Research on Kubeflow Distributed Machine Learning

As the accuracy of machine learning improves, the depth of the model will become deeper and the amount of input data will become larger and larger, and calculation will be too large to complete with one computer. Also, multiple computers are required for gaining more speed. To achieve distributed machine learning, Kubeflow and Kubernetes can be better managed through a graphical interface. This research used a cluster and the Docker virtual containers, With Jupyter Notebook to do distributed training with TensorFlow, this study compare the impact of the difference in the computing speeds of the devices on the overall computing time while maintaining the same accuracy and analyze what speed can be achieved with different devices working together.