Riccardo Taiello
Research Subjects
Riccardo Taiello has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Melek Önen
EURECOM
6 shared publications - Marco Lorenzi
Epione Research Group, Inria Center of Université Côte d’Azur, Nice , France
5 shared publications - Clémentine Gritti
CITI Lab, INSA Lyon, Building Hedy Lamarr, 6 Avenue des Art, 69100 Villeurbanne, France
2 shared publications Marco Lorenzi
Epione Research Group Inria Sophia Antipolis France
2 shared publicationsOlivier Humbert
Université Côte d'Azur France
2 shared publications- Olivier Humbert
Centre Antoine-Lacassagne
2 shared publications
Latest Publications
- 8 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Privacy-preserving machine learning for large-scale collaborative healthcare data analysis
Apprentissage automatique sécurisé pour l'analyse collaborative des données de santé à grande échelle: Cette thèse de doctorat explore l'intégration de la préservation de la confidentialité, de l'imagerie médicale et de l'apprentissage fédéré (FL) à l'aide de méthodes cryptographiques avancées. Dans le cadre de l'analyse d'images médicales, nous développons un cadre de recalage d'images préservant la confidentialité (PPIR). Ce cadre aborde le défi du recalage des images de manière confidentielle, sans révéler leur contenu. En étendant les paradigmes de recalage classiques, nous incorporons des outils cryptographiques tels que le calcul multipartite sécurisé et le chiffrement homomorphe pour effectuer ces opérations en toute sécurité. Ces outils sont essentiels car ils empêchent les fuites de données pendant le traitement. Étant donné les défis associés à la performance et à l'évolutivité des méthodes cryptographiques dans les données de haute dimension, nous optimisons nos opérations de recalage d'images en utilisant des approximations de gradient. Notre attention se porte sur des méthodes de recalage de plus en plus complexes, telles que les approches rigides, affines et non linéaires utilisant des splines cubiques ou des difféomorphismes, paramétrées par des champs de vitesses variables dans le temps. Nous démontrons comment ces méthodes de recalage sophistiquées peuvent intégrer des mécanismes de préservation de la confidentialité de manière efficace dans diverses tâches.Parallèlement, la thèse aborde le défi des retardataires dans l'apprentissage fédéré, en mettant l'accent sur le rôle de l'agrégation sécurisée (SA) dans l'entraînement collaboratif des modèles. Nous introduisons "Eagle", un schéma SA synchrone conçu pour optimiser la participation des dispositifs arrivant tardivement, améliorant ainsi considérablement les efficacités computationnelle et de communication. Nous présentons également "Owl", adapté aux environnements FL asynchrones tamponnés, surpassant constamment les solutions antérieures. En outre, dans le domaine de la Buffered AsyncSA, nous proposons deux nouvelles approches : "Buffalo" et "Buffalo+". "Buffalo" fait progresser les techniques de SA pour la Buffered AsyncSA, tandis que "Buffalo+" contrecarre les attaques sophistiquées que les méthodes traditionnelles ne parviennent pas à détecter. Cette solution exploite les propriétés des fonctions de hachage incrémentielles et explore la parcimonie dans la quantification des gradients locaux des modèles clients. "Buffalo" et "Buffalo+" sont validés théoriquement et expérimentalement, démontrant leur efficacité dans une nouvelle tâche de FL inter-dispositifs pour les dispositifs médicaux.Enfin, cette thèse a accordé une attention particulière à la traduction des outils de préservation de la confidentialité dans des applications réelles, notamment grâce au cadre open-source FL Fed-BioMed. Les contributions concernent l'introduction de l'une des premières implémentations pratiques de SA spécifiquement conçues pour le FL inter-silos entre hôpitaux, mettant en évidence plusieurs cas d'utilisation pratiques.
