Fabio Giampaolo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Fabio Giampaolo has a total of 94 co-authors. Here are the top co-authors by number of shared publications:
- Francesco Piccialli
University of Naples Federico II, Naples, Italy
25 shared publications - Salvatore Cuomo
Department of Mathematics and Applications “R. Caccioppoli”, University of Naples Federico II, Naples, Italy
12 shared publications - Stefano Izzo
University of Naples Federico II, Naples, Italy
9 shared publications - Edoardo Prezioso
University of Naples Federico II, Napoli, Italy
8 shared publications Salvatore Cuomo
5 shared publicationsGiampaolo Casolla
4 shared publications
Latest Publications
- 25 August 2026
- Published by Springer Nature
- p. 250-267
FLIP4Tab: Federated Latent In-Context Priors for Tabular Data Synthesis
- 25 August 2026
- Published by Springer Nature
- p. 250-267
FLIP4Tab: Federated Latent In-Context Priors for Tabular Data Synthesis
- Open Access
- 21 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 320
MAKES-QA: A multi-agent framework for knowledge graph construction and enrichment over scientific literature for question answering
- 21 March 2026
- Published by Elsevier in Expert Systems with Applications
- Vol. 320
Open AccessMAKES-QA: A multi-agent framework for knowledge graph construction and enrichment over scientific literature for question answering
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP, 1-14
FuGuard: Client-Level Federated Unlearning via Generative Surrogates and Optimal Transport
Federated learning (FL) is a widely adopted paradigm that enables collaborative model training while preserving data privacy. As concerns around data poisoning and the "right to be forgotten" continue to grow, federated unlearning, which is the ability to remove the influence of specific training data from a trained FL model, has become increasingly critical. However, existing unlearning methods often require expensive retraining or fail to achieve good forgetting effects, limiting their practicality in real-world FL systems. In this work, we propose FuGuard, a dual-strategy federated unlearning framework, designed for efficient and ideal client-level data removal. FuGuard combines the generative surrogate, which approximates the contribution of the target client, with optimal transport regularization that softly constrains model parameter drift during unlearning. This approach effectively removes the influence of the target client while preserving the stability and performance of the global model. To evaluate the forgetting capability, we conduct testing using backdoor attacks and member inference attacks (MIAs) for residual data influence. Empirical results on different benchmarks demonstrate that FuGuard significantly reduces the impact of the target client's data while maintaining the performance of nontarget clients, consistently outperforming state-of-the-art baselines in both forgetting effectiveness and accuracy retention. Our code is accessible at: https://anonymous.4open.science/r/FuGuard-0263.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP, 1-14
FuGuard: Client-Level Federated Unlearning via Generative Surrogates and Optimal Transport
Federated learning (FL) is a widely adopted paradigm that enables collaborative model training while preserving data privacy. As concerns around data poisoning and the "right to be forgotten" continue to grow, federated unlearning, which is the ability to remove the influence of specific training data from a trained FL model, has become increasingly critical. However, existing unlearning methods often require expensive retraining or fail to achieve good forgetting effects, limiting their practicality in real-world FL systems. In this work, we propose FuGuard, a dual-strategy federated unlearning framework, designed for efficient and ideal client-level data removal. FuGuard combines the generative surrogate, which approximates the contribution of the target client, with optimal transport regularization that softly constrains model parameter drift during unlearning. This approach effectively removes the influence of the target client while preserving the stability and performance of the global model. To evaluate the forgetting capability, we conduct testing using backdoor attacks and member inference attacks (MIAs) for residual data influence. Empirical results on different benchmarks demonstrate that FuGuard significantly reduces the impact of the target client's data while maintaining the performance of nontarget clients, consistently outperforming state-of-the-art baselines in both forgetting effectiveness and accuracy retention. Our code is accessible at: https://anonymous.4open.science/r/FuGuard-0263.
