Google Cloud AutoML is a family of cloud services, announced on January 17, 2018, that lets organizations build custom machine-learning models without deep machine-learning expertise. The pitch was to democratize AI: a business with domain data but no in-house ML team could upload labeled examples, let Google’s systems train a model, and deploy it, rather than hiring specialists to design and tune networks by hand.
The first product, AutoML Vision, focused on image classification with a drag-and-drop workflow for uploading images, training a model, and deploying it on Google Cloud. Google said the service was built on its leading image-recognition methods, including transfer learning and neural architecture search, the same automated-design techniques developed by its research teams, and that models could be created in minutes for a pilot or brought to production within a day. The announcement cited early users such as Urban Outfitters, Disney, and the Zoological Society of London, which used it to identify wildlife species from camera-trap images. The lineup later expanded to natural language, translation, and tabular data.
For a business reader, Cloud AutoML was an early, prominent example of packaging cutting-edge research like neural architecture search into a service that non-experts can actually use.