Riccardo Fida
Design, Implementation, and Experimental Evaluation of a Privacy-Preserving Federated Learning Framework for Medical Imaging Classification.
Rel. Carla Fabiana Chiasserini, Sergio Micalizzi. Politecnico di Torino, Corso di laurea magistrale in Ict For Smart Societies (Ict Per La Società Del Futuro), 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (10MB) | Preview |
Abstract
Privacy-Preserving Machine Learning (PPML) con Apprendimento Federato e Crittografia OmomorfaHealthcare is one of the most promising domains for the application of Artificial Intelligence (AI). However, the effective use of healthcare data is often limited by fragmentation across institutions, interoperability issues, and strict privacy requirements. Federated Learning (FL) has emerged as a collaborative learning paradigm that enables multiple organizations to train machine learning models without directly sharing sensitive patient data. Despite its advantages, FL remains vulnerable to privacy leakage and security threats that may compromise both data confidentiality and model integrity. This thesis investigates privacy-preserving federated learning in healthcare, focusing on the effectiveness of Differential Privacy (DP) and Homomorphic Encryption (HE) as mechanisms for enhancing privacy and security.
The study analyzes three representative threats: gradient inversion attacks, which attempt to reconstruct training samples from shared gradients; membership inference attacks, which seek to determine whether a specific sample participated in training; and backdoor attacks, in which malicious clients manipulate local data to alter the behavior of the global model
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
