Review: Google Cloud Vertex AI irons out ML platform wrinkles

Tracking ML metadata

Vertex ML Metadata (preview) describes your ML system’s metadata as a graph. It can store and track artifacts (pieces of data that ML systems consume or produce, including datasets, models, and logs), executions, events, and contexts. The stored metadata can help you understand your model’s lineage. Vertex ML Metadata integrates with Vertex Pipelines.

Running ML experiments

Vertex Experiments (preview) use Vertex TensorBoard, which is an enterprise-ready, managed version of TensorBoard, which is in turn a Google open source project for experimental visualization. The Vertex Experiments pane lists your experiments, your Vizier studies, and your TensorBoard instances.

Optimization with Vertex Vizier

Vertex Vizier is a black-box optimization service that helps you tune hyperparameters in complex machine learning models. Black-box optimization is the optimization of a system that either doesn’t have a known objective function to evaluate, or more commonly is too costly to evaluate by using the objective function, usually due to the complexity of the system.

Managing model features

Vertex Feature Store (preview) provides a centralized repository for organizing, storing, and serving ML features. By using a central feature store, your organization can efficiently share, discover, and re-use features at scale. Vertex Feature Store is similar to Amazon SageMaker Feature Store.

Performing vector similarity searches

Vertex Matching Engine allows you to perform vector similarity search so that you can perform efficient, accurate searches on large amounts of data. ML models transform data inputs, such as text and images, into embeddings that represent high-dimensional vectors. Vertex Matching Engine uses Google-developed algorithms, including the Two-Tower algorithm for matching pairs of items, the Swivel embedding pipeline, and Scalable Nearest Neighbors with Anisotropic Vector Quantization.

Exporting models

You can export AutoML Tabular classification and regression models to run on local servers in Docker. You can also export image and video AutoML Edge models to run in edge or mobile devices, Edge TPU devices, Docker containers, iOS and macOS devices, and in a browser or in Node.js; different devices require different model formats.

As promised, Vertex AI brings Google Cloud AutoML and Google Cloud AI and Machine Learning Platform together into a unified API, client library, and user interface. It also makes Google’s ML capabilities more competitive with Amazon’s and Azure’s offerings, which have been improving while Google rebuilt its software plumbing.

Although Vertex AI is a big improvement over what it replaced, it is by no means perfect. It’s not just that so many services are still in the preview phase. It’s also that there are so many more capabilities that Vertex AI could be offering, and so many improvements that could be made to its model building, user interface, model management, and automation of the machine learning lifecycle.

—

Cost: See Google Cloud Vertex AI pricing for details. The price structure is complicated, although the costs for Vertex AI remain the same as they are for the existing products that Vertex AI supersedes.

Platform: All services run on the Google Cloud Platform. A few kinds of models can be exported to run on-premises, in containers, or on edge or mobile devices.