• by Laurie Sullivan , Staff Writer @lauriesullivan, November 10, 2015

Google engineers have built an entirely new machine learning system dubbed TensorFlow, which replaces the old system. It allows Google to build and train neural networks up to five times faster than its first-generation system, so engineers can use it to improve Google products such as search and Gmail. TensorFlow will improve software as well as hardware, allowing devices with little processing power to return data and complete tasks more quickly.

TensorFlow is an interface for expressing machine-learning algorithms. Google built a library of the open-source code, and the initial release of this code -- which Google made available Monday -- aims to help developers build on technology. The goal, for Google, is to build an active open-source community that drives the future of the library, both by providing feedback and by actively contributing to the source code.

The system, according to Google, represents a significant shift in the way the company's technology processes information. "While DistBelief was very successful, it had some limitations," wrote Jeff Dean, senior Google Fellow, and Rajat Monga, Technical Lead, wrote 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."

The new system, TensorFlow, has the capability of running one smartphone or across thousands of computers in data centers, supporting everything from Google search, and speech recognition in the Google app, to Smart Reply in Inbox, and Google Photos. Using artificial intelligence, the system can improve on tasks that require a better understanding of perception and language, so software and hardware can learn each time it is used.

Deep learning is a brand of machine learning that is inspired by how the brain works. The technology trains computers to see and understand what's in an image of short video clip. for example, so it not only supports software, but hardware as well.

Google explains that TensorFlow is not an out-of-the box solution similar to a Web service that a developer can use in a product or service. It's an open-source project that developers can build on -- like feeding information into your brain to process.

It can provide better and faster search results and more accurate ad targeting. More often than not, when someone uses an app on their phone supported by deep-learning technology, it cannot run without sending information back to the data center to process. Google says in some cases using AI it can process the request on the phone without sending it back to the data center similar to what was requires of mainframe computers, where all the processing was done on a main machine. This is a difficult task, considering the limited processing power of phones.

Google hopes TensorFlow will make the process easier, which is one reason the company decided to open source the project so others can build on its success.