Federated Learning in Distributed Networks: Leveraging Collaborative Machine Learning without Compromising Data Privacy

Abstract

This thesis is part of our end-of-cycle project to obtain a master's degree in network and data engineering. The project explores Federated Learning in Distributed Networks, specifically leveraging collaborative machine learning without compromising data privacy. We evaluated the FedGA-ICPS approach developed by Ms. Badra Guendouzi on various datasets to assess its performance in heterogeneous environments. Extensive testing was conducted on benchmarks such as Fashion MNIST, EMNIST, MNIST, and CIFAR -10, using models like Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs), Artificial Neural Networks (ANNs), and DenseNet. In addition, we tested two optimization algorithms, Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO), with the EMNIST and MNIST datasets. These tests were conducted to evaluate the effectiveness of these optimization techniques in improving model performance. The results underscore the versatility and efficacy of PSO and ACO in various scenarios, demonstrating their potential to enhance the accuracy and efficiency of our models within the FedGA-ICPS framework. The project implementation was done in Python, as it is the most suitable language for this type of project, offering rich libraries in the domain.