SortAlign: a score-based aggregation technique for neural networks in one-round federated learning

Abstract

In recent years, the growth of data generated on a daily basis in critical domains, such as industrial processes, where data privacy plays a key role, has led to the strong development of Federated Learning. In turn, the need for communication-efficient approaches has given particular importance to One-Round Federated Learning, where a central server coordinates the learning process of a global model using a federated network of clients, or nodes, in a single round of communication. In this study, a novel alignment strategy based on nodes similarity matching for Neural Networks in One-Round Federated Learning is proposed. This method was compared with various federated models and validated using a real-world use case of machining process.

Keywords

• Federated learning
• one-round federated learning
• deep learning
• machining process