At the recent Google I/O 2021 conference, the cloud provider announced the general availability of Vertex AI, a managed machine learning platform designed to accelerate the deployment and maintenance of artificial intelligence models.
Using Vertex AI, engineers can manage image, video, text, and tabular datasets, and build machine learning pipelines to train and evaluate models using Google Cloud algorithms or custom training code. They can then deploy models for online or batch use cases all on scalable managed infrastructure.
The new service provides Docker images that developers run for serving predictions from trained model artifacts, with prebuilt containers for TensorFlow, XGBoost, and Scikit-learn prediction. If data needs to stay on-site or on a device, Vertex ML Edge Manager, currently experimental, can deploy and monitor models on the edge.
Vertex AI replaces legacy services such as AI Platform Data Labeling, AI Platform Training, and Prediction, AutoML Natural Language, AutoML Video, AutoML Vision, AutoML Tables, and AI Platform Deep Learning Containers.
Andrew Moore, vice president and general manager of Cloud AI at Google Cloud, explains why the cloud provider decided to introduce a new platform: “We had two guiding lights while building Vertex AI: get data scientists and engineers out of the orchestration weeds and create an industry-wide shift that would make everyone get serious about moving AI out of pilot purgatory and into full-scale production.”
In a separate article, Google explains how to streamline Machine Learning training workflows with Vertex AI, avoiding running model training on local environments like notebook computers or desktops and working instead with Vertex AI custom training service. Using a pre-built TensorFlow 2 image as an example, the authors cover how to package the code for a training job, submit a training job, configure which machines to use, and access the trained model.
The pricing model of Vertex AI matches the existing ML products that it will supersede. For AutoML models, users pay for training the model, deploying it to an endpoint, and using it to make predictions.
This article was originally published by InfoQ. Algoworks does not take any credit and is not responsible for the information shared in the article.