AFRODITE - Optimizing Federated Learning in Resource-Constrained Environments Through Data Sampling Techniques
Mayara Aragão, Pedro Henrique González, Cláudio de Farias, Flávia C. Delicato · 2024
Federated Learning (FL) has emerged as a promising solution to address challenges in traditional machine learning (ML) regarding data privacy and security. However, training federated models in resource-constrained environments, such as IoT devices, presents challenges due to limited computational resources and complex data. This paper proposes data sampling techniques to optimize federated training in such environments, aiming to reduce training time while maintaining model quality. The study evaluates the impact of data sampling on federated model performance and compares it with traditional approaches. The methodology involves implementing random data selection in client datasets within the context of federated learning and conducting experiments across different configurations to analyze results. The findings provide insights for practical application in real-world scenarios with computational constraints.