Google hasn’t been shy about sharing how it uses advanced neural networks (informally known as AI) in some of its products. The company has been teaching its machine learning tools a slew of new tricks in recent months. Google Photos uses it to easily find specific images based on your search, they equipped YouTube with the ability to better select thumbnails, reply to your emails from Gmail and made Google Translate far better at reading signs. And now, it wants to share its machine learning engine with developers, to make it even better…
Sundar Pichai, Google’s CEO announced in a blog post this morning that it has built an entirely new machine learning system called ‘TensorFlow’. It claims the network is faster, smarter and more flexible than its old one and can be adapted for use in more new products and research. Google describes it like this:
It’s a highly scalable machine learning system—it can run on a single smartphone or across thousands of computers in datacenters. We use TensorFlow for everything from speech recognition in the Google app, to Smart Reply in Inbox, to search in Google Photos. It allows us to build and train neural nets up to five times faster than our first-generation system, so we can use it to improve our products much more quickly.
At the moment, Google says the machine isn’t smart enough. It can’t even do things that a 4 year-old child can do, like recognize an object after only seeing it a couple of times. Or, understanding that ‘I saw the Grand Canyon flying to Chicago’ doesn’t mean that the Canyon itself was hurtling through the air towards the Windy City. By open sourcing the project, Google is hoping the work other researchers, engineers and hobbyists do can help speed the machine’s learning along and help get it to a much smarter level in less time.
If you want to join in and get involved with TensorFlow, head on over to TensorFlow.org to find out more.