show image

Google wants to make machine learning mainstream

Google has unveiled a new machine learning tool that lets developers with no AI expertise apply the technology to image banks hosted in the cloud.

The product, AutoML Vision, has been designed to enable Google Cloud users to control how their images are categorised without having to write any code.

Google hopes it will appeal to businesses that want to capitalise on ML but are not able to hire from the small and coveted global pool of machine learning experts.

The firm’s engineers explained on a call with reporters yesterday that the service would be initially confined to imagery but that it would be expanded to translation, video and natural language processing in due course.

Fei-Fei Li, chief scientist for Google Cloud AI, said the firm started with vision because it is “one of the killer fields of ML application”: “It’s critical to many use cases, including the classification of products.”

The Zoological Society of London has been working with Google to develop the system to help it identify animals photographed in the wild, while Disney is using it to annotate its range of merchandise.

Google already offers building blocks to create machine learning models for specific tasks, but, unlike AutoML, they don’t let people customise their models using their own data.

“Cloud AutoML helps businesses with limited ML expertise start building their own high-quality custom models by using advanced techniques like learning2learn and transfer learning from Google,” Li and Jia Li, head of R&D at Google Cloud AI, said in a blog. “We believe Cloud AutoML will make AI experts even more productive, advance new fields in AI and help less-skilled engineers build powerful AI systems they previously only dreamed of.”

At the moment, the product is only available to developers, who have to apply for access as part of an alpha launch. Google is yet to reveal how much it will cost.