Learning is something that helps something to grow and mature into something more than what it previously was. Google has been developing ways for machines to learn, starting with DistBelief back in 2011. DistBelief was a first generation device that could do some amazing things, like learn a concept like “cat” from unlabelled YouTube images. As with all first generation devices, there were limitations to be figured out.
TensorFlow is a second generation development from Google that takes DistBelief to a different level. One issue with Distbelief was its reliance on Google’s internal infrastructure. This reliance made it near impossible to share research code with those on the outside. TensorFlow is released as open source; Google announced on the Google Research blog.
Google claims that TensorFlow is easy to use, portable, flexible and has improved speed, scalability and production readiness. It’s not like Deep Thought from Hitch Hiker’s Guide to the Galaxy, and it’s unlikely to be able to tell you the meaning of life. TensorFlow is capable of deep learning. TensorFlow uses the Python interface to give users the ability to express ideas that were previously difficult to compute.
Research using TensorFlow is helpful, but Google are proud of the readiness of TensorFlow for real products. The fact that TensorFlow is open source means that a wide range of people can contribute to the development of new ideas and ways to use TensorFlow. For those who want to take part, or just use TensorFlow, there will be a library of tutorials, examples and associated tools to help users get the most out of this technology.
At Google, TensorFlow is an important part of deep learning research experiments. Deep neural networks are infused with Google Search. Google have considered that digital technology changes rapidly, so growth and the future are planned in its design. Any institute will be able to use TensorFlow, as it uses the Apache 2.0 licence.
It’s hard to tell where technology will go, what uses people will find for developments like this or what unintended consequences will come about when more people start using and developing for TensorFlow. The user and TensorFlow will learn together, and information will be shared in the community of users. What question TensorFlow will be able to answer, what problems TensorFlow will be able to compute will be up to what is developed by users. Google is giving researchers and many others a powerful tool that could lead to something even more impressive in the future.