Efficient asynchronous federated neuromorphic learning of spiking neural networks

Spiking Neural Networks (SNNs) can be trained on resource-constrained devices at low computational costs. There has been little attention to training them on a large-scale distributed system like federated learning. Federated Learning (FL) can be exploited to perform collaborative training for higher accuracy, involving multiple resource-constrained devices. In this paper, we introduce SNNs into asynchronous federated learning (AFL), which adapts to the statistical heterogeneity of users and complex communication environments. A novel fusion weight based on information age and average spike rate is designed, which aims to reduce the impact of model staleness. Numerical experiments validate SNNs on federated learning with MNIST, FashionMNIST, CIFAR10 and SVHN benchmarks, achieving better accuracy and desirable convergence under Non-IID settings.