Google LLC today released a set of enhancements for its AI Platform that will give customers more flexibility in how they manage their artificial intelligence projects. 

AI Platform is a relatively new addition to the company’s enterprise portfolio. Introduced in April, the offering bundles a number of the AI services available on Google Cloud into a suite that covers every major step of machine learning development, from model creation to deployment. Today’s update improves upon two core components of the toolkit.

The first is AI Platform Prediction, a serverless AI hosting service that enables developers to run neural networks without having to manually manage the underlying infrastructure. Previously, there were only two hardware options: a model could be deployed on an instance with two virtual central processing units or four. Google is now adding over a dozen additional instance types with up to 32 vCPUs and introducing the ability to attach graphics cards for an extra speed boost.

To help developers monitor their deployments, the company is also rolling out new diagnostics capabilities. “You can now log your prediction requests and responses to BigQuery, where you can analyze them to detect skew and outliers, or decide if it’s time to retrain your model,” Henry Tappen, the project manager for AI Platform, wrote in a blog post.

The other service being updated is AI Platform Training. As the name suggests, it provides a managed environment where developers can hone newly built neural networks’ accuracy using sample data.

AI Platform Training is receiving more instance options too, along with an option for engineers to deploy custom Docker software containers on the platform. Containers enable applications to be installed on different kinds of infrastructure with little to no modification. In this case, developers can use the technology to bring  over their existing AI models and training scripts to Google Cloud without having to make any time-consuming code changes. 

Photo: SiliconANGLE