In the world of machine-learning, there are few companies putting as much effort into its progression than Google. We learned just a few weeks ago about one of Google's "signals" called RankBrain that helps handle our most outlandish search requests, and today, we learn of TensorFlow, an important learning library that Google wants to share with the world.
In a new blog post, Google refreshes our memory about what wasn't possible just a couple of years ago. Trying to talk to your phone while on a busy sidewalk? Good luck. Translate a sign that's in a different language? Hah! In a very short amount of time, though, Google's machine- and deep-learning work has dramatically improved the situation in areas like these.
Ultimately, a software library called TensorFlow was born; one that's far more flexible and more scalable than Google's old system. The company notes that it can run just fine on a single smartphone or scale thousands of servers. Today, TensorFlow acts as a backbone for speech recognition, Smart Reply in Inbox, and also for sorting things out in Google Photos.
With TensorFlow, Google says that it can build and train neural networks up to five times faster than its first generation system, and given improvements to both software and hardware, that's sure to become even faster as time goes on.
Google admits that TensorFlow isn't perfect, and in fact, it might never be. However, it can always be improved with the right hands at the wheel, and for that reason, the company has decided to open source the library. With more people looking at what's in effect some extremely thorough R&D by Google, who knows what's possible? Google's chosen to host the project at GitHub, but the official website (linked below) has a wealth of information to peruse should you be interested in diving into working with TensorFlow.
With this move being made, we can't help but wonder if other companies will follow-suit with some of their own machine-learning projects. Time will tell.