"HE- based Privacy preserving training of ML models through Microsoft SEAL
Abstract:
Homomorphic encryption is a form of encryption that allows specific mathematical operations to be performed on encrypted data without decrypting it first. This capability is particularly significant in the context of machine learning (ML) training, where data privacy is a critical concern. By applying homomorphic encryption to ML training, organizations can harness the benefits of cloud-based computation while preserving the confidentiality of sensitive data.
This abstract explores the concept of homomorphic encryption and its application to machine learning training. It begins by explaining the fundamental principles of homomorphic encryption, with practical examples implemented thrugh the Microsoft SEAL library. The talk then discusses how homomorphic encryption addresses these challenges by allowing ML models to be trained directly on encrypted data. This approach eliminates the need to share plaintext data with third-party service providers, thus mitigating the risk of data breaches and unauthorized access.
Speaker:
Gaia Anastasi has a grant at the University of Pisa, where she collaborates together with Prof. Gianluca Dini. Her research activities focus on Homomorphic Encryption and its application to Machine Learning.