Description: A Technical Deep Dive and Demo of Kubeflow on Kubernetes within the ARL SCOUT HPC system. The first part of this training provides an introduction to machine learning and deep learning as well as an overview of Kubernetes, and shows how it is implemented on SCOUT via Kubespray. The second part of the training covers Kubeflow, a machine learning toolkit for Kubernetes, which provides an easy to use web interface for launching Jupyter notebooks on SCOUT, and provides pipelines for automating the end-to-end flow of a machine learning through the AI lifecycle. For example, Kubeflow provides support for notebook servers, distributed training, hyperparameter tuning, deploying models, and managing models in production. The final part of this training walks the user through a demo of using Kubeflow on SCOUT using three publicly available example notebooks: (1) a simple hello pipeline example, (2) an end-to-end monkey classification example, and (3) an end-to-end music genre classification example.

CUI