Center on Computer Vision and eXtended Reality (xRAI).
Use the tag cloud to filter papers based on specific research topics, or use the menus to filter by year, type of publication, or authors.
For each paper, you have the option to view additional details such as the Abstract, Links, and BibTex record.
Research is formalized curiosity. It is poking and prying with a purpose
Zora Neale Hurston
2026
Mondal, Semanto; Ferraro, Antonino; Pecorelli, Fabiano; Iammarino, Martina; Pietro, Giuseppe De
Concept and rule guided neural network for early crop leaf nutrient deficiency diagnosis Journal Article
In: Computers and Electronics in Agriculture, vol. 248, pp. 111735, 2026, ISSN: 01681699.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Deep Learning, Neurosymbolic
@article{mondal_concept_2026,
title = {Concept and rule guided neural network for early crop leaf nutrient deficiency diagnosis},
author = {Semanto Mondal and Antonino Ferraro and Fabiano Pecorelli and Martina Iammarino and Giuseppe De Pietro},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0168169926003303},
doi = {10.1016/j.compag.2026.111735},
issn = {01681699},
year = {2026},
date = {2026-07-01},
urldate = {2026-06-16},
journal = {Computers and Electronics in Agriculture},
volume = {248},
pages = {111735},
abstract = {Crop nutrition deficiency poses a major challenge to achieving optimal yield, particularly in smallholder farming systems where timely expert diagnosis is limited. Early detection is crucial to minimize losses and reduce unnecessary fertilizer or pesticide usage. While deep learning offers potential for automated visual diagnosis, most existing approaches operate as black boxes and lack interpretability, explainability, or actionable recommendations. In this work, we present a neurosymbolic framework for early nutrient deficiency detection in ash gourd leaves using the EarlyNSD dataset. Our approach integrates a ResNet-50 backbone with a dual-head design: a classification head for deficiency prediction and a concept-prediction head that quantifies physiologically meaningful visual patterns such as yellowing, edge discoloration, spots, and vein greenness. These concept scores are combined with predefined domain rules to guide the learning of the neural component and to generate transparent, human-aligned explanations for each diagnosis. Building on the model outputs, we incorporate a Retrieval Augmented Generation (RAG)-based pipeline along with an agricultural knowledge base to generate targeted recommendations. This approach overcomes key shortcomings of pure neural models by incorporating domain knowledge in the form of differentiable fuzzy logic rules. The study demonstrates that the proposed framework improves both classification performance and interpretability compared to standard ResNet baselines. Grad-CAM analysis demonstrates that concept-guided attention aligns with symptom-specific regions, such as yellowed areas for Nitrogen deficiency or marginal discoloration for Potassium deficiency, providing visual validation of the reasoning process. Since EarlyNSD is limited in scale and visual diversity, the results are not directly comparable to large open-field datasets. Overall, our results establish a proof of concept for integrating neural detection with symbolic reasoning, enabling interpretable, actionable, and domain-informed nutrient management for practical applications.},
keywords = {Artificial Intelligence, Deep Learning, Neurosymbolic},
pubstate = {published},
tppubtype = {article}
}
Crop nutrition deficiency poses a major challenge to achieving optimal yield, particularly in smallholder farming systems where timely expert diagnosis is limited. Early detection is crucial to minimize losses and reduce unnecessary fertilizer or pesticide usage. While deep learning offers potential for automated visual diagnosis, most existing approaches operate as black boxes and lack interpretability, explainability, or actionable recommendations. In this work, we present a neurosymbolic framework for early nutrient deficiency detection in ash gourd leaves using the EarlyNSD dataset. Our approach integrates a ResNet-50 backbone with a dual-head design: a classification head for deficiency prediction and a concept-prediction head that quantifies physiologically meaningful visual patterns such as yellowing, edge discoloration, spots, and vein greenness. These concept scores are combined with predefined domain rules to guide the learning of the neural component and to generate transparent, human-aligned explanations for each diagnosis. Building on the model outputs, we incorporate a Retrieval Augmented Generation (RAG)-based pipeline along with an agricultural knowledge base to generate targeted recommendations. This approach overcomes key shortcomings of pure neural models by incorporating domain knowledge in the form of differentiable fuzzy logic rules. The study demonstrates that the proposed framework improves both classification performance and interpretability compared to standard ResNet baselines. Grad-CAM analysis demonstrates that concept-guided attention aligns with symptom-specific regions, such as yellowed areas for Nitrogen deficiency or marginal discoloration for Potassium deficiency, providing visual validation of the reasoning process. Since EarlyNSD is limited in scale and visual diversity, the results are not directly comparable to large open-field datasets. Overall, our results establish a proof of concept for integrating neural detection with symbolic reasoning, enabling interpretable, actionable, and domain-informed nutrient management for practical applications.
Gallo, Luigi; Carruba, Maria Concetta; Ferraro, Antonino; Lund, Henrik Hautop; Rega, Angelo; Triberti, Stefano
Editorial: AI innovations in education: adaptive learning and beyond Journal Article
In: Frontiers in Computer Science, vol. 8, pp. 1822456, 2026, ISSN: 2624-9898.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Education
@article{gallo_editorial_2026,
title = {Editorial: AI innovations in education: adaptive learning and beyond},
author = {Luigi Gallo and Maria Concetta Carruba and Antonino Ferraro and Henrik Hautop Lund and Angelo Rega and Stefano Triberti},
url = {https://www.frontiersin.org/articles/10.3389/fcomp.2026.1822456/full},
doi = {10.3389/fcomp.2026.1822456},
issn = {2624-9898},
year = {2026},
date = {2026-03-01},
urldate = {2026-04-03},
journal = {Frontiers in Computer Science},
volume = {8},
pages = {1822456},
publisher = {Frontiers Media SA},
abstract = {Artificial Intelligence (AI) is gradually transforming educational practices. AI-powered teaching assistants, large language models, and multimodal analytics platforms are reshaping how learning experiences are designed and assessed. However, AI integration is not merely a technological matter: it is also heavily influenced by pedagogical, psychological, and sociocultural factors. Beyond technical implementation, AI systems can be framed within a human augmentation perspective, where technologies enhance sensory, motor, and cognitive processes in hybrid environments (Augello et al., 2022), including immersive contexts in which presence and cognition dynamically interact (Palombi et al., 2023). At the same time, advances in adaptive and data-driven AI, including explainable and diversity-aware approaches, highlight the role of algorithmic choices in shaping users' experiences and behaviors (Ferraro et al., 2025). Recent studies further show that, while educators are using AI tools for instructional purposes (e.g., creating teaching materials), concerns remain about the risk of unfair AI use and the difficulty of detecting it (Amato et al., 2023; Carruba et al., 2025). Accordingly, research on AI in education is becoming increasingly focused on factors that support implementation and adoption in real-life contexts, beyond mere improvement of algorithms from a computer science perspective (Triberti et al., 2024; Acosta-Enriquez et al., 2025; Galindo-Domĺnguez et al., 2024).
Against this backdrop, this Research Topic (RT), which spans four Frontiers journals, focuses on empirical and theoretical aspects of personalized and adaptive learning. More specifically, it examines the role of AI in fostering inclusive, data-driven, and emotionally responsive educational ecosystems, with particular attention to motivation, beliefs, creativity, self-regulation, and ethics.
For analytical clarity, the contributions are organized into six interrelated thematic areas: AI Adoption, Acceptance, and Self-Regulation; Teacher AI Literacy and Sustainable Integration; Adaptive, Immersive, and AI-Enhanced Learning Environments; Multimodal Analytics, Assessment, and Predictive AI; Learner Psychological, Cognitive, and Sociocultural Factors; Conceptual, Ethical, and Human-AI Synergy Perspectives. These areas reflect three broader dimensions of current research: the adoption of AI by learners and teachers, the design of intelligent learning environments, and the broader psychological and ethical implications of AI-supported education.},
keywords = {Artificial Intelligence, Education},
pubstate = {published},
tppubtype = {article}
}
Artificial Intelligence (AI) is gradually transforming educational practices. AI-powered teaching assistants, large language models, and multimodal analytics platforms are reshaping how learning experiences are designed and assessed. However, AI integration is not merely a technological matter: it is also heavily influenced by pedagogical, psychological, and sociocultural factors. Beyond technical implementation, AI systems can be framed within a human augmentation perspective, where technologies enhance sensory, motor, and cognitive processes in hybrid environments (Augello et al., 2022), including immersive contexts in which presence and cognition dynamically interact (Palombi et al., 2023). At the same time, advances in adaptive and data-driven AI, including explainable and diversity-aware approaches, highlight the role of algorithmic choices in shaping users' experiences and behaviors (Ferraro et al., 2025). Recent studies further show that, while educators are using AI tools for instructional purposes (e.g., creating teaching materials), concerns remain about the risk of unfair AI use and the difficulty of detecting it (Amato et al., 2023; Carruba et al., 2025). Accordingly, research on AI in education is becoming increasingly focused on factors that support implementation and adoption in real-life contexts, beyond mere improvement of algorithms from a computer science perspective (Triberti et al., 2024; Acosta-Enriquez et al., 2025; Galindo-Domĺnguez et al., 2024).
Against this backdrop, this Research Topic (RT), which spans four Frontiers journals, focuses on empirical and theoretical aspects of personalized and adaptive learning. More specifically, it examines the role of AI in fostering inclusive, data-driven, and emotionally responsive educational ecosystems, with particular attention to motivation, beliefs, creativity, self-regulation, and ethics.
For analytical clarity, the contributions are organized into six interrelated thematic areas: AI Adoption, Acceptance, and Self-Regulation; Teacher AI Literacy and Sustainable Integration; Adaptive, Immersive, and AI-Enhanced Learning Environments; Multimodal Analytics, Assessment, and Predictive AI; Learner Psychological, Cognitive, and Sociocultural Factors; Conceptual, Ethical, and Human-AI Synergy Perspectives. These areas reflect three broader dimensions of current research: the adoption of AI by learners and teachers, the design of intelligent learning environments, and the broader psychological and ethical implications of AI-supported education.
Against this backdrop, this Research Topic (RT), which spans four Frontiers journals, focuses on empirical and theoretical aspects of personalized and adaptive learning. More specifically, it examines the role of AI in fostering inclusive, data-driven, and emotionally responsive educational ecosystems, with particular attention to motivation, beliefs, creativity, self-regulation, and ethics.
