Facebook makes pattern recognition algorithms in images open source

Facebook has made software intended for machine learning available under an open source license. Among other things, the tools should accelerate machine learning algorithms for pattern recognition in images and videos.

On its Research blog, Facebook announced that it would donate the tools to Torch, a framework mainly used by scientists and intended for machine learning and building ‘neural’ networks. Facebook claims that the software it makes available under an open source license can significantly accelerate the machine learning capabilities of algorithms. Scientists and application developers, among others, could benefit from the tools, according to Facebook.

Facebook states that its tools can be up to 23 times faster than what is currently built into the Torch framework. This mainly concerns algorithms that