Bristech MLOps: Stefano Fioravanzo - Kubeflow + Rok + Kale (Sept 2020) скачать в хорошем качестве

Bristech MLOps: Stefano Fioravanzo - Kubeflow + Rok + Kale (Sept 2020)

Kubeflow + Rok + Kale: The easiest path to reproducible ML - Stefano Fioravanzo, Arrikto Video Chapters (please click on Chapter Heading to start playing from that time) 0:00 Introduction by Nic Hemley 0:57 Talk by Stefano 24:08 Q+A chaired by Nic Hemley Kubeflow is an open source toolkit for Machine Learning on Kubernetes, designed to make deployments of Machine Learning workflows on Kubernetes simple, portable, and scalable. It is an exponentially growing project and very popular among data scientists, with outstanding community and industry support. Using Kubeflow makes it easier for Data Scientists to automate and operationalize common Machine Learning workflows, like distributed training, hyperparameter tuning, running complex data pipelines, logging and storing metadata and artifacts, as well as working in shared JupyterLab environments. Kubeflow strives to provide all the bells and whistles for a comprehensive ML environment, but given the inherent complexity of running Machine Learning workflows at scale, Kubeflow remains more suited to Software/ML Engineers that possess a fair understanding of Kubernetes concepts, specific SDKs, and practices. In this talk, we will lower that complexity bar, taking Kubeflow’s MLOps paradigm to the next level, empowering Data Scientists to leverage on the Ops while focusing just on the ML. Easy Notebook deployments, reproducibility and collaboration are the key aspects to nail down for a seamless experience. Kale is the conversion engine that provides an integrated JupyterLab experience to deploy pipelines to Kubeflow. Rok is the storage and data management technology that takes care of making all your workflows portable, versioned, and reproducible. When put together, they allow for a dramatic improvement in the time and effort for scaling up Machine Learning workloads on-prem or in the Cloud! Caution: This talk’s contents are highly addictive, any extended use might cause distress, pain or anxiety towards other less innovative, disruptive and cumbersome technologies. Come and listen at your own risk. About Stefano Stefano is a Software Engineer at Arrikto. Previously he worked as a Research Software Engineer at Fondazione Bruno Kessler, an ICT research center leading AI research in Italy for 40 years, where he worked with brilliant ML researchers to democratize the use of Cloud-based infrastructure. He is now leveraging on this experience to build Cloud Native tools that will empower Data Scientists to scale up their workflows in the Cloud effortlessly, with a focus on reproducibility and collaboration across teams. Twitter: @sfioravanzo LinkedIn: Stefano Fioravanzo