Google launches new features for their AI platform.

2019-10-28

Google's AI platform is a cloud-hosted service that facilitates machine learning and data science workflows. Recently, it has gained a new feature leveraging powerful Nvidia graphics chip backend models, allowing users to run training scripts on any hardware.


For those new to machine learning, the AI platform enables developers to quickly and easily prepare, build, run, and share machine learning models in the cloud. With built-in data labeling services, they can annotate model training images, videos, audio, and text corpora through applied classification, object detection, and entity extraction. The hosted Jupyter Notebook service supports various machine learning frameworks, including Google TensorFlow, while the dashboard in the Google Cloud Platform console provides controls for managing, experimenting, and deploying models in the cloud or locally.


Now, AI Platform Prediction (a component of the AI platform that offers model serving for online predictions in a serverless environment) allows developers to choose from a set of machine types in the Google Compute Engine service to run models. Thanks to the new backend built on Google Kubernetes Engine, they can add graphics chips such as Nvidia T4 and enable AI Platform Prediction to handle provisioning, scaling, and serving. (Previously, "online prediction" only allowed selection from one or four vCPU machine types.)


In addition, prediction requests and responses can now be logged to Google's BigQuery for analysis to detect biases and outliers.As for AI Platform Training, it enables data scientists to run training scripts on various hardware without managing the underlying machines. It now supports custom frameworks, allowing researchers to launch any Docker framework so that they can train models in any language, framework, or dependency. Furthermore, AI Platform Training has gained Compute Engine machine types for training, enabling the selection of any combination of CPUs, RAM, and accelerators.


Cloud AI Platform product manager Henry Tappen wrote in a blog post, "Cloud AI Platform simplifies the training and deployment of models, allowing you to focus on using AI to solve the most challenging problems. From optimizing mobile games to detecting diseases to 3D modeling, businesses have been seeking new creative machine learning use cases. With the emergence of more inference hardware and training software, we look forward to seeing you tackle even more challenges with AI in the future."

RECENT AI NEWS

RECENT AI TOOLS