FL-ODP: An Optimized Differential Privacy Enabled Privacy Preserving Federated Learning

Abstract

Privacy-preserving methods and techniques aim to safeguard the privacy of individuals and groups while facilitating data sharing for specific purposes. Federated Learning (FL) is a machine learning approach that allows multiple devices or systems to collaboratively train a model without directly sharing their data with each other or a central server. This paper proposes an Optimized Differential Privacy (ODP) approach to ensure the privacy of individual data points while enabling the extraction of valuable information. The proposed model is validated using the MNIST dataset and is analyzed with the FedAvg aggregator. Differential Privacy (DP) with FL is optimized by varying the noise and delta values. The analysis of data privacy is conducted in three phases: the first phase evaluates simple FL, and the second and third phases utilize different DP parameters to achieve optimized results. These optimized results show improved accuracy and privacy, confirming the efficacy of the proposed FL-ODP approach.

Keywords

• Federated learning
• differential privacy
• data privacy
• privacy preserved
• optimized differential privacy