For analytical clarity, the contributions are organized into six interrelated thematic areas: AI Adoption, Acceptance, and Self-Regulation; Teacher AI Literacy and Sustainable Integration; Adaptive, Immersive, and AI-Enhanced Learning Environments; Multimodal Analytics, Assessment, and Predictive AI; Learner Psychological, Cognitive, and Sociocultural Factors; Conceptual, Ethical, and Human-AI Synergy Perspectives. These areas reflect three broader dimensions of current research: the adoption of AI by learners and teachers, the design of intelligent learning environments, and the broader psychological and ethical implications of AI-supported education.
2025
Mondal, Semanto; Ferraro, Antonino; Pecorelli, Fabiano; Pietro, Giuseppe De
A Logic Tensor Network-Based Neurosymbolic Framework for Explainable Diabetes Prediction Journal Article
In: Applied Sciences, vol. 15, no. 21, pp. 11806, 2025, ISSN: 2076-3417.
Abstract | Links | BibTeX | Tags:
@article{mondal_logic_2025-1,
title = {A Logic Tensor Network-Based Neurosymbolic Framework for Explainable Diabetes Prediction},
author = {Semanto Mondal and Antonino Ferraro and Fabiano Pecorelli and Giuseppe De Pietro},
url = {https://www.mdpi.com/2076-3417/15/21/11806},
doi = {10.3390/app152111806},
issn = {2076-3417},
year = {2025},
date = {2025-11-01},
urldate = {2025-11-06},
journal = {Applied Sciences},
volume = {15},
number = {21},
pages = {11806},
abstract = {Neurosymbolic AI is an emerging paradigm that combines neural network learning capabilities with the structured reasoning capacity of symbolic systems. Although machine learning has achieved cutting-edge outcomes in diverse fields, including healthcare, agriculture, and environmental science, it has potential limitations. Machine learning and neural models excel at identifying intricate data patterns, yet they often lack transparency, depend on large labelled datasets, and face challenges with logical reasoning and tasks that require explainability. These challenges reduce their reliability in high-stakes applications such as healthcare. To address these limitations, we propose a hybrid framework that integrates symbolic knowledge expressed in First-Order Logic into neural learning via a Logic Tensor Network (LTN). In this framework, expert-defined medical rules are embedded as logical axioms with learnable thresholds. As a result, the model gains predictive power, interpretability, and explainability through reasoning over the logical rules. We have utilized this neurosymbolic method for predicting diabetes by employing the Pima Indians Diabetes Dataset. Our experimental setup evaluates the LTN-based model against several conventional methods, including Support Vector Machines (SVM), Logistic Regression (LR), K-Nearest Neighbors (K-NN), Random Forest Classifiers (RF), Naive Bayes (NB), and a Standalone Neural Network (NN). The findings demonstrate that the neurosymbolic framework not only surpasses traditional models in predictive accuracy but also offers improved explainability and robustness. Notably, the LTN-based neurosymbolic framework achieves an excellent balance between recall and precision, along with a higher AUC-ROC score. These results underscore its potential for trustworthy medical diagnostics. This work highlights how integrating symbolic reasoning with data-driven models can bridge the gap between explainability, interpretability, and performance, offering a promising direction for AI systems in domains where both accuracy and explainability are critical.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neurosymbolic AI is an emerging paradigm that combines neural network learning capabilities with the structured reasoning capacity of symbolic systems. Although machine learning has achieved cutting-edge outcomes in diverse fields, including healthcare, agriculture, and environmental science, it has potential limitations. Machine learning and neural models excel at identifying intricate data patterns, yet they often lack transparency, depend on large labelled datasets, and face challenges with logical reasoning and tasks that require explainability. These challenges reduce their reliability in high-stakes applications such as healthcare. To address these limitations, we propose a hybrid framework that integrates symbolic knowledge expressed in First-Order Logic into neural learning via a Logic Tensor Network (LTN). In this framework, expert-defined medical rules are embedded as logical axioms with learnable thresholds. As a result, the model gains predictive power, interpretability, and explainability through reasoning over the logical rules. We have utilized this neurosymbolic method for predicting diabetes by employing the Pima Indians Diabetes Dataset. Our experimental setup evaluates the LTN-based model against several conventional methods, including Support Vector Machines (SVM), Logistic Regression (LR), K-Nearest Neighbors (K-NN), Random Forest Classifiers (RF), Naive Bayes (NB), and a Standalone Neural Network (NN). The findings demonstrate that the neurosymbolic framework not only surpasses traditional models in predictive accuracy but also offers improved explainability and robustness. Notably, the LTN-based neurosymbolic framework achieves an excellent balance between recall and precision, along with a higher AUC-ROC score. These results underscore its potential for trustworthy medical diagnostics. This work highlights how integrating symbolic reasoning with data-driven models can bridge the gap between explainability, interpretability, and performance, offering a promising direction for AI systems in domains where both accuracy and explainability are critical.
Mondal, Semanto; Ferraro, Antonino; Pecorelli, Fabiano; Pietro, Giuseppe De
A Logic Tensor Network-Based Neurosymbolic Framework for Explainable Diabetes Prediction Journal Article
In: Applied Sciences, vol. 15, no. 21, pp. 11806, 2025, ISSN: 2076-3417.
Abstract | Links | BibTeX | Tags:
@article{mondal_logic_2025,
title = {A Logic Tensor Network-Based Neurosymbolic Framework for Explainable Diabetes Prediction},
author = {Semanto Mondal and Antonino Ferraro and Fabiano Pecorelli and Giuseppe De Pietro},
url = {https://www.mdpi.com/2076-3417/15/21/11806},
doi = {10.3390/app152111806},
issn = {2076-3417},
year = {2025},
date = {2025-11-01},
urldate = {2025-11-25},
journal = {Applied Sciences},
volume = {15},
number = {21},
pages = {11806},
abstract = {Neurosymbolic AI is an emerging paradigm that combines neural network learning capabilities with the structured reasoning capacity of symbolic systems. Although machine learning has achieved cutting-edge outcomes in diverse fields, including healthcare, agriculture, and environmental science, it has potential limitations. Machine learning and neural models excel at identifying intricate data patterns, yet they often lack transparency, depend on large labelled datasets, and face challenges with logical reasoning and tasks that require explainability. These challenges reduce their reliability in high-stakes applications such as healthcare. To address these limitations, we propose a hybrid framework that integrates symbolic knowledge expressed in First-Order Logic into neural learning via a Logic Tensor Network (LTN). In this framework, expert-defined medical rules are embedded as logical axioms with learnable thresholds. As a result, the model gains predictive power, interpretability, and explainability through reasoning over the logical rules. We have utilized this neurosymbolic method for predicting diabetes by employing the Pima Indians Diabetes Dataset. Our experimental setup evaluates the LTN-based model against several conventional methods, including Support Vector Machines (SVM), Logistic Regression (LR), K-Nearest Neighbors (K-NN), Random Forest Classifiers (RF), Naive Bayes (NB), and a Standalone Neural Network (NN). The findings demonstrate that the neurosymbolic framework not only surpasses traditional models in predictive accuracy but also offers improved explainability and robustness. Notably, the LTN-based neurosymbolic framework achieves an excellent balance between recall and precision, along with a higher AUC-ROC score. These results underscore its potential for trustworthy medical diagnostics. This work highlights how integrating symbolic reasoning with data-driven models can bridge the gap between explainability, interpretability, and performance, offering a promising direction for AI systems in domains where both accuracy and explainability are critical.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Neurosymbolic AI is an emerging paradigm that combines neural network learning capabilities with the structured reasoning capacity of symbolic systems. Although machine learning has achieved cutting-edge outcomes in diverse fields, including healthcare, agriculture, and environmental science, it has potential limitations. Machine learning and neural models excel at identifying intricate data patterns, yet they often lack transparency, depend on large labelled datasets, and face challenges with logical reasoning and tasks that require explainability. These challenges reduce their reliability in high-stakes applications such as healthcare. To address these limitations, we propose a hybrid framework that integrates symbolic knowledge expressed in First-Order Logic into neural learning via a Logic Tensor Network (LTN). In this framework, expert-defined medical rules are embedded as logical axioms with learnable thresholds. As a result, the model gains predictive power, interpretability, and explainability through reasoning over the logical rules. We have utilized this neurosymbolic method for predicting diabetes by employing the Pima Indians Diabetes Dataset. Our experimental setup evaluates the LTN-based model against several conventional methods, including Support Vector Machines (SVM), Logistic Regression (LR), K-Nearest Neighbors (K-NN), Random Forest Classifiers (RF), Naive Bayes (NB), and a Standalone Neural Network (NN). The findings demonstrate that the neurosymbolic framework not only surpasses traditional models in predictive accuracy but also offers improved explainability and robustness. Notably, the LTN-based neurosymbolic framework achieves an excellent balance between recall and precision, along with a higher AUC-ROC score. These results underscore its potential for trustworthy medical diagnostics. This work highlights how integrating symbolic reasoning with data-driven models can bridge the gap between explainability, interpretability, and performance, offering a promising direction for AI systems in domains where both accuracy and explainability are critical.
Ferraro, Antonino; Orlando, Gian Marco; Russo, Diego
Generative Agent-Based Modeling with Large Language Models for insider threat detection Journal Article
In: Engineering Applications of Artificial Intelligence, vol. 157, pp. 111343, 2025, ISSN: 09521976.
