Experimental results of the "Ok-Topk-SP: A Novel Sparse Allreduce for Distributed Deep Neural Network Training on CPUs" article

This repository contains the experimental results presented in the article: “Ok-Topk-SP: A Novel Sparse Allreduce for Distributed Deep Neural Network Training on CPUs”, submitted to The 24th IEEE International Symposium on Parallel and Distributed Computing (ISPDC 2025). The experiments focus on the training of a VGG16 model using the CIFAR-10 dataset. The objective is to evaluate the performance and effectiveness of the proposed Ok-Topk-SP optimiser in comparison with Ok-Topk and the standard SGD baseline in the PyDTNN framework.