Compression de réseaux de neurones pour l'apprentissage fédéré


Federated Learning (FL) is a promising distributed method for edge-level machine learning, particularly for privacysensitive applications such as those in military and medical domains, where client data cannot be shared or transferred to a cloud computing server. The communication cost is a major challenge in FL due to its intensive network usage. Client devices, such as smartphones or Internet of Things (IoT) nodes, have limited resources in terms of energy, computation, and memory. To address these hardware constraints, lightweight models and compression techniques such as quantization and pruning are commonly adopted. In this study, we investigate the impact of compression techniques on FL for image classification using CIFAR10 and a ResNet-12 architecture. Our experimental results demonstrate the effectiveness of compression techniques in reducing communication costs while maintaining reasonable accuracy.

XXIXème Colloque GRETSI