Google Ups its AI Game
Google Cloud is rolling out an “AI Hub” supplying machine learning content ranging from data pipelines and TensorFlow modules. It also announced a new pipeline component for the Google-backed Kubeflow open-source project, the machine learning stack built on Kubernetes that among other things packages machine learning code for reuse.
The AI marketplace and the Kubeflow pipeline are intended to accelerate development and deployment of AI applications, Google said Thursday (Nov. 8). The new services follow related AI efforts such as expanding access to updated Tensor processing units (TPUs) on the Google Cloud.
The AI Hub is described as a community for accessing “plug-and-play” machine learning content. The clearinghouse also includes Jupyter notebooks, the open-source web application that allows developers to share web applications containing live code and other components. Together, the ML content, data pipelines and notebook are designed to accelerate deployment of AI applications in production.
The initial version of the hub provides content, pipelines and other machine learning resources developed by the search giant along private controls for sharing those resources. A beta version will include additional “asset types” along with more public ML content.
Meanwhile, Kubeflow includes standard tooling for building and managing complete AI pipelines. The open-source project launched last December now includes a “workbench” to compose and deploy machine learning workflows running on the Kubernetes cluster orchestrator, the company (NASDAQ: GOOGL) said in a blog post on Thursday (Nov. 8).
According to the Kubeflow project page, its goal is to provide a “way to deploy best-of-breed open-source systems for [machine learning] to diverse infrastructures. Anywhere you are running Kubernetes, you should be able to run Kubeflow.”
Kubeflow pipelines are available hereon GitHub.
Google said other companies contributing to Kubeflow include Cisco Systems (NASDAQ: CSCO) and Nvidia (NASDAQ: NVDA). The GPU leader is currently integrating its RAPIDS data science libraries into Kubeflow. The libraries use GPUs to accelerate data preparation and machine learning, the partners said.
Cisco’s contributions to the open-source project focus on applying machine learning models to enterprise infrastructure management, including hybrid and multi-cloud life-cycle management using AI tools.
In July, Google released its second-generation TPUs, custom ASICS designed for machine learning workloads. The TPUs are generally available to Google Cloud users. That was followed last month by the release of third-generation cloud TPUs along with the release on Google Cloud of PyTorch, the machine learning library for the Python programming language.
Google said this week it expects to PyTorch to be available soon for use on its cloud TPUs.
The company’s machine learning push accelerated in January with the release of its Cloud AutoML aimed at helping businesses build their own custom machine learning models.