Links | BibTeX | Tags: Cybersecurity, Generative Agent-Based Modeling, Insider threat detection, Large Language Models, Multi-Agent Systems
@article{ferraro_generative_2025,
title = {Generative Agent-Based Modeling with Large Language Models for insider threat detection},
author = {Antonino Ferraro and Gian Marco Orlando and Diego Russo},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0952197625013454},
doi = {10.1016/j.engappai.2025.111343},
issn = {09521976},
year = {2025},
date = {2025-10-01},
urldate = {2025-09-30},
journal = {Engineering Applications of Artificial Intelligence},
volume = {157},
pages = {111343},
keywords = {Cybersecurity, Generative Agent-Based Modeling, Insider threat detection, Large Language Models, Multi-Agent Systems},
pubstate = {published},
tppubtype = {article}
}
Ferraro, Antonino; Galli, Antonio; Gatta, Valerio La; Postiglione, Marco; Russo, Diego; Orlando, Gian Marco; Riccio, Giuseppe; Romano, Antonio; Moscato, Vincenzo
From Explanation to Exploration: Promoting DivErsity in Recommendation Systems Book Section
In: Boratto, Ludovico; Filippo, Allegra De; Lex, Elisabeth; Ricci, Francesco (Ed.): Recommender Systems for Sustainability and Social Good, vol. 2470, pp. 135–150, Springer Nature Switzerland, Cham, 2025, ISBN: 978-3-031-87653-0 978-3-031-87654-7, (Series Title: Communications in Computer and Information Science).
Links | BibTeX | Tags: Diversity, Explanation-Driven, Rabbit Hole Effect, Recommendation Diversity
@incollection{boratto_explanation_2025,
title = {From Explanation to Exploration: Promoting DivErsity in Recommendation Systems},
author = {Antonino Ferraro and Antonio Galli and Valerio La Gatta and Marco Postiglione and Diego Russo and Gian Marco Orlando and Giuseppe Riccio and Antonio Romano and Vincenzo Moscato},
editor = {Ludovico Boratto and Allegra De Filippo and Elisabeth Lex and Francesco Ricci},
url = {https://link.springer.com/10.1007/978-3-031-87654-7_13},
doi = {10.1007/978-3-031-87654-7_13},
isbn = {978-3-031-87653-0 978-3-031-87654-7},
year = {2025},
date = {2025-01-01},
urldate = {2025-09-30},
booktitle = {Recommender Systems for Sustainability and Social Good},
volume = {2470},
pages = {135–150},
publisher = {Springer Nature Switzerland},
address = {Cham},
note = {Series Title: Communications in Computer and Information Science},
keywords = {Diversity, Explanation-Driven, Rabbit Hole Effect, Recommendation Diversity},
pubstate = {published},
tppubtype = {incollection}
}
Ferraro, Antonino; Galli, Antonio; Gatta, Valerio La; Postiglione, Marco; Orlando, Gian Marco; Russo, Diego; Riccio, Giuseppe; Romano, Antonio; Moscato, Vincenzo
Agent-Based Modelling Meets Generative AI in Social Network Simulations Book Section
In: Aiello, Luca Maria; Chakraborty, Tanmoy; Gaito, Sabrina (Ed.): Social Networks Analysis and Mining, vol. 15211, pp. 155–170, Springer Nature Switzerland, Cham, 2025, ISBN: 978-3-031-78540-5 978-3-031-78541-2, (Series Title: Lecture Notes in Computer Science).
Links | BibTeX | Tags: Agent-Based Modelling, Generative Artificial Intelligence, Social media simulation
@incollection{aiello_agent-based_2025,
title = {Agent-Based Modelling Meets Generative AI in Social Network Simulations},
author = {Antonino Ferraro and Antonio Galli and Valerio La Gatta and Marco Postiglione and Gian Marco Orlando and Diego Russo and Giuseppe Riccio and Antonio Romano and Vincenzo Moscato},
editor = {Luca Maria Aiello and Tanmoy Chakraborty and Sabrina Gaito},
url = {https://link.springer.com/10.1007/978-3-031-78541-2_10},
doi = {10.1007/978-3-031-78541-2_10},
isbn = {978-3-031-78540-5 978-3-031-78541-2},
year = {2025},
date = {2025-01-01},
urldate = {2025-09-30},
booktitle = {Social Networks Analysis and Mining},
volume = {15211},
pages = {155–170},
publisher = {Springer Nature Switzerland},
address = {Cham},
note = {Series Title: Lecture Notes in Computer Science},
keywords = {Agent-Based Modelling, Generative Artificial Intelligence, Social media simulation},
pubstate = {published},
tppubtype = {incollection}
}
2024
Ferraro, Antonino; Sperlì, Giancarlo
How does user-generated content on Social Media affect stock predictions? A case study on GameStop Journal Article
In: Online Social Networks and Media, vol. 43-44, pp. 100293, 2024, ISSN: 24686964.
Links | BibTeX | Tags: Deep Learning, Financial systems, GameStop, Online social networks, Stock forecasting
@article{ferraro_how_2024,
title = {How does user-generated content on Social Media affect stock predictions? A case study on GameStop},
author = {Antonino Ferraro and Giancarlo Sperlì},
url = {https://linkinghub.elsevier.com/retrieve/pii/S2468696424000181},
doi = {10.1016/j.osnem.2024.100293},
issn = {24686964},
year = {2024},
date = {2024-11-01},
urldate = {2025-09-30},
journal = {Online Social Networks and Media},
volume = {43-44},
pages = {100293},
keywords = {Deep Learning, Financial systems, GameStop, Online social networks, Stock forecasting},
pubstate = {published},
tppubtype = {article}
}
Ferraro, Antonino; Gatta, Valerio La; Postiglione, Marco
Empowering Network Security with Autoencoders Proceedings Article
In: 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), pp. 346–350, IEEE, Seoul, Korea, Republic of, 2024, ISBN: 979-8-3503-7451-3.
Links | BibTeX | Tags: Autoencoders, Deep Learning, network anomaly detection, network security, network traffic analysis
@inproceedings{ferraro_empowering_2024,
title = {Empowering Network Security with Autoencoders},
author = {Antonino Ferraro and Valerio La Gatta and Marco Postiglione},
url = {https://ieeexplore.ieee.org/document/10627534/},
doi = {10.1109/ICASSPW62465.2024.10627534},
isbn = {979-8-3503-7451-3},
year = {2024},
date = {2024-04-01},
urldate = {2025-09-30},
booktitle = {2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW)},
pages = {346–350},
publisher = {IEEE},
address = {Seoul, Korea, Republic of},
keywords = {Autoencoders, Deep Learning, network anomaly detection, network security, network traffic analysis},
pubstate = {published},
tppubtype = {inproceedings}
}
Ferraro, Antonino; Galli, Antonio; Gatta, Valerio La; Minocchi, Mario; Moscato, Vincenzo; Postiglione, Marco
Few Shot NER on Augmented Unstructured Text from Cardiology Records Book Section
In: Barolli, Leonard (Ed.): Advances in Internet, Data & Web Technologies, vol. 193, pp. 1–12, Springer Nature Switzerland, Cham, 2024, ISBN: 978-3-031-53554-3 978-3-031-53555-0, (Series Title: Lecture Notes on Data Engineering and Communications Technologies).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Data Augmentation, Healthcare, Named-Entity Recognition
@incollection{ferraroFewShotNER2024,
title = {Few Shot NER on Augmented Unstructured Text from Cardiology Records},
author = {Antonino Ferraro and Antonio Galli and Valerio La Gatta and Mario Minocchi and Vincenzo Moscato and Marco Postiglione},
editor = {Leonard Barolli},
url = {https://link.springer.com/10.1007/978-3-031-53555-0_1},
doi = {10.1007/978-3-031-53555-0_1},
isbn = {978-3-031-53554-3 978-3-031-53555-0},
year = {2024},
date = {2024-01-01},
urldate = {2024-07-12},
booktitle = {Advances in Internet, Data & Web Technologies},
volume = {193},
pages = {1–12},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {The principal challenge encountered in the realm of Named-Entity Recognition lies in the acquisition of high-caliber annotated data. In certain languages and specialized domains, the availability of substantial datasets suitable for training models via traditional machine learning methodologies can prove to be a formidable obstacle [10]. In an effort to address this issue, we have explored a Policy-based Active Learning approach aimed at meticulously selecting the most advantageous instances generated through a Data Augmentation procedure [3, 6]. This endeavor was undertaken within the context of a few-shot scenario in the biomedical field. Our study has revealed the superiority of this strategy in comparison to active learning techniques relying on fixed metrics or random instance selection, guaranteeing the privacy of patients from whose medical records the source data were obtained and used. However, it is imperative to note that this approach entails heightened computational demands and necessitates a longer execution duration [7].},
note = {Series Title: Lecture Notes on Data Engineering and Communications Technologies},
keywords = {Artificial Intelligence, Data Augmentation, Healthcare, Named-Entity Recognition},
pubstate = {published},
tppubtype = {incollection}
}
The principal challenge encountered in the realm of Named-Entity Recognition lies in the acquisition of high-caliber annotated data. In certain languages and specialized domains, the availability of substantial datasets suitable for training models via traditional machine learning methodologies can prove to be a formidable obstacle [10]. In an effort to address this issue, we have explored a Policy-based Active Learning approach aimed at meticulously selecting the most advantageous instances generated through a Data Augmentation procedure [3, 6]. This endeavor was undertaken within the context of a few-shot scenario in the biomedical field. Our study has revealed the superiority of this strategy in comparison to active learning techniques relying on fixed metrics or random instance selection, guaranteeing the privacy of patients from whose medical records the source data were obtained and used. However, it is imperative to note that this approach entails heightened computational demands and necessitates a longer execution duration [7].
