Secure Multi-Party Computation in Distributed Deep Learning Networks

The goal of this study is to present a novel strategy for improving the data privacy and security of distributed deep learning networks through the use of secure multi-party computation (SMPC). The approach in issue employs three key processes: privacy-preserving gradient descent (PPGD), secure aggregation (SA), and homomorphic encryption (HE). Several criteria are utilized to compare the efficacy of our proposed strategy to six industry-standard techniques, including accuracy, precision, recall, F1-score, training time, computational complexity, memory utilization, scalability, resilience, and privacy protection. The findings show that our method surpasses standard strategies in terms of protecting private information and sensitive data while retaining competitive model accuracy and training efficiency. This paper offers a complete architecture for secure collaborative model training and underlines the need for privacy-preserving strategies in distributed deep learning.