- Open Access
- 18 November 2025
- Published by Elsevier in Neurocomputing
- Vol. 664
Towards one-shot federated learning: Advances, challenges, and future directions
- 18 November 2025
- Published by Elsevier in Neurocomputing
- Vol. 664
Open AccessTowards one-shot federated learning: Advances, challenges, and future directions
- Open Access
- 4 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 16-22
Deciphering One-Shot Federated Learning: The Pivotal Role of Pretrained Models
- 4 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 16-22
Open AccessDeciphering One-Shot Federated Learning: The Pivotal Role of Pretrained Models
- 20 September 2024
- Published by Seismological Society of America (SSA) in Seismological Research Letters
A Robust and Rapid Grid-Based Machine Learning Approach for Inside and Off-Network Earthquakes Classification in Dynamically Changing Seismic Networks
Earthquake location and magnitude estimation are critical for seismic monitoring and emergency response. However, accurately determining the location and the magnitude of off-network earthquakes remains challenging. Seismic stations receive signals from various sources, and it is crucial to quickly discern whether events originated within the area of interest. Location determination relies on obtaining ample P- and S-wave readings to ensure accurate and dependable results. Seismic networks vary due to station changes or outages, and their variable geometry represents a constraint for traditional machine learning models, which rely on fixed data structures. This study presents a novel approach for real-time classification of local and off-network earthquakes using the first three associated P picks within an early warning scenario, and also identifying the event’s direction. To handle variable network geometry, we employ a grid structure over the seismic area. The effectiveness of our method was initially validated with data from the Italian National Seismic Network, selecting Central Italy and Messina Strait subnetworks, and from a subnetwork of the Southern California Seismic Network; it achieves an inside–outside accuracy of 95%, 93%, and 96%, and a location region accuracy of 93%, 82%, and 97%, respectively. Its robustness was further demonstrated using picks from an earthquake early warning (EEW) system, the PRobabilistic and Evolutionary early warning SysTem (PRESTo) software, to simulate real and noncataloged input data. Our method outperforms PRESTo’s first localization, showing an inside versus outside classification improvement of 9.1% for Central Italy and 20.7% for the Messina Strait. This approach provides advanced seismic monitoring that can be implemented in systems devoted to reduce the impact of damaging events as the EEW system, but also shows promise for enhancing emergency response. Indeed, being able to quickly classify earthquakes is crucial for responding promptly and effectively during emergencies, minimizing risks, and for limiting false alarms.
- 20 September 2024
- Published by Seismological Society of America (SSA) in Seismological Research Letters
A Robust and Rapid Grid-Based Machine Learning Approach for Inside and Off-Network Earthquakes Classification in Dynamically Changing Seismic Networks
Earthquake location and magnitude estimation are critical for seismic monitoring and emergency response. However, accurately determining the location and the magnitude of off-network earthquakes remains challenging. Seismic stations receive signals from various sources, and it is crucial to quickly discern whether events originated within the area of interest. Location determination relies on obtaining ample P- and S-wave readings to ensure accurate and dependable results. Seismic networks vary due to station changes or outages, and their variable geometry represents a constraint for traditional machine learning models, which rely on fixed data structures. This study presents a novel approach for real-time classification of local and off-network earthquakes using the first three associated P picks within an early warning scenario, and also identifying the event’s direction. To handle variable network geometry, we employ a grid structure over the seismic area. The effectiveness of our method was initially validated with data from the Italian National Seismic Network, selecting Central Italy and Messina Strait subnetworks, and from a subnetwork of the Southern California Seismic Network; it achieves an inside–outside accuracy of 95%, 93%, and 96%, and a location region accuracy of 93%, 82%, and 97%, respectively. Its robustness was further demonstrated using picks from an earthquake early warning (EEW) system, the PRobabilistic and Evolutionary early warning SysTem (PRESTo) software, to simulate real and noncataloged input data. Our method outperforms PRESTo’s first localization, showing an inside versus outside classification improvement of 9.1% for Central Italy and 20.7% for the Messina Strait. This approach provides advanced seismic monitoring that can be implemented in systems devoted to reduce the impact of damaging events as the EEW system, but also shows promise for enhancing emergency response. Indeed, being able to quickly classify earthquakes is crucial for responding promptly and effectively during emergencies, minimizing risks, and for limiting false alarms.
