Since most Machine Learning systems are fed with data that are sensitive and personal, issues related to data security and privacy preservation are critical in this area. However, it may be difficult or impossible to learn a model over encrypted data in an efficient way.
This kind of problem can be addressed using homomorphic encryption, a method that allows anyone to compute on encrypted data without the need of decrypting it. The primary objective of this project is to develop a homomorphic encryption toolbox for the construction of Machine Learning algorithms on encrypted data. This could be achieved by adapting existing homomorphic schemes or developing new approaches to this problem based on homomorphic encryption. Starting with simple learning algorithms, such as neural networks and decision trees, we will then extend these techniques to distributed applications like federated learning.
Datasets used for experiments
Classical UCI dataset