Google releases machine learning software TensorFlow to the open source community

The Google logo at the headquarters in Mountain View, California.
The Google logo at the headquarters in Mountain View, California.PHOTO: AFP

TOKYO (JAPAN) - Some parts of the technology behind software giant Google's most popular apps have been released to the public to the first time, giving data scientists and programmers access to the company's machine learning software.

Early this week, Google released its machine learning system, called TensorFlow, as an open-source library for anyone to download and use to program.

Machine learning is the technology which powers Google's apps that require user interaction. It allows apps like Google Translate or Photos to "learn" and improve its functions by feeding it lots of data and refining the eventual result.

For instance, machine learning works in Google Photos when the app automatically tries to recognise and tags the pictures in a user's smartphone which contain cats as "cats". The app learns what an image of a cat is or is not when the user corrects the app if it mislabeled another animal as a cat. The app would then retain this information for future images, thus learning what a cat looks like and leading to better image recognition in the future.

TensorFlow would allow data scientists, app developers or the interested user to develop their own programs which make use of machine learning.

At a press event in Tokyo on Tuesday, Google executive chairman Eric Schmidt said machine learning is the beginning of a new future in programing, and that there are currently very few programers and computer scientists who understand machine learning.

"This is the beginning of a completely new thing," said Mr Schmidt. "Programers, just like everybody else, don't like change. When I was a programer, I was very good at figuring out algorithms and writing them all down... today, I think I would try to figure out how to program a computer to learn something. That's a completely different idea."

lesterh@sph.com.sg