- 8 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Privacy-preserving machine learning for large-scale collaborative healthcare data analysis
Apprentissage automatique sécurisé pour l'analyse collaborative des données de santé à grande échelle: Cette thèse de doctorat explore l'intégration de la préservation de la confidentialité, de l'imagerie médicale et de l'apprentissage fédéré (FL) à l'aide de méthodes cryptographiques avancées. Dans le cadre de l'analyse d'images médicales, nous développons un cadre de recalage d'images préservant la confidentialité (PPIR). Ce cadre aborde le défi du recalage des images de manière confidentielle, sans révéler leur contenu. En étendant les paradigmes de recalage classiques, nous incorporons des outils cryptographiques tels que le calcul multipartite sécurisé et le chiffrement homomorphe pour effectuer ces opérations en toute sécurité. Ces outils sont essentiels car ils empêchent les fuites de données pendant le traitement. Étant donné les défis associés à la performance et à l'évolutivité des méthodes cryptographiques dans les données de haute dimension, nous optimisons nos opérations de recalage d'images en utilisant des approximations de gradient. Notre attention se porte sur des méthodes de recalage de plus en plus complexes, telles que les approches rigides, affines et non linéaires utilisant des splines cubiques ou des difféomorphismes, paramétrées par des champs de vitesses variables dans le temps. Nous démontrons comment ces méthodes de recalage sophistiquées peuvent intégrer des mécanismes de préservation de la confidentialité de manière efficace dans diverses tâches.Parallèlement, la thèse aborde le défi des retardataires dans l'apprentissage fédéré, en mettant l'accent sur le rôle de l'agrégation sécurisée (SA) dans l'entraînement collaboratif des modèles. Nous introduisons "Eagle", un schéma SA synchrone conçu pour optimiser la participation des dispositifs arrivant tardivement, améliorant ainsi considérablement les efficacités computationnelle et de communication. Nous présentons également "Owl", adapté aux environnements FL asynchrones tamponnés, surpassant constamment les solutions antérieures. En outre, dans le domaine de la Buffered AsyncSA, nous proposons deux nouvelles approches : "Buffalo" et "Buffalo+". "Buffalo" fait progresser les techniques de SA pour la Buffered AsyncSA, tandis que "Buffalo+" contrecarre les attaques sophistiquées que les méthodes traditionnelles ne parviennent pas à détecter. Cette solution exploite les propriétés des fonctions de hachage incrémentielles et explore la parcimonie dans la quantification des gradients locaux des modèles clients. "Buffalo" et "Buffalo+" sont validés théoriquement et expérimentalement, démontrant leur efficacité dans une nouvelle tâche de FL inter-dispositifs pour les dispositifs médicaux.Enfin, cette thèse a accordé une attention particulière à la traduction des outils de préservation de la confidentialité dans des applications réelles, notamment grâce au cadre open-source FL Fed-BioMed. Les contributions concernent l'introduction de l'une des premières implémentations pratiques de SA spécifiquement conçues pour le FL inter-silos entre hôpitaux, mettant en évidence plusieurs cas d'utilisation pratiques.
- 27 April 2025
- Published by Springer Nature
- p. 19-41
Fed-BioMed: Open, Transparent and Trusted Federated Learning for Real-world Healthcare Applications
- 27 April 2025
- Published by Springer Nature
- p. 19-41
Fed-BioMed: Open, Transparent and Trusted Federated Learning for Real-world Healthcare Applications
- 17 January 2025
- Published by Springer Nature
- p. 204-214
Enhancing Privacy in Federated Learning: Secure Aggregation for Real-World Healthcare Applications
- 17 January 2025
- Published by Springer Nature
- p. 204-214
Enhancing Privacy in Federated Learning: Secure Aggregation for Real-World Healthcare Applications
- Open Access
- 30 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Let Them Drop: Scalable and Efficient Federated Learning Solutions Agnostic to Stragglers
Secure Aggregation (SA) stands as a crucial component in modern Federated Learning (FL) systems, facilitating collaborative training of a global machine learning model while protecting the privacy of individual clients’ local datasets. Many existing SA protocols described in the FL literature operate synchronously, leading to notable runtime slowdowns due to the presence of stragglers (i.e. late-arriving clients). To address this challenge, one common approach is to consider stragglers as client failures and use SA solutions that are robust against dropouts. While this approach indeed seems to work, it unfortunately affects the performance of the protocol as its cost strongly depends on the dropout ratio and this ratio has increased significantly when taking stragglers into account. Another approach explored in the literature to address stragglers is to introduce asynchronicity into the FL system. Very few SA solutions exist in this setting and currently suffer from high overhead. In this paper, similar to related work, we propose to handle stragglers as client failures but design SA solutions that do not depend on the dropout ratio so that an unavoidable increase on this metric does not affect the performance of the solution. We first introduce Eagle, a synchronous SA scheme designed not to depend on the client failures but on the online users’ inputs only. This approach offers better computation and communication costs compared to existing solutions under realistic settings where the number of stragglers is high. We then propose Owl, the first SA solution that is suitable for the asynchronous setting and once again considers online clients’ contributions only. We implement both solutions and show that: (i) in a synchronous FL with realistic dropout rates (taking potential stragglers into account), Eagle outperforms the best SA solution, namely Flamingo, by × 4; (ii) In the asynchronous setting, Owl exhibits the best performance compared to the state-of-the-art solution LightSecAgg.
