Getting Your AI/ML Workloads Into the Kubeflow (DeveloperWeek Global 2020)

Learn how to simplify your AI/ML workloads with Kubeflow, an open-source platform for machine learning operations, built on Kubernetes, with features like centralized dashboard, training operators, and pipelines.

Key takeaways
  • The speaker introduces Kubeflow as a platform for machine learning operations, simplifying the process of deploying and managing machine learning models.
  • Machine learning workflows can be complex, involving data collection, data cleaning, feature engineering, model training, and deployment, and Kubeflow aims to streamline this process.
  • Kubeflow uses existing tools and frameworks, such as Jupyter Notebooks, TensorFlow, and scikit-learn, and provides a declarative interface for defining workflows.
  • The platform is built on top of Kubernetes and includes features like centralized dashboard, training operators, and pipelines.
  • Machine learning engineers can use Kubeflow to define workflows, automate tasks, and focus on model development, while data scientists can work with data and experiment with models without requiring infrastructure expertise.
  • The platform supports workflows that involve data streaming, data storage, data transformation, and feature engineering.
  • Kubeflow provides features like distributed training, hyperparameter tuning, and pipeline execution, making it easier to develop and deploy machine learning models.
  • The platform is scalable, flexible, and modular, allowing users to customize and extend its capabilities.
  • Kubeflow has a large community of contributors and users, with over 9,000 members and more than 5,000 community-driven projects.
  • The platform is widely adopted, with users from over 15 organizations in various industries, including finance, retail, and healthcare.
  • Kubeflow is open source and available on GitHub, with a growing ecosystem of contributors and integrations with other tools and frameworks.