Evolutionary Federated Learning Using Particle Swarm Optimization

Ender Minyard, Steven Kolawole, Nayan Saxena

Keywords: federated learning, evolutionary algorithms, particle swarm optimization, communication efficiency, resource-constrained clients, privacy-preserving machine learning, client-server architecture, TensorFlow Federated, Flower, accelerated particle swarm optimization, (de)-serialization of weights, privacy preservation

Abstract: Efficient communication is a key challenge in federated learning, where multiple clients contribute to a shared model. To address this issue, reducing local computation is an effective solution. This paper proposes an innovative federated learning algorithm that utilizes Particle Swarm Optimization, a powerful evolutionary algorithm, to minimize the computational demands on federated learning clients. Our results show that this algorithm results in significant enhancements in accuracy and faster convergence of loss compared to traditional federated learning methods.