Ferraro, Antonino; Galli, Antonio; Gatta, Valerio La; Minocchi, Mario; Moscato, Vincenzo; Postiglione, Marco
Few Shot NER on Augmented Unstructured Text from Cardiology Records Book Section
In: Barolli, Leonard (Ed.): Advances in Internet, Data & Web Technologies, vol. 193, pp. 1–12, Springer Nature Switzerland, Cham, 2024, ISBN: 978-3-031-53554-3 978-3-031-53555-0, (Series Title: Lecture Notes on Data Engineering and Communications Technologies).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Data Augmentation, Healthcare, Named-Entity Recognition
@incollection{barolli_few_2024,
title = {Few Shot NER on Augmented Unstructured Text from Cardiology Records},
author = {Antonino Ferraro and Antonio Galli and Valerio La Gatta and Mario Minocchi and Vincenzo Moscato and Marco Postiglione},
editor = {Leonard Barolli},
url = {https://link.springer.com/10.1007/978-3-031-53555-0_1},
doi = {10.1007/978-3-031-53555-0_1},
isbn = {978-3-031-53554-3 978-3-031-53555-0},
year = {2024},
date = {2024-01-01},
urldate = {2024-07-12},
booktitle = {Advances in Internet, Data & Web Technologies},
volume = {193},
pages = {1–12},
publisher = {Springer Nature Switzerland},
address = {Cham},
abstract = {The principal challenge encountered in the realm of Named-Entity Recognition lies in the acquisition of high-caliber annotated data. In certain languages and specialized domains, the availability of substantial datasets suitable for training models via traditional machine learning methodologies can prove to be a formidable obstacle [10]. In an effort to address this issue, we have explored a Policy-based Active Learning approach aimed at meticulously selecting the most advantageous instances generated through a Data Augmentation procedure [3, 6]. This endeavor was undertaken within the context of a few-shot scenario in the biomedical field. Our study has revealed the superiority of this strategy in comparison to active learning techniques relying on fixed metrics or random instance selection, guaranteeing the privacy of patients from whose medical records the source data were obtained and used. However, it is imperative to note that this approach entails heightened computational demands and necessitates a longer execution duration [7].},
note = {Series Title: Lecture Notes on Data Engineering and Communications Technologies},
keywords = {Artificial Intelligence, Data Augmentation, Healthcare, Named-Entity Recognition},
pubstate = {published},
tppubtype = {incollection}
}
The principal challenge encountered in the realm of Named-Entity Recognition lies in the acquisition of high-caliber annotated data. In certain languages and specialized domains, the availability of substantial datasets suitable for training models via traditional machine learning methodologies can prove to be a formidable obstacle [10]. In an effort to address this issue, we have explored a Policy-based Active Learning approach aimed at meticulously selecting the most advantageous instances generated through a Data Augmentation procedure [3, 6]. This endeavor was undertaken within the context of a few-shot scenario in the biomedical field. Our study has revealed the superiority of this strategy in comparison to active learning techniques relying on fixed metrics or random instance selection, guaranteeing the privacy of patients from whose medical records the source data were obtained and used. However, it is imperative to note that this approach entails heightened computational demands and necessitates a longer execution duration [7].
2023
Russo, Raffaele; Giuseppe, Giuliano Di; Vanacore, Alessandro; Gatta, Valerio La; Ferraro, Antonino; Galli, Antonio; Postiglione, Marco; Moscato, Vincenzo
Graph-Based Approach for European Law Classification Proceedings Article
In: 2023 IEEE International Conference on Big Data (BigData), pp. 1–9, IEEE, Sorrento, Italy, 2023, ISBN: 979-8-3503-2445-7.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Semantics
@inproceedings{russoGraphBasedApproachEuropean2023,
title = {Graph-Based Approach for European Law Classification},
author = {Raffaele Russo and Giuliano Di Giuseppe and Alessandro Vanacore and Valerio La Gatta and Antonino Ferraro and Antonio Galli and Marco Postiglione and Vincenzo Moscato},
url = {https://ieeexplore.ieee.org/document/10386684/},
doi = {10.1109/BigData59044.2023.10386684},
isbn = {979-8-3503-2445-7},
year = {2023},
date = {2023-12-01},
urldate = {2024-07-12},
booktitle = {2023 IEEE International Conference on Big Data (BigData)},
pages = {1–9},
publisher = {IEEE},
address = {Sorrento, Italy},
abstract = {Deep learning, owing to its transformative influence across a myriad of sectors, has recently made its foray into the legal domain, instigated by the surge in digitization. Among the multitude of applications in this space, legal document classification emerges as a pivotal yet complex undertaking. Legal texts, characterized by unique domain-centric semantics and intricate linguistic patterns, necessitate precision-driven classification systems for numerous practical implications. This paper illuminates the challenges and opportunities in automating the classification of European Union (EU) legal documents, emphasizing the interrelationships among statutes and the hierarchical nature of legal references. In this context, we introduce a novel graph data modeling technique that adeptly marries content-centric indicators with the relational dynamics inherent among diverse legal documents. Central to our approach is a framework that melds text embeddings with graph neural networks for the classification of legal documents aligned with their subject-based directories. Empirical evaluations on the EU law dataset underline the efficacy of our model across varying granularities, from general thematic categories to intricate subtopics. This endeavor not only augments the comprehensibility and accessibility of EU jurisprudence but also holds significant implications across regulatory compliance, legal research, and policy formulation, underscoring the potential of deep learning in reshaping legal paradigms.},
keywords = {Artificial Intelligence, Semantics},
pubstate = {published},
tppubtype = {inproceedings}
}
Deep learning, owing to its transformative influence across a myriad of sectors, has recently made its foray into the legal domain, instigated by the surge in digitization. Among the multitude of applications in this space, legal document classification emerges as a pivotal yet complex undertaking. Legal texts, characterized by unique domain-centric semantics and intricate linguistic patterns, necessitate precision-driven classification systems for numerous practical implications. This paper illuminates the challenges and opportunities in automating the classification of European Union (EU) legal documents, emphasizing the interrelationships among statutes and the hierarchical nature of legal references. In this context, we introduce a novel graph data modeling technique that adeptly marries content-centric indicators with the relational dynamics inherent among diverse legal documents. Central to our approach is a framework that melds text embeddings with graph neural networks for the classification of legal documents aligned with their subject-based directories. Empirical evaluations on the EU law dataset underline the efficacy of our model across varying granularities, from general thematic categories to intricate subtopics. This endeavor not only augments the comprehensibility and accessibility of EU jurisprudence but also holds significant implications across regulatory compliance, legal research, and policy formulation, underscoring the potential of deep learning in reshaping legal paradigms.
Russo, Raffaele; Giuseppe, Giuliano Di; Vanacore, Alessandro; Gatta, Valerio La; Ferraro, Antonino; Galli, Antonio; Postiglione, Marco; Moscato, Vincenzo
Graph-Based Approach for European Law Classification Proceedings Article
In: 2023 IEEE International Conference on Big Data (BigData), pp. 1–9, IEEE, Sorrento, Italy, 2023, ISBN: 979-8-3503-2445-7.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Semantics
@inproceedings{russo_graph-based_2023,
title = {Graph-Based Approach for European Law Classification},
author = {Raffaele Russo and Giuliano Di Giuseppe and Alessandro Vanacore and Valerio La Gatta and Antonino Ferraro and Antonio Galli and Marco Postiglione and Vincenzo Moscato},
url = {https://ieeexplore.ieee.org/document/10386684/},
doi = {10.1109/BigData59044.2023.10386684},
isbn = {979-8-3503-2445-7},
year = {2023},
date = {2023-12-01},
urldate = {2024-07-12},
booktitle = {2023 IEEE International Conference on Big Data (BigData)},
pages = {1–9},
publisher = {IEEE},
address = {Sorrento, Italy},
abstract = {Deep learning, owing to its transformative influence across a myriad of sectors, has recently made its foray into the legal domain, instigated by the surge in digitization. Among the multitude of applications in this space, legal document classification emerges as a pivotal yet complex undertaking. Legal texts, characterized by unique domain-centric semantics and intricate linguistic patterns, necessitate precision-driven classification systems for numerous practical implications. This paper illuminates the challenges and opportunities in automating the classification of European Union (EU) legal documents, emphasizing the interrelationships among statutes and the hierarchical nature of legal references. In this context, we introduce a novel graph data modeling technique that adeptly marries content-centric indicators with the relational dynamics inherent among diverse legal documents. Central to our approach is a framework that melds text embeddings with graph neural networks for the classification of legal documents aligned with their subject-based directories. Empirical evaluations on the EU law dataset underline the efficacy of our model across varying granularities, from general thematic categories to intricate subtopics. This endeavor not only augments the comprehensibility and accessibility of EU jurisprudence but also holds significant implications across regulatory compliance, legal research, and policy formulation, underscoring the potential of deep learning in reshaping legal paradigms.},
keywords = {Artificial Intelligence, Semantics},
pubstate = {published},
tppubtype = {inproceedings}
}
Deep learning, owing to its transformative influence across a myriad of sectors, has recently made its foray into the legal domain, instigated by the surge in digitization. Among the multitude of applications in this space, legal document classification emerges as a pivotal yet complex undertaking. Legal texts, characterized by unique domain-centric semantics and intricate linguistic patterns, necessitate precision-driven classification systems for numerous practical implications. This paper illuminates the challenges and opportunities in automating the classification of European Union (EU) legal documents, emphasizing the interrelationships among statutes and the hierarchical nature of legal references. In this context, we introduce a novel graph data modeling technique that adeptly marries content-centric indicators with the relational dynamics inherent among diverse legal documents. Central to our approach is a framework that melds text embeddings with graph neural networks for the classification of legal documents aligned with their subject-based directories. Empirical evaluations on the EU law dataset underline the efficacy of our model across varying granularities, from general thematic categories to intricate subtopics. This endeavor not only augments the comprehensibility and accessibility of EU jurisprudence but also holds significant implications across regulatory compliance, legal research, and policy formulation, underscoring the potential of deep learning in reshaping legal paradigms.
Luca, Marco De; Fasolino, Anna Rita; Ferraro, Antonino; Moscato, Vincenzo; Sperlí, Giancarlo; Tramontana, Porfirio
A community detection approach based on network representation learning for repository mining Journal Article
In: Expert Systems with Applications, vol. 231, pp. 120597, 2023, ISSN: 09574174.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Community detection, Developer social network, Graph-embedding, Repository mining, Social network analysis, Social Networks
@article{de_luca_community_2023,
title = {A community detection approach based on network representation learning for repository mining},
author = {Marco De Luca and Anna Rita Fasolino and Antonino Ferraro and Vincenzo Moscato and Giancarlo Sperlí and Porfirio Tramontana},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0957417423010990},
doi = {10.1016/j.eswa.2023.120597},
issn = {09574174},
year = {2023},
date = {2023-11-01},
urldate = {2024-07-12},
journal = {Expert Systems with Applications},
volume = {231},
pages = {120597},
abstract = {In this paper, we propose a novel heterogeneous graph-based model for capturing and handling all the complex and strongly-correlated information of a software Developer Social Network (DSN) to support several analytic tasks. In particular, we challenge the problem of automatically discovering communities of software developers sharing interests for similar projects by relying on Social Network Analysis (SNA) findings. To overcome the huge graph-size issue, we leverage different graph embedding techniques. Eventually, we evaluate the proposed approach with respect to state-of-the-art approaches from an efficiency and an effectiveness point of view by carrying out an experiment involving the GitHub dataset.},
keywords = {Artificial Intelligence, Community detection, Developer social network, Graph-embedding, Repository mining, Social network analysis, Social Networks},
pubstate = {published},
tppubtype = {article}
}
In this paper, we propose a novel heterogeneous graph-based model for capturing and handling all the complex and strongly-correlated information of a software Developer Social Network (DSN) to support several analytic tasks. In particular, we challenge the problem of automatically discovering communities of software developers sharing interests for similar projects by relying on Social Network Analysis (SNA) findings. To overcome the huge graph-size issue, we leverage different graph embedding techniques. Eventually, we evaluate the proposed approach with respect to state-of-the-art approaches from an efficiency and an effectiveness point of view by carrying out an experiment involving the GitHub dataset.
