Nowadays, the generality of computing devices and the diversity of digital services constantly contribute to the generation and collection of huge amounts of data. Machine learning allows you to explore data in order to extract useful information. The unprecedented knowledge produced by this process can e extremely beneficial in many areas of application. However, in the case of sensitive data, the use of conventional methods of machine learning may result in privacy risks. To overcome this problem, Privacy Preserving Machine Learnng is proposed. It generally relies on cryptographic techniques such as homomorphic encryption or non-cryptographic techniques such as data perturbation.
The objective of this Ph.D. thesis is to propose machine learning methods that operate on encrypted data by providing end-to-end data protection
sensitive and extracted information from them via machine learning mechanisms in an efficient manner and which adapts to large amounts of data.