With the aim of advancing the development of machine learning artificial intelligence (AI), Google has released its TensorFlow system as open source material.
Google is describing TensorFlow as the second generation of its machine learning system developed from DistBelief, with claims that the new iteration will improve upon the latter’s shortcomings.
“While DistBelief was very successful, it had some limitations,” said Google’s developers in a blog post. “It was narrowly targeted to neural networks, it was difficult to configure, and it was tightly coupled to Google’s internal infrastructure – making it nearly impossible to share research code externally.”
With TensorFlow, to develop a program’s ability to recognise images or translate languages, such as that seen in Google Translate, then a programmer will need a capable GPU to run and train its machine learning program.
Google said that any gradient-based machine learning algorithm will benefit from TensorFlow’s auto-differentiation and suite of first-rate optimisers using TensorFlow’s flexible Python interface, but will also accept C++.
And flexible it certainly appears to be as it has been designed to work across different scales of platform, such as being able to transition quickly between running on a desktop or laptop, down to a smartphone and back again.
From a time-consuming point of view also, Google claims that developers using it will find that they can both build and train their machine learning algorithms faster, at a rate of five times that which was seen with DistBelief.
Google are now using the system internally to develop Google Search with signals derived from their deep neural networks, but now the company will be able to tap into the wider engineering community to speed up the machine-learning process.
Machine learning image via Shutterstock