Flo is a Developer Advocate at Cisco focusing on IoT, machine learning and network programmability. With a software and networking background he has been working since a couple of years on many IoT, ML and network automation projects. He is the most passionate about connecting things and getting information out of data in any way possible. In his current role, he is working on awesome showcases with Cisco and Open-Source technologies and providing learning content to the developer community.
MLOps 101: Building a ML Pipeline with Kubeflow
The code of a machine learning model is just a small part of the whole application or ML-based function. However, other components such as data sourcing, feature engineering, model development and model serving are vitally important as well. Therefore, this session will focus on the operations part known as MLOps and will provide an example of how an end-to-end ML architecture can be built with the machine learning toolkit Kubeflow, which leverages Kubernetes.
After a general overview of MLOps Open-Source tools and the MLOps architecture, a sample ML application will be presented. Then, the focus will be on Kubeflow and the ML workflow of this application. It will be shown how the data is being sourced and validated, and how the ML model is being developed with the specific Kubeflow components. The performance of the model, experiment handling and model serving will be elaborated and discussed.
After this session, the attendee will be more familiar with the architecture of end-to-end ML workflows and the capabilities of Kubeflow.