Federated Learning with Differential Privacy on Personal Opinions: A Privacy-Preserving Approach
Mirwais Ahmadzai, Giang Nguyen · Procedia Computer Science · 2023
Data sharing poses a number of privacy risks, including reconstruction attacks, model inversion attacks, and membership inference attacks. Federated learning is a distributed paradigm that aims to improve privacy by sharing parameter gradients rather than the original data. The purpose of this study was to use federated learning and differential privacy techniques to train a model on sensitive personal opinion data while protecting the privacy of individual participants. The experiments carried out evaluate the effect of various noise levels on the model's performance. As the number of training rounds increases, the model's performance improves continuously, with noise multipliers and clients per round adjusted to achieve a balance between privacy and accuracy. This study adds to the literature on data analysis in scenarios with limited data sharing protocols and illustrates the ability to solve data privacy difficulties and restrictions efficiently using federated learning and differential privacy techniques.