- 30 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessLet Them Drop: Scalable and Efficient Federated Learning Solutions Agnostic to Stragglers
Secure Aggregation (SA) stands as a crucial component in modern Federated Learning (FL) systems, facilitating collaborative training of a global machine learning model while protecting the privacy of individual clients’ local datasets. Many existing SA protocols described in the FL literature operate synchronously, leading to notable runtime slowdowns due to the presence of stragglers (i.e. late-arriving clients). To address this challenge, one common approach is to consider stragglers as client failures and use SA solutions that are robust against dropouts. While this approach indeed seems to work, it unfortunately affects the performance of the protocol as its cost strongly depends on the dropout ratio and this ratio has increased significantly when taking stragglers into account. Another approach explored in the literature to address stragglers is to introduce asynchronicity into the FL system. Very few SA solutions exist in this setting and currently suffer from high overhead. In this paper, similar to related work, we propose to handle stragglers as client failures but design SA solutions that do not depend on the dropout ratio so that an unavoidable increase on this metric does not affect the performance of the solution. We first introduce Eagle, a synchronous SA scheme designed not to depend on the client failures but on the online users’ inputs only. This approach offers better computation and communication costs compared to existing solutions under realistic settings where the number of stragglers is high. We then propose Owl, the first SA solution that is suitable for the asynchronous setting and once again considers online clients’ contributions only. We implement both solutions and show that: (i) in a synchronous FL with realistic dropout rates (taking potential stragglers into account), Eagle outperforms the best SA solution, namely Flamingo, by × 4; (ii) In the asynchronous setting, Owl exhibits the best performance compared to the state-of-the-art solution LightSecAgg.
- Open Access
- 4 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 197-208
Buffalo: A Practical Secure Aggregation Protocol for Buffered Asynchronous Federated Learning
- 4 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 197-208
Open AccessBuffalo: A Practical Secure Aggregation Protocol for Buffered Asynchronous Federated Learning
- Open Access
Privacy Preserving Image Registration
Image registration is a key task in medical imaging applications, allowing to represent medical images in a common spatial reference frame. Current approaches to image registration are generally based on the assumption that the content of the images is usually accessible in clear form, from which the spatial transformation is subsequently estimated. This common assumption may not be met in practical applications, since the sensitive nature of medical images may ultimately require their analysis under privacy constraints, preventing to openly share the image content.In this work, we formulate the problem of image registration under a privacy preserving regime, where images are assumed to be confidential and cannot be disclosed in clear. We derive our privacy preserving image registration framework by extending classical registration paradigms to account for advanced cryptographic tools, such as secure multi-party computation and homomorphic encryption, that enable the execution of operations without leaking the underlying data. To overcome the problem of performance and scalability of cryptographic tools in high dimensions, we propose several techniques to optimize the image registration operations by using gradient approximations, and by revisiting the use of homomorphic encryption trough packing, to allow the efficient encryption and multiplication of large matrices. We demonstrate our privacy preserving framework in linear and non-linear registration problems, evaluating its accuracy and scalability with respect to standard, non-private counterparts. Our results show that privacy preserving image registration is feasible and can be adopted in sensitive medical imaging applications.
Open AccessPrivacy Preserving Image Registration
Image registration is a key task in medical imaging applications, allowing to represent medical images in a common spatial reference frame. Current approaches to image registration are generally based on the assumption that the content of the images is usually accessible in clear form, from which the spatial transformation is subsequently estimated. This common assumption may not be met in practical applications, since the sensitive nature of medical images may ultimately require their analysis under privacy constraints, preventing to openly share the image content.In this work, we formulate the problem of image registration under a privacy preserving regime, where images are assumed to be confidential and cannot be disclosed in clear. We derive our privacy preserving image registration framework by extending classical registration paradigms to account for advanced cryptographic tools, such as secure multi-party computation and homomorphic encryption, that enable the execution of operations without leaking the underlying data. To overcome the problem of performance and scalability of cryptographic tools in high dimensions, we propose several techniques to optimize the image registration operations by using gradient approximations, and by revisiting the use of homomorphic encryption trough packing, to allow the efficient encryption and multiplication of large matrices. We demonstrate our privacy preserving framework in linear and non-linear registration problems, evaluating its accuracy and scalability with respect to standard, non-private counterparts. Our results show that privacy preserving image registration is feasible and can be adopted in sensitive medical imaging applications.
- 1 March 2024
- Published by Elsevier in Medical Image Analysis
- Vol. 94, 103129
Privacy preserving image registration
- 1 March 2024
- Published by Elsevier in Medical Image Analysis
- Vol. 94, 103129
Privacy preserving image registration