Google celebrates one year of open sourcing machine learning framework TensorFlow

TensorFlow is now the most popular machine learning project posted on GitHub, with more than 3000 related repositories.

On 9 November last year, Google open sourced its machine learning framework, TensorFlow. Over the course of the year, Google added distributed computing features to TensorFlow and updated the framework with iOS support.

TensorFlow is now the most popular machine learning project posted on GitHub, with more than 3000 related repositories. Google uses TensorFlow to power many of its projects, including cutting edge natural language processing capabilities. TensorFlow is available as a cloud service on Google Cloud Platform.

Project Magenta by Google is an initiative to enable machines to create art. The eventual aim of project Magenta is to generate original video content, but for now the project is focusing on creating music. The DeepMind Artificial Intelligence by Google is powered by TensorFlow. After defeating world champion Go player, DeepMind is now learning to play Starcraft. In what could be baby steps towards an artificial general intelligence, DeepMind can learn things on its own.

TensorFlow works on Tensor Processing Units, specialised hardware created by Google specifically for TensorFlow to accelerate computing. At the Made By Google event, CEO Sundar Pichai heralded artificial intelligence as the most disruptive technological transformation for the decade ahead. After desktop computers, the internet, smartphones, it is now time for the advent of machine learning, and TensorFlow is at the heart of the machine learning capabilities by Google.

Google celebrates one year of open sourcing machine learning framework TensorFlow

Google CEO Sundar Pichai at the Made By Google Event in October

Ever since TensorFlow became open source, people around the world are finding innovative ways to use the framework. Australian marine biologists are using TensorFlow to monitor the population levels of the threatened Sea Cows. Originally, the population levels were monitored by humans in Sea Planes, a cumbersome process. Drones taking photos reduced the time needed, but required researchers to scan through thousands of photos to identify the sea cows. This is where TensorFlow was used to simplify the operations much further. A TensorFlow implementation was used to track Sea Cows that appear as small specs in a body of water.

find_the_sea_cow_solution.width-1000

TensorFlow spotting a solitary sea cow.

A cucumber farmer named Makoto Koike from Japan inspired by how well DeepMind managed to play Go, used TensorFlow to sort the cucumbers into different categories in his cucumber farm. Sorting is a time-consuming process, time that could be used more productively elsewhere on the farm. Makoto's mother would sort the cucumbers by hand before the AI-based sorting system was developed. Makoto says, "Farmers want to focus and spend their time on growing delicious vegetables. I'd like to automate the sorting tasks before taking the farm business over from my parents."

TensorFlow sorting cucumbers.

TensorFlow sorting cucumbers.

Data scientists in Silicon Valley hacked together a camera, a microphone, a Raspberry Pi and TensorFlow to physically track the erratic Caltrain. The device finds out when the train is passing, and at what speed and direction it is moving in. Radiologists have adapted TensorFlow to automate the detection of Parkinson's disease in medical scans.

The Tensor Processing Unit

The Tensor Processing Unit

TensorFlow is one of those projects that keeps getting better with time. There have been 10,000 code commits on GitHub after just a year of being open source. The Machine Learning community is contributing directly to the codebase. TensorFlow is available for use in everything from standalone products, to startups, to research initiatives, to school projects.

Loading...




also see

science