PRIVATE-AI: A Hybrid Approach to privacy-preserving AI
Siddanth Krishna, Siri S, Saif Kamalsha, S Amruth, Shruti Jadon · 2023
Machine learning algorithms based on Deep Neural Networks (NN) have achieved remarkable results and are being extensively used in different domains. A crucial enabler for Artificial Intelligence is the quantum of data, and these models are as good as the quality of data that is used to train them. However, gathering data and using them to prognosticate behaviours presents great challenges to the privacy of individuals and organizations, such as data breaches, privacy loss, and the corresponding financial and reputational damages. Privacy-preserving machine learning (PPML) aims at bridging the gap between preserving privacy and reaping the benefits of ML. It is a key enabler for privatizing collected data and complying with data protection regulations. The goal of the project is to implement a hybrid approach to Privacy Preserving AI by using these three Privacy Preserving techniques namely: Federated Learning, Differential Privacy, and Homomorphic Encryption to achieve maximum user data privacy while not affecting the overall accuracy of the predictions and computations made by the AI model.