Google’s evolution from consumer service to enterprise tech provider continues today with an expansion of its artificial intelligence services, called Cloud AutoML. The company is moving beyond services for data scientists towards point-and-click products for non-experts, beginning with an image-recognition tool called AutoML Vision. There are plenty of generic image-recognition models (including Google’s own) that can distinguish, say a shoe from a hat. But that’s not good enough for a retailer that sells many kinds of shoes, hats, or other gear. They need machine-learning models that are custom trained to recognize particular models and attributes of products. Urban Outfitters, for instance, is experimenting with AutoML Vision to train models for better search filters and recommendations on its site.

“Developing a custom model often requires rare expertise and extensive resources,” said Fei-Fei Li, Google Cloud AI’s chief scientist. So her team developed a drag-and-drop interface. Users can just upload labeled photos of the items they want to be recognize. Jia Li, Cloud AI’s head of R&D, says that a model can be created in a day with just a few examples of each item it needs to recognize. “The smallest quantity we’ve tried so far is like tens of images,” she says.


Related: A New Point-and-Click Revolution Brings AI To The Masses


This isn’t new, however. In 2016, for instance, a company called Clarifai (founded by a former Google intern) started offering a drag-and-drop model maker called simply Custom Training. Rajen Sheth, Google Cloud AI’s senior director of product management,…

[SOURCE]