Federated Learning Platform on Embedded Many-core Processor with Flower

In the emerging field of autonomous vehicle development, the role of artificial intelligence, particularly deep learning (DL), has garnered significant interest. This growing interest has led to extensive studies in using cameras and other onboard sensors for the critical task of object detection and recognition in these vehicles. A major concern, however, is the sensitive nature of the training data, which poses privacy risks when centralized on a server. Furthermore, as the demand for privacy protection increases, power consumption may increase rapidly. To address these issues, this paper proposes Federated Learning (FL) for DL on an embedded many-core processor. This study provides an implementation of FL, aiming to enhance privacy protection and energy efficiency in autonomous vehicle development. The experimental results of the proposed FL platform demonstrate that it offers significant improvements in processing speed and power consumption, suggesting an enhanced performance compared to existing edge devices.