Ferraro, Antonino; Galli, Antonio; Gatta, Valerio La; Postiglione, Marco
Benchmarking open source and paid services for speech to text: an analysis of quality and input variety Journal Article
In: Frontiers in Big Data, vol. 6, pp. 1210559, 2023, ISSN: 2624-909X.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, ASR, Benchmark, Multimedia, Speech Recognition, Speech to text
@article{ferraro_benchmarking_2023,
title = {Benchmarking open source and paid services for speech to text: an analysis of quality and input variety},
author = {Antonino Ferraro and Antonio Galli and Valerio La Gatta and Marco Postiglione},
url = {https://www.frontiersin.org/articles/10.3389/fdata.2023.1210559/full},
doi = {10.3389/fdata.2023.1210559},
issn = {2624-909X},
year = {2023},
date = {2023-09-01},
urldate = {2024-07-12},
journal = {Frontiers in Big Data},
volume = {6},
pages = {1210559},
abstract = {Introduction
Speech to text (STT) technology has seen increased usage in recent years for automating transcription of spoken language. To choose the most suitable tool for a given task, it is essential to evaluate the performance and quality of both open source and paid STT services.
Methods
In this paper, we conduct a benchmarking study of open source and paid STT services, with a specific focus on assessing their performance concerning the variety of input text. We utilizes ix datasets obtained from diverse sources, including interviews, lectures, and speeches, as input for the STT tools. The evaluation of the instruments employs the Word Error Rate (WER), a standard metric for STT evaluation.
Results
Our analysis of the results demonstrates significant variations in the performance of the STT tools based on the input text. Certain tools exhibit superior performance on specific types of audio samples compared to others. Our study provides insights into STT tool performance when handling substantial data volumes, as well as the challenges and opportunities posed by the multimedia nature of the data.
Discussion
Although paid services generally demonstrate better accuracy and speed compared to open source alternatives, their performance remains dependent on the input text. The study highlights the need for considering specific requirements and characteristics of the audio samples when selecting an appropriate STT tool.},
keywords = {Artificial Intelligence, ASR, Benchmark, Multimedia, Speech Recognition, Speech to text},
pubstate = {published},
tppubtype = {article}
}
Introduction
Speech to text (STT) technology has seen increased usage in recent years for automating transcription of spoken language. To choose the most suitable tool for a given task, it is essential to evaluate the performance and quality of both open source and paid STT services.
Methods
In this paper, we conduct a benchmarking study of open source and paid STT services, with a specific focus on assessing their performance concerning the variety of input text. We utilizes ix datasets obtained from diverse sources, including interviews, lectures, and speeches, as input for the STT tools. The evaluation of the instruments employs the Word Error Rate (WER), a standard metric for STT evaluation.
Results
Our analysis of the results demonstrates significant variations in the performance of the STT tools based on the input text. Certain tools exhibit superior performance on specific types of audio samples compared to others. Our study provides insights into STT tool performance when handling substantial data volumes, as well as the challenges and opportunities posed by the multimedia nature of the data.
Discussion
Although paid services generally demonstrate better accuracy and speed compared to open source alternatives, their performance remains dependent on the input text. The study highlights the need for considering specific requirements and characteristics of the audio samples when selecting an appropriate STT tool.
Speech to text (STT) technology has seen increased usage in recent years for automating transcription of spoken language. To choose the most suitable tool for a given task, it is essential to evaluate the performance and quality of both open source and paid STT services.
Methods
In this paper, we conduct a benchmarking study of open source and paid STT services, with a specific focus on assessing their performance concerning the variety of input text. We utilizes ix datasets obtained from diverse sources, including interviews, lectures, and speeches, as input for the STT tools. The evaluation of the instruments employs the Word Error Rate (WER), a standard metric for STT evaluation.
Results
Our analysis of the results demonstrates significant variations in the performance of the STT tools based on the input text. Certain tools exhibit superior performance on specific types of audio samples compared to others. Our study provides insights into STT tool performance when handling substantial data volumes, as well as the challenges and opportunities posed by the multimedia nature of the data.
Discussion
Although paid services generally demonstrate better accuracy and speed compared to open source alternatives, their performance remains dependent on the input text. The study highlights the need for considering specific requirements and characteristics of the audio samples when selecting an appropriate STT tool.
Ferraro, Antonino; Galli, Antonio; Moscato, Vincenzo; Sperlì, Giancarlo
Evaluating eXplainable artificial intelligence tools for hard disk drive predictive maintenance Journal Article
In: Artificial Intelligence Review, vol. 56, no. 7, pp. 7279–7314, 2023, ISSN: 0269-2821, 1573-7462.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Explainable AI, Industry 4.0
@article{ferraroEvaluatingEXplainableArtificial2023,
title = {Evaluating eXplainable artificial intelligence tools for hard disk drive predictive maintenance},
author = {Antonino Ferraro and Antonio Galli and Vincenzo Moscato and Giancarlo Sperlì},
url = {https://link.springer.com/10.1007/s10462-022-10354-7},
doi = {10.1007/s10462-022-10354-7},
issn = {0269-2821, 1573-7462},
year = {2023},
date = {2023-07-01},
urldate = {2024-07-12},
journal = {Artificial Intelligence Review},
volume = {56},
number = {7},
pages = {7279–7314},
abstract = {In the last years, one of the main challenges in Industry 4.0 concerns maintenance operations optimization, which has been widely dealt with several predictive maintenance frameworks aiming to jointly reduce maintenance costs and downtime intervals. Nevertheless, the most recent and effective frameworks mainly rely on deep learning models, but their internal representations (black box) are too complex for human understanding making difficult explain their predictions. This issue can be challenged by using eXplainable artificial intelligence (XAI) methodologies, the aim of which is to explain the decisions of data-driven AI models, characterizing the strengths and weaknesses of the decision-making process by making results more understandable by humans. In this paper, we focus on explanation of the predictions made by a recurrent neural networks based model, which requires a tree-dimensional dataset because it exploits spatial and temporal features for estimating remaining useful life (RUL) of hard disk drives (HDDs). In particular, we have analyzed in depth as explanations about RUL prediction provided by different XAI tools, compared using different metrics and showing the generated dashboards, can be really useful for supporting predictive maintenance task by means of both global and local explanations. For this aim, we have realized an explanation framework able to investigate local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) tools w.r.t. to the Backblaze Dataset and a long short-term memory (LSTM) prediction model. The achieved results show how SHAP outperforms LIME in almost all the considered metrics, resulting a suitable and effective solution for HDD predictive maintenance applications.},
keywords = {Artificial Intelligence, Explainable AI, Industry 4.0},
pubstate = {published},
tppubtype = {article}
}
In the last years, one of the main challenges in Industry 4.0 concerns maintenance operations optimization, which has been widely dealt with several predictive maintenance frameworks aiming to jointly reduce maintenance costs and downtime intervals. Nevertheless, the most recent and effective frameworks mainly rely on deep learning models, but their internal representations (black box) are too complex for human understanding making difficult explain their predictions. This issue can be challenged by using eXplainable artificial intelligence (XAI) methodologies, the aim of which is to explain the decisions of data-driven AI models, characterizing the strengths and weaknesses of the decision-making process by making results more understandable by humans. In this paper, we focus on explanation of the predictions made by a recurrent neural networks based model, which requires a tree-dimensional dataset because it exploits spatial and temporal features for estimating remaining useful life (RUL) of hard disk drives (HDDs). In particular, we have analyzed in depth as explanations about RUL prediction provided by different XAI tools, compared using different metrics and showing the generated dashboards, can be really useful for supporting predictive maintenance task by means of both global and local explanations. For this aim, we have realized an explanation framework able to investigate local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) tools w.r.t. to the Backblaze Dataset and a long short-term memory (LSTM) prediction model. The achieved results show how SHAP outperforms LIME in almost all the considered metrics, resulting a suitable and effective solution for HDD predictive maintenance applications.