- Open Access
- 20 June 2024
- Published by Elsevier in Journal of Network and Computer Applications
- Vol. 229
Synthetic and privacy-preserving traffic trace generation using generative AI models for training Network Intrusion Detection Systems
- 20 June 2024
- Published by Elsevier in Journal of Network and Computer Applications
- Vol. 229
Open AccessSynthetic and privacy-preserving traffic trace generation using generative AI models for training Network Intrusion Detection Systems
- 14 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 0651-0658
Graph Neural Networks in Emergency Management: A Case Study on Fire Call Assignment
This paper introduces a Graph Neural Network (GNN) approach to tackle the problem of emergency fire call assignment. The dataset utilized comprises emergency fire calls received by the 911 service in San Francisco. Each call is processed and assigned to a specific fire station. To capture the geographical nuances, we preprocess the dataset into a bipartite graph representation. The right nodes of the graph represent the calls with their geographical coordinates (latitude and longitude), while the left nodes signify the stations and their current occupancy. The graph edges represent the distances between calls and stations. Our GNN model predicts the optimal assignment of calls to stations based on unit availability. Instead of the conventional method of routing calls to the nearest station, our model bypasses this system to directly forecast the station with available units for each call. Through rigorous experimentation, we showcase the superiority of our method in enhancing the call assignment process. By leveraging the intrinsic graph structure, the GNN model taps into the rich information within the dataset. This study advances the field of emergency management by offering a groundbreaking optimization framework that can markedly improve resource allocation in emergency response systems.
- 14 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 0651-0658
Graph Neural Networks in Emergency Management: A Case Study on Fire Call Assignment
This paper introduces a Graph Neural Network (GNN) approach to tackle the problem of emergency fire call assignment. The dataset utilized comprises emergency fire calls received by the 911 service in San Francisco. Each call is processed and assigned to a specific fire station. To capture the geographical nuances, we preprocess the dataset into a bipartite graph representation. The right nodes of the graph represent the calls with their geographical coordinates (latitude and longitude), while the left nodes signify the stations and their current occupancy. The graph edges represent the distances between calls and stations. Our GNN model predicts the optimal assignment of calls to stations based on unit availability. Instead of the conventional method of routing calls to the nearest station, our model bypasses this system to directly forecast the station with available units for each call. Through rigorous experimentation, we showcase the superiority of our method in enhancing the call assignment process. By leveraging the intrinsic graph structure, the GNN model taps into the rich information within the dataset. This study advances the field of emergency management by offering a groundbreaking optimization framework that can markedly improve resource allocation in emergency response systems.
- 14 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 0738-0746
A Privacy Preserving Service-Oriented Approach for Data Anonymization Through Deep Learning
In the realm of safeguarding data privacy, particularly when it comes to sensitive personal information acquired via sensors, anonymization plays a crucial role. Within this paper, we present a novel architecture that operates on a service-oriented basis, enabling the real-time anonymization of such data. Our primary objective is to ensure that authorized users can access the data while upholding privacy standards. To achieve this, our architecture involves annotating the data upon ingestion, assigning privacy levels to groups of columns. The anonymization process relies on an autoencoder model which returns a lower-dimensional encoding of the original data. In our architecture, pretrained models are fetched from a library and updated according to a set of policies, or a new model is trained if the user requests a new set of columns. Our solution can manage any situation where anonymization policies must be managed on the fly, and provides a flexible approach to balancing privacy protection and data access. The proposed architecture can be a valuable tool for data infrastructures providers, researchers, and other organizations dealing with sensitive personal information.
- 14 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 0738-0746
A Privacy Preserving Service-Oriented Approach for Data Anonymization Through Deep Learning
In the realm of safeguarding data privacy, particularly when it comes to sensitive personal information acquired via sensors, anonymization plays a crucial role. Within this paper, we present a novel architecture that operates on a service-oriented basis, enabling the real-time anonymization of such data. Our primary objective is to ensure that authorized users can access the data while upholding privacy standards. To achieve this, our architecture involves annotating the data upon ingestion, assigning privacy levels to groups of columns. The anonymization process relies on an autoencoder model which returns a lower-dimensional encoding of the original data. In our architecture, pretrained models are fetched from a library and updated according to a set of policies, or a new model is trained if the user requests a new set of columns. Our solution can manage any situation where anonymization policies must be managed on the fly, and provides a flexible approach to balancing privacy protection and data access. The proposed architecture can be a valuable tool for data infrastructures providers, researchers, and other organizations dealing with sensitive personal information.