Google Introduces Kubeflow Pipelines and AI Hub to Further Democratize AI
by Adrian Bowles
Google Cloud announced two new services on November 8th, 2018: Kubeflow Pipelines and AI Hub. This blog overviews what these services mean for the future of AI technology.
Google Cloud Has a Strong Commitment to AI
“Google is an AI company” and Google Cloud is on a mission to “make every company a machine learning company.”* Google has gone all-in to leverage artificial intelligence (AI) in their own products and services, and they have made a significant commitment to developing tools that help their customers leverage AI, too. From AutoML to TPU hardware, Google now offers support to help customers improve their AI journey, regardless of their starting point or their skillset.
What’s New in Google Cloud?
Google Cloud’s 11/8 announcement of new services—Kubeflow Pipelines and AI Hub—is consistent with this mission and represents two significant advances on their AI-democratization journey.
Kubeflow Pipelines extends Kubeflow (Google-derived open-source project dedicated to simplifying deployment of ML workflows on Kubernetes). Kubeflow Pipelines is essentially a workbench environment that allows users to package ML workflow code into components that can be composed and reconfigured and deployed using Kubernetes. This type of environment encourages experimentation and makes it much easier to update models or parts of a complete ML workflow as conditions or data changes. Kubeflow Pipelines represents a significant and open advance to ML-driven development.
AI Hub is a repository of machine learning (ML) content, including pipelines, Jupyter notebooks (Jupyter is an open-source project supporting data science and scientific computing R&D) and TensorFlow modules. The initial version of the Hub has been populated with content from Google Cloud AI, Google Research, and other groups and is available to the public. Enterprises can also create private, secure instances of the Hub to share ML resources internally. This allows firms to create their own ML resources as reusable components, which should speed development, lower costs, and improve the reliability and quality of ML-driven applications.
The AI Hub as it exists today is a great start. We expect it to become an invaluable resource in the future as it inevitably becomes a marketplace for ML content/components.
What’s Next for Google Cloud?
Google Cloud’s efforts to democratize AI, along with a commitment to make ML application components reusable and composable, promises to change the way the market develops AI-based applications.
Just as the structured development movement of the 1970s and 1980s advanced modular software development, and the object-oriented movement of the 1980s to 1990s refined and extended those constructs to focus on modular units with associated behaviors, Google is now at the forefront of a new application development model with composable ML components at its core.
*From SWOT Analysis: Google Cloud’s Artificial Intelligence Strategy, publishing soon from Aragon Research.