New services, aimed at data scientists and ML engineers, to simplify Machine Learning Operations (MLOps) have been announced by Google. 

The thinking behind the new services is that Machine learning (ML) systems are tricky to manage - Google says they "have a special capacity for creating technical debt if not managed well".

This is because they have all of the maintenance problems of traditional code, then add their own ML-specific issues including unusual hardware and software dependencies, the need to test and validate data as well as code, and a tendency for models to degrade over time as data changes making the assumptions used by models less valid. Google says that in reality, creating an ML model is the easy part, and it's the operation and management of the ML model that's the hard part.

In practical terms, the services start with a fully managed service for ML pipelines that will be available in preview by October. This follows Google's announcement of a hosted offering for building and managing ML pipelines on its AI Platform earlier this year. The new managed service will let customers build ML pipelines using TensorFlow Extended (TFX’s) pre-built components and templates.

Alongside this, there's a Continuous Evaluation service on Google's AI platform that samples prediction input and output from deployed ML models, then analyzes the model’s performance against ground-truth labels.A new Continuous Monitoring service is being added that will monitor model performance in production to let you know if it is going stale, or if there are any outliers, skews, or concept drifts. This is designed to simplify the management of models at scale.

Underpinning the new services is Google's new ML Metadata Management service in AI Platform. This service lets AI teams track important artifacts and experiments, providing a curated ledger of actions and detailed model lineage.

The final part of the announcement is the addition of a Feature Store in the AI Platform that will act as a centralized repository of historical and latest feature values to make it easier to reuse information.

More Information

An Introduction to MLOps on Google Cloud

Architecture for MLOps using TFX, Kubeflow Pipelines, and Cloud Build

Related Articles

Google Open Sources Albert NLP

Google Provides Free Machine Learning For All

Google's Teachable Machine - What it really signifies

More Machine Learning Applied to Google Sheets

Get On The Machine Learning Bandwagon With Google

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info