Ferraro, Antonino; Galli, Antonio; Moscato, Vincenzo; Sperlì, Giancarlo
Evaluating eXplainable artificial intelligence tools for hard disk drive predictive maintenance Journal Article
In: Artificial Intelligence Review, vol. 56, no. 7, pp. 7279–7314, 2023, ISSN: 0269-2821, 1573-7462.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Explainable AI, Industry 4.0
@article{ferraro_evaluating_2023,
title = {Evaluating eXplainable artificial intelligence tools for hard disk drive predictive maintenance},
author = {Antonino Ferraro and Antonio Galli and Vincenzo Moscato and Giancarlo Sperlì},
url = {https://link.springer.com/10.1007/s10462-022-10354-7},
doi = {10.1007/s10462-022-10354-7},
issn = {0269-2821, 1573-7462},
year = {2023},
date = {2023-07-01},
urldate = {2024-07-12},
journal = {Artificial Intelligence Review},
volume = {56},
number = {7},
pages = {7279–7314},
abstract = {In the last years, one of the main challenges in Industry 4.0 concerns maintenance operations optimization, which has been widely dealt with several predictive maintenance frameworks aiming to jointly reduce maintenance costs and downtime intervals. Nevertheless, the most recent and effective frameworks mainly rely on deep learning models, but their internal representations (black box) are too complex for human understanding making difficult explain their predictions. This issue can be challenged by using eXplainable artificial intelligence (XAI) methodologies, the aim of which is to explain the decisions of data-driven AI models, characterizing the strengths and weaknesses of the decision-making process by making results more understandable by humans. In this paper, we focus on explanation of the predictions made by a recurrent neural networks based model, which requires a tree-dimensional dataset because it exploits spatial and temporal features for estimating remaining useful life (RUL) of hard disk drives (HDDs). In particular, we have analyzed in depth as explanations about RUL prediction provided by different XAI tools, compared using different metrics and showing the generated dashboards, can be really useful for supporting predictive maintenance task by means of both global and local explanations. For this aim, we have realized an explanation framework able to investigate local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) tools w.r.t. to the Backblaze Dataset and a long short-term memory (LSTM) prediction model. The achieved results show how SHAP outperforms LIME in almost all the considered metrics, resulting a suitable and effective solution for HDD predictive maintenance applications.},
keywords = {Artificial Intelligence, Explainable AI, Industry 4.0},
pubstate = {published},
tppubtype = {article}
}
In the last years, one of the main challenges in Industry 4.0 concerns maintenance operations optimization, which has been widely dealt with several predictive maintenance frameworks aiming to jointly reduce maintenance costs and downtime intervals. Nevertheless, the most recent and effective frameworks mainly rely on deep learning models, but their internal representations (black box) are too complex for human understanding making difficult explain their predictions. This issue can be challenged by using eXplainable artificial intelligence (XAI) methodologies, the aim of which is to explain the decisions of data-driven AI models, characterizing the strengths and weaknesses of the decision-making process by making results more understandable by humans. In this paper, we focus on explanation of the predictions made by a recurrent neural networks based model, which requires a tree-dimensional dataset because it exploits spatial and temporal features for estimating remaining useful life (RUL) of hard disk drives (HDDs). In particular, we have analyzed in depth as explanations about RUL prediction provided by different XAI tools, compared using different metrics and showing the generated dashboards, can be really useful for supporting predictive maintenance task by means of both global and local explanations. For this aim, we have realized an explanation framework able to investigate local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) tools w.r.t. to the Backblaze Dataset and a long short-term memory (LSTM) prediction model. The achieved results show how SHAP outperforms LIME in almost all the considered metrics, resulting a suitable and effective solution for HDD predictive maintenance applications.
De Santo, Aniello; Ferraro, Antonino; Moscato, Vincenzo; Sperlí, Giancarlo
An action–reaction influence model relying on OSN user-generated content Journal Article
In: Knowledge and Information Systems, vol. 65, no. 5, pp. 2251–2280, 2023, ISSN: 0219-1377, 0219-3116.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Diffusion model, Heterogeneous online social network, Influence analysis, Influence maximization, Social network analysis, Social Networks
@article{desanto_actionreaction_2023,
title = {An action–reaction influence model relying on OSN user-generated content},
author = {Aniello De Santo and Antonino Ferraro and Vincenzo Moscato and Giancarlo Sperlí},
url = {https://link.springer.com/10.1007/s10115-023-01833-6},
doi = {10.1007/s10115-023-01833-6},
issn = {0219-1377, 0219-3116},
year = {2023},
date = {2023-05-01},
urldate = {2024-07-12},
journal = {Knowledge and Information Systems},
volume = {65},
number = {5},
pages = {2251–2280},
abstract = {Due to the sustained popularization of Online Social Networks (OSNs), it has become of interest for a variety of domains of applications to correctly characterize how the behavior of an individual user can be influenced by the actions of other users in a network. Additionally, the richness of available features in modern OSNs highlights the growing importance of user-generated data in establishing user relations. In this paper, we follow a data-driven methodology and propose a diffusion algorithm designed around user-to-content relationships and an action–reaction paradigm. Crucially, we design our approach by integrating different cross-disciplinary theories of how users influence each other. Thus, we enrich the influence maximization task with a psychological dimension and define a model that ties influence diffusion to recurrent users’ behavior from OSN logs, considering relationships between users mediated by user-generated content. We evaluate our approach over the Yahoo Flickr Creative Commons 100 Million real-world dataset. We measure efficiency and effectiveness by analyzing scalability and spread efficacy and show how our model outperforms existing state-of-the-art methods.},
keywords = {Artificial Intelligence, Diffusion model, Heterogeneous online social network, Influence analysis, Influence maximization, Social network analysis, Social Networks},
pubstate = {published},
tppubtype = {article}
}
Due to the sustained popularization of Online Social Networks (OSNs), it has become of interest for a variety of domains of applications to correctly characterize how the behavior of an individual user can be influenced by the actions of other users in a network. Additionally, the richness of available features in modern OSNs highlights the growing importance of user-generated data in establishing user relations. In this paper, we follow a data-driven methodology and propose a diffusion algorithm designed around user-to-content relationships and an action–reaction paradigm. Crucially, we design our approach by integrating different cross-disciplinary theories of how users influence each other. Thus, we enrich the influence maximization task with a psychological dimension and define a model that ties influence diffusion to recurrent users’ behavior from OSN logs, considering relationships between users mediated by user-generated content. We evaluate our approach over the Yahoo Flickr Creative Commons 100 Million real-world dataset. We measure efficiency and effectiveness by analyzing scalability and spread efficacy and show how our model outperforms existing state-of-the-art methods.
Luca, Roberto De; Ferraro, Antonino; Galli, Antonio; Gallo, Mosè; Moscato, Vincenzo; Sperlì, Giancarlo
A deep attention based approach for predictive maintenance applications in IoT scenarios Journal Article
In: Journal of Manufacturing Technology Management, vol. 34, no. 4, pp. 535–556, 2023, ISSN: 1741-038X.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Decision Making, Decision Support Systems, Deep Learning, Deep neural networks, Industry 4.0, Predictive Maintenance
@article{de_luca_deep_2023,
title = {A deep attention based approach for predictive maintenance applications in IoT scenarios},
author = {Roberto De Luca and Antonino Ferraro and Antonio Galli and Mosè Gallo and Vincenzo Moscato and Giancarlo Sperlì},
url = {https://www.emerald.com/insight/content/doi/10.1108/JMTM-02-2022-0093/full/html},
doi = {10.1108/JMTM-02-2022-0093},
issn = {1741-038X},
year = {2023},
date = {2023-05-01},
urldate = {2024-07-12},
journal = {Journal of Manufacturing Technology Management},
volume = {34},
number = {4},
pages = {535–556},
abstract = {Purpose
The recent innovations of Industry 4.0 have made it possible to easily collect data related to a production environment. In this context, information about industrial equipment – gathered by proper sensors – can be profitably used for supporting predictive maintenance (PdM) through the application of data-driven analytics based on artificial intelligence (AI) techniques. Although deep learning (DL) approaches have proven to be a quite effective solutions to the problem, one of the open research challenges remains – the design of PdM methods that are computationally efficient, and most importantly, applicable in real-world internet of things (IoT) scenarios, where they are required to be executable directly on the limited devices’ hardware.
Design/methodology/approach
In this paper, the authors propose a DL approach for PdM task, which is based on a particular and very efficient architecture. The major novelty behind the proposed framework is to leverage a multi-head attention (MHA) mechanism to obtain both high results in terms of remaining useful life (RUL) estimation and low memory model storage requirements, providing the basis for a possible implementation directly on the equipment hardware.
Findings
The achieved experimental results on the NASA dataset show how the authors’ approach outperforms in terms of effectiveness and efficiency the majority of the most diffused state-of-the-art techniques.
Research limitations/implications
A comparison of the spatial and temporal complexity with a typical long-short term memory (LSTM) model and the state-of-the-art approaches was also done on the NASA dataset. Despite the authors’ approach achieving similar effectiveness results with respect to other approaches, it has a significantly smaller number of parameters, a smaller storage volume and lower training time.
Practical implications
The proposed approach aims to find a compromise between effectiveness and efficiency, which is crucial in the industrial domain in which it is important to maximize the link between performance attained and resources allocated. The overall accuracy performances are also on par with the finest methods described in the literature.
Originality/value
The proposed approach allows satisfying the requirements of modern embedded AI applications (reliability, low power consumption, etc.), finding a compromise between efficiency and effectiveness.},
keywords = {Artificial Intelligence, Decision Making, Decision Support Systems, Deep Learning, Deep neural networks, Industry 4.0, Predictive Maintenance},
pubstate = {published},
tppubtype = {article}
}
Purpose
The recent innovations of Industry 4.0 have made it possible to easily collect data related to a production environment. In this context, information about industrial equipment – gathered by proper sensors – can be profitably used for supporting predictive maintenance (PdM) through the application of data-driven analytics based on artificial intelligence (AI) techniques. Although deep learning (DL) approaches have proven to be a quite effective solutions to the problem, one of the open research challenges remains – the design of PdM methods that are computationally efficient, and most importantly, applicable in real-world internet of things (IoT) scenarios, where they are required to be executable directly on the limited devices’ hardware.
Design/methodology/approach
In this paper, the authors propose a DL approach for PdM task, which is based on a particular and very efficient architecture. The major novelty behind the proposed framework is to leverage a multi-head attention (MHA) mechanism to obtain both high results in terms of remaining useful life (RUL) estimation and low memory model storage requirements, providing the basis for a possible implementation directly on the equipment hardware.
Findings
The achieved experimental results on the NASA dataset show how the authors’ approach outperforms in terms of effectiveness and efficiency the majority of the most diffused state-of-the-art techniques.
Research limitations/implications
A comparison of the spatial and temporal complexity with a typical long-short term memory (LSTM) model and the state-of-the-art approaches was also done on the NASA dataset. Despite the authors’ approach achieving similar effectiveness results with respect to other approaches, it has a significantly smaller number of parameters, a smaller storage volume and lower training time.
Practical implications
The proposed approach aims to find a compromise between effectiveness and efficiency, which is crucial in the industrial domain in which it is important to maximize the link between performance attained and resources allocated. The overall accuracy performances are also on par with the finest methods described in the literature.
Originality/value
The proposed approach allows satisfying the requirements of modern embedded AI applications (reliability, low power consumption, etc.), finding a compromise between efficiency and effectiveness.
The recent innovations of Industry 4.0 have made it possible to easily collect data related to a production environment. In this context, information about industrial equipment – gathered by proper sensors – can be profitably used for supporting predictive maintenance (PdM) through the application of data-driven analytics based on artificial intelligence (AI) techniques. Although deep learning (DL) approaches have proven to be a quite effective solutions to the problem, one of the open research challenges remains – the design of PdM methods that are computationally efficient, and most importantly, applicable in real-world internet of things (IoT) scenarios, where they are required to be executable directly on the limited devices’ hardware.
Design/methodology/approach
In this paper, the authors propose a DL approach for PdM task, which is based on a particular and very efficient architecture. The major novelty behind the proposed framework is to leverage a multi-head attention (MHA) mechanism to obtain both high results in terms of remaining useful life (RUL) estimation and low memory model storage requirements, providing the basis for a possible implementation directly on the equipment hardware.
Findings
The achieved experimental results on the NASA dataset show how the authors’ approach outperforms in terms of effectiveness and efficiency the majority of the most diffused state-of-the-art techniques.
Research limitations/implications
A comparison of the spatial and temporal complexity with a typical long-short term memory (LSTM) model and the state-of-the-art approaches was also done on the NASA dataset. Despite the authors’ approach achieving similar effectiveness results with respect to other approaches, it has a significantly smaller number of parameters, a smaller storage volume and lower training time.
Practical implications
The proposed approach aims to find a compromise between effectiveness and efficiency, which is crucial in the industrial domain in which it is important to maximize the link between performance attained and resources allocated. The overall accuracy performances are also on par with the finest methods described in the literature.
Originality/value
The proposed approach allows satisfying the requirements of modern embedded AI applications (reliability, low power consumption, etc.), finding a compromise between efficiency and effectiveness.
Barolli, Leonard; Ferraro, Antonino
A Prediction Approach in Health Domain Combining Encoding Strategies and Neural Networks Book Section
In: Barolli, Leonard (Ed.): Advances on P2P, Parallel, Grid, Cloud and Internet Computing, vol. 571, pp. 129–136, Springer International Publishing, Cham, 2023, ISBN: 978-3-031-19944-8 978-3-031-19945-5, (Series Title: Lecture Notes in Networks and Systems).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Deep Learning, Depression, Encoding strategies, Gramian angular field, Healthcare
@incollection{barolli_prediction_2023,
title = {A Prediction Approach in Health Domain Combining Encoding Strategies and Neural Networks},
author = {Leonard Barolli and Antonino Ferraro},
editor = {Leonard Barolli},
url = {https://link.springer.com/10.1007/978-3-031-19945-5_12},
doi = {10.1007/978-3-031-19945-5_12},
isbn = {978-3-031-19944-8 978-3-031-19945-5},
year = {2023},
date = {2023-01-01},
urldate = {2024-07-12},
booktitle = {Advances on P2P, Parallel, Grid, Cloud and Internet Computing},
volume = {571},
pages = {129–136},
publisher = {Springer International Publishing},
address = {Cham},
abstract = {Healthcare has always been of paramount importance in the world of scientific research, and the advent of Artificial Intelligence (AI) has contributed to enormous strides in the field of prevention. In particular, research has focused on developing Machine Learning (ML)-based approaches to provide accurate prediction mechanisms to prevent and minimize any health complications [1]. This paper proposes an approach based on Gramian Angular Field (GAF) coding and a convolutional neural network (CNN) to solve a health prediction problem, specifically inherent to a subject’s depressive state.
Specifically, GAF is applied to transform information about a subject’s health condition, modeled as a time series, into images to be used as input for CNN to improve prediction performance. Experiments demonstrate superior performance to the best approach presented to the scientific community.},
note = {Series Title: Lecture Notes in Networks and Systems},
keywords = {Artificial Intelligence, Deep Learning, Depression, Encoding strategies, Gramian angular field, Healthcare},
pubstate = {published},
tppubtype = {incollection}
}
Healthcare has always been of paramount importance in the world of scientific research, and the advent of Artificial Intelligence (AI) has contributed to enormous strides in the field of prevention. In particular, research has focused on developing Machine Learning (ML)-based approaches to provide accurate prediction mechanisms to prevent and minimize any health complications [1]. This paper proposes an approach based on Gramian Angular Field (GAF) coding and a convolutional neural network (CNN) to solve a health prediction problem, specifically inherent to a subject’s depressive state.
Specifically, GAF is applied to transform information about a subject’s health condition, modeled as a time series, into images to be used as input for CNN to improve prediction performance. Experiments demonstrate superior performance to the best approach presented to the scientific community.
Specifically, GAF is applied to transform information about a subject’s health condition, modeled as a time series, into images to be used as input for CNN to improve prediction performance. Experiments demonstrate superior performance to the best approach presented to the scientific community.
Amato, Flora; Barolli, Leonard; Cozzolino, Giovanni; Ferraro, Antonino; Giacalone, Marco
An Intelligent Interface for Human-Computer Interaction in Legal Domain Book Section
In: Barolli, Leonard (Ed.): Advances on P2P, Parallel, Grid, Cloud and Internet Computing, vol. 571, pp. 240–248, Springer International Publishing, Cham, 2023, ISBN: 978-3-031-19944-8 978-3-031-19945-5, (Series Title: Lecture Notes in Networks and Systems).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Chatbot
@incollection{amatoIntelligentInterfaceHumanComputer2023,
title = {An Intelligent Interface for Human-Computer Interaction in Legal Domain},
author = {Flora Amato and Leonard Barolli and Giovanni Cozzolino and Antonino Ferraro and Marco Giacalone},
editor = {Leonard Barolli},
url = {https://link.springer.com/10.1007/978-3-031-19945-5_24},
doi = {10.1007/978-3-031-19945-5_24},
isbn = {978-3-031-19944-8 978-3-031-19945-5},
year = {2023},
date = {2023-01-01},
urldate = {2024-07-12},
booktitle = {Advances on P2P, Parallel, Grid, Cloud and Internet Computing},
volume = {571},
pages = {240–248},
publisher = {Springer International Publishing},
address = {Cham},
abstract = {Technological evolution and advances in the field of artificial intelligence have brought about considerable transformations in every area of our lives, also changing our various needs. In particular, the ever-increasing development and use of messaging applications have enabled the growth of services closer to users, as they can be seen as excellent means of advertising, sales and customer service. This is precisely why business models have changed drastically, moving towards new technologies such as ChatBots. The messaging applications nowadays are used daily while ensuring that they can keep up with the pace of this increasingly hectic and demanding world thanks to their 24/7 availability, low costs and customised real-time services. This paper aims to provide a general description and design principle of a ChatBot, designed and developed for the CREA2 (Conflict Resolution with Equitative Algorithms) platform, which includes the management and automatic resolution of disputes concerning the division of assets, trying to avoid costs and bureaucracy.},
note = {Series Title: Lecture Notes in Networks and Systems},
keywords = {Artificial Intelligence, Chatbot},
pubstate = {published},
tppubtype = {incollection}
}
Technological evolution and advances in the field of artificial intelligence have brought about considerable transformations in every area of our lives, also changing our various needs. In particular, the ever-increasing development and use of messaging applications have enabled the growth of services closer to users, as they can be seen as excellent means of advertising, sales and customer service. This is precisely why business models have changed drastically, moving towards new technologies such as ChatBots. The messaging applications nowadays are used daily while ensuring that they can keep up with the pace of this increasingly hectic and demanding world thanks to their 24/7 availability, low costs and customised real-time services. This paper aims to provide a general description and design principle of a ChatBot, designed and developed for the CREA2 (Conflict Resolution with Equitative Algorithms) platform, which includes the management and automatic resolution of disputes concerning the division of assets, trying to avoid costs and bureaucracy.
Amato, Flora; Barolli, Leonard; Cozzolino, Giovanni; Ferraro, Antonino; Giacalone, Marco
An Intelligent Interface for Human-Computer Interaction in Legal Domain Book Section
In: Barolli, Leonard (Ed.): Advances on P2P, Parallel, Grid, Cloud and Internet Computing, vol. 571, pp. 240–248, Springer International Publishing, Cham, 2023, ISBN: 978-3-031-19944-8 978-3-031-19945-5, (Series Title: Lecture Notes in Networks and Systems).
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Chatbot
@incollection{barolli_intelligent_2023,
title = {An Intelligent Interface for Human-Computer Interaction in Legal Domain},
author = {Flora Amato and Leonard Barolli and Giovanni Cozzolino and Antonino Ferraro and Marco Giacalone},
editor = {Leonard Barolli},
url = {https://link.springer.com/10.1007/978-3-031-19945-5_24},
doi = {10.1007/978-3-031-19945-5_24},
isbn = {978-3-031-19944-8 978-3-031-19945-5},
year = {2023},
date = {2023-01-01},
urldate = {2024-07-12},
booktitle = {Advances on P2P, Parallel, Grid, Cloud and Internet Computing},
volume = {571},
pages = {240–248},
publisher = {Springer International Publishing},
address = {Cham},
abstract = {Technological evolution and advances in the field of artificial intelligence have brought about considerable transformations in every area of our lives, also changing our various needs. In particular, the ever-increasing development and use of messaging applications have enabled the growth of services closer to users, as they can be seen as excellent means of advertising, sales and customer service. This is precisely why business models have changed drastically, moving towards new technologies such as ChatBots. The messaging applications nowadays are used daily while ensuring that they can keep up with the pace of this increasingly hectic and demanding world thanks to their 24/7 availability, low costs and customised real-time services. This paper aims to provide a general description and design principle of a ChatBot, designed and developed for the CREA2 (Conflict Resolution with Equitative Algorithms) platform, which includes the management and automatic resolution of disputes concerning the division of assets, trying to avoid costs and bureaucracy.},
note = {Series Title: Lecture Notes in Networks and Systems},
keywords = {Artificial Intelligence, Chatbot},
pubstate = {published},
tppubtype = {incollection}
}
Technological evolution and advances in the field of artificial intelligence have brought about considerable transformations in every area of our lives, also changing our various needs. In particular, the ever-increasing development and use of messaging applications have enabled the growth of services closer to users, as they can be seen as excellent means of advertising, sales and customer service. This is precisely why business models have changed drastically, moving towards new technologies such as ChatBots. The messaging applications nowadays are used daily while ensuring that they can keep up with the pace of this increasingly hectic and demanding world thanks to their 24/7 availability, low costs and customised real-time services. This paper aims to provide a general description and design principle of a ChatBot, designed and developed for the CREA2 (Conflict Resolution with Equitative Algorithms) platform, which includes the management and automatic resolution of disputes concerning the division of assets, trying to avoid costs and bureaucracy.
2022
Fiore, Emanuele Di; Ferraro, Antonino; Galli, Antonio; Moscato, Vincenzo; Sperlì, Giancarlo
An anomalous sound detection methodology for predictive maintenance Journal Article
In: Expert Systems with Applications, vol. 209, pp. 118324, 2022, ISSN: 09574174.
Abstract | Links | BibTeX | Tags: Anomalous Sound detection, Artificial Intelligence, Cybersecurity, Deep Learning, Predictive Maintenance
@article{di_fiore_anomalous_2022,
title = {An anomalous sound detection methodology for predictive maintenance},
author = {Emanuele Di Fiore and Antonino Ferraro and Antonio Galli and Vincenzo Moscato and Giancarlo Sperlì},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0957417422014518},
doi = {10.1016/j.eswa.2022.118324},
issn = {09574174},
year = {2022},
date = {2022-12-01},
urldate = {2024-07-12},
journal = {Expert Systems with Applications},
volume = {209},
pages = {118324},
abstract = {In the last decade, Anomalous Sound Detection (ASD) is becoming an increasingly challenging task for a plethora of applications due to the widespread diffusion of Deep Neural Networks. Nevertheless, the arise of recent cyber–physical attacks (i.e. Triton or Stuxnet), that deceive monitoring platforms, pose novel and challenging issues. For this reason, advanced predictive maintenance techniques are starting to exploit sounds generated by particular industrial equipment, whose analysis can unveil symptom of possible failures. For this kind of context, it is very easy to collect data related to normal and abnormal behavior of a given machinery, thus several kinds of deep neural architectures can be effectively trained to predict eventual downtime situations. In this paper, we propose a novel deep learning-based methodology for anomalous sound detection task, having flexibility, modularity and efficiency characteristics. The proposed methodology analyzes audio clips based on the mel-spectogram and ID equipment information, while a one-hot encoding method extracts features that are, successively, used to train an ID Conditioned Network. In particular, the main novelty of the proposed methodology concerns the conditioning of an autoencoder by jointly analyzing the relationships between mel-spectogram and the related machine identifier through an encoder–decoder architecture for computing an anomaly score related to the input sequence. Several experiments have been made for investigating the efficiency and effectiveness of the proposed methodology on multiple instances of different industrial machines (pumps, valves, slide rails and fans), achieving low inference time and memory requirements w.r.t. the other approaches in the literature.},
keywords = {Anomalous Sound detection, Artificial Intelligence, Cybersecurity, Deep Learning, Predictive Maintenance},
pubstate = {published},
tppubtype = {article}
}
In the last decade, Anomalous Sound Detection (ASD) is becoming an increasingly challenging task for a plethora of applications due to the widespread diffusion of Deep Neural Networks. Nevertheless, the arise of recent cyber–physical attacks (i.e. Triton or Stuxnet), that deceive monitoring platforms, pose novel and challenging issues. For this reason, advanced predictive maintenance techniques are starting to exploit sounds generated by particular industrial equipment, whose analysis can unveil symptom of possible failures. For this kind of context, it is very easy to collect data related to normal and abnormal behavior of a given machinery, thus several kinds of deep neural architectures can be effectively trained to predict eventual downtime situations. In this paper, we propose a novel deep learning-based methodology for anomalous sound detection task, having flexibility, modularity and efficiency characteristics. The proposed methodology analyzes audio clips based on the mel-spectogram and ID equipment information, while a one-hot encoding method extracts features that are, successively, used to train an ID Conditioned Network. In particular, the main novelty of the proposed methodology concerns the conditioning of an autoencoder by jointly analyzing the relationships between mel-spectogram and the related machine identifier through an encoder–decoder architecture for computing an anomaly score related to the input sequence. Several experiments have been made for investigating the efficiency and effectiveness of the proposed methodology on multiple instances of different industrial machines (pumps, valves, slide rails and fans), achieving low inference time and memory requirements w.r.t. the other approaches in the literature.
Santo, Aniello De; Ferraro, Antonino; Galli, Antonio; Moscato, Vincenzo; Sperlì, Giancarlo
Evaluating time series encoding techniques for Predictive Maintenance Journal Article
In: Expert Systems with Applications, vol. 210, pp. 118435, 2022, ISSN: 09574174.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Deep Learning, Failure prediction, Industry 4.0, Predictive Maintenance, Time series Encoding techniques
@article{de_santo_evaluating_2022,
title = {Evaluating time series encoding techniques for Predictive Maintenance},
author = {Aniello De Santo and Antonino Ferraro and Antonio Galli and Vincenzo Moscato and Giancarlo Sperlì},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0957417422015342},
doi = {10.1016/j.eswa.2022.118435},
issn = {09574174},
year = {2022},
date = {2022-12-01},
urldate = {2024-07-12},
journal = {Expert Systems with Applications},
volume = {210},
pages = {118435},
abstract = {Predictive Maintenance has become an important component in modern industrial scenarios, as a way to minimize down-times and fault rate for different equipment. In this sense, while machine learning and deep learning approaches are promising due to their accurate predictive abilities, their data-heavy requirements make them significantly limited in real world applications. Since one of the main issues to overcome is lack of consistent training data, recent work has explored the possibility of adapting well-known deep-learning models for image recognition, by exploiting techniques to encode time series as images. In this paper, we propose a framework for evaluating some of the best known time series encoding techniques, together with Convolutional Neural Network-based image classifiers applied to predictive maintenance tasks. We conduct an extensive empirical evaluation of these approaches for the failure prediction task on two real-world datasets (PAKDD2020 Alibaba AI OPS Competition and NASA bearings), also comparing their performances with respect to the state-of-the-art approaches. We further discuss advantages and limitation of the exploited models when coupled with proper data augmentation techniques.},
keywords = {Artificial Intelligence, Deep Learning, Failure prediction, Industry 4.0, Predictive Maintenance, Time series Encoding techniques},
pubstate = {published},
tppubtype = {article}
}
Predictive Maintenance has become an important component in modern industrial scenarios, as a way to minimize down-times and fault rate for different equipment. In this sense, while machine learning and deep learning approaches are promising due to their accurate predictive abilities, their data-heavy requirements make them significantly limited in real world applications. Since one of the main issues to overcome is lack of consistent training data, recent work has explored the possibility of adapting well-known deep-learning models for image recognition, by exploiting techniques to encode time series as images. In this paper, we propose a framework for evaluating some of the best known time series encoding techniques, together with Convolutional Neural Network-based image classifiers applied to predictive maintenance tasks. We conduct an extensive empirical evaluation of these approaches for the failure prediction task on two real-world datasets (PAKDD2020 Alibaba AI OPS Competition and NASA bearings), also comparing their performances with respect to the state-of-the-art approaches. We further discuss advantages and limitation of the exploited models when coupled with proper data augmentation techniques.
Ferraro, Antonino; Galli, Antonio; Gatta, Valerio La; Postiglione, Marco
A Deep Learning pipeline for Network Anomaly Detection based on Autoencoders Proceedings Article
In: 2022 IEEE International Conference on Metrology for Extended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE), pp. 260–264, IEEE, Rome, Italy, 2022, ISBN: 978-1-6654-8574-6.
Abstract | Links | BibTeX | Tags: Artificial Intelligence, Deep Learning
@inproceedings{ferraroDeepLearningPipeline2022,
title = {A Deep Learning pipeline for Network Anomaly Detection based on Autoencoders},
author = {Antonino Ferraro and Antonio Galli and Valerio La Gatta and Marco Postiglione},
url = {https://ieeexplore.ieee.org/document/9967598/},
doi = {10.1109/MetroXRAINE54828.2022.9967598},
isbn = {978-1-6654-8574-6},
year = {2022},
date = {2022-10-01},
urldate = {2024-07-12},
booktitle = {2022 IEEE International Conference on Metrology for Extended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE)},
pages = {260–264},
publisher = {IEEE},
address = {Rome, Italy},
abstract = {During the last decade, we have witnessed an ever-increasing growth of inter-connected devices (e.g. IoT, Cloud) and the security assessment of such networks has become more and more essential. Identifying network anomalies represents a promising strategy to detect network intrusions, thefts to users privacy, system damage and fraudulent activities. Thanks to their ability to learn complex anomalies patterns in a complete data-driven fashion, deep neural networks have recently received an increasing attention. However, the application of such techniques is constrained by the peculiar characteristics of network traffic data, which is very sparse and noisy — due to the high number of devices generating data and Internet applications — and suffer from a high imbalance, i.e. anomalies typically occur 0.001-1% of the time. In this work, we handle the above-mentioned challenges with a simple pipeline: first, we identify samples with anomalous behavior by means of an autoencoder (AE); then, an attack classifier is used to assign anomalies to their attack type. We experiment our framework on a million-scale dataset of real-world network traffic data for anomaly detection, showing promising performance in terms of Precision, Recall and F1 scores.},
keywords = {Artificial Intelligence, Deep Learning},
pubstate = {published},
tppubtype = {inproceedings}
}
During the last decade, we have witnessed an ever-increasing growth of inter-connected devices (e.g. IoT, Cloud) and the security assessment of such networks has become more and more essential. Identifying network anomalies represents a promising strategy to detect network intrusions, thefts to users privacy, system damage and fraudulent activities. Thanks to their ability to learn complex anomalies patterns in a complete data-driven fashion, deep neural networks have recently received an increasing attention. However, the application of such techniques is constrained by the peculiar characteristics of network traffic data, which is very sparse and noisy — due to the high number of devices generating data and Internet applications — and suffer from a high imbalance, i.e. anomalies typically occur 0.001-1% of the time. In this work, we handle the above-mentioned challenges with a simple pipeline: first, we identify samples with anomalous behavior by means of an autoencoder (AE); then, an attack classifier is used to assign anomalies to their attack type. We experiment our framework on a million-scale dataset of real-world network traffic data for anomaly detection, showing promising performance in terms of Precision, Recall and F1 scores.