On Monday, Google announced that they have decided to open source their machine learning engine TensorFlow. This engine has been a major part of the many advances that Google has had in automated search as well as analysis, particularly in apps like Google Photos, and Google Search. Google has reported that the engine has improved speech recognition in the Google app by as much as 25%, it’s also formed the image search in Google Photos, which was announced this past summer at Google I/O. Additionally, it’s worked to make it easier to index pictures of cats, dogs and other things on YouTube, similar to what it’s done with Google Photos. So it’s a pretty big deal to be open sourcing this machine learning engine for everyone to use, edit and modify.
“Deep Learning has had a huge impact on computer science, making it possible explore new frontiers of research and to develop amazingly useful products that millions of people use every day. Our internal deep learning infrastructure DistBelief, developed in 2011, has allowed Googlers to build ever larger neural networks and scale training to thousands of cores in our datacenters.”
Google is much more than just a search engine, although arguably TensorFlow has definitely helped improve their search engine, especially when it comes to speech recognition and searching through photos, as we mentioned above already. Google has built large neural networks thanks to help from DistBelief in their datacenters. TensorFlow is said to be about twice as fast as DistBelief, and has much better scalability when compared to DistBelief. Google also notes that TensorFlow was built from the ground up. And by doing this, it means that the engine is built to be fast, portable ready for production service. Google also notes that you can seamlessly move an idea from training on your desktop GPU to running on your mobile phone. The company also plans to release their complete, to shelf ImageNet computer vision model on TensorFlow soon.
TensorFlow will be available as a standalone library a long with associated tools, tutorials and examples, all under the Apache 2.0 license. Which means you are free to use it anyway you like, thanks to that license. Including at colleges and Universities. It should be a great tool for institutions, and make a great future. You can find all of the links to GitHub and much more information in the blog post below, from Google. Or by going to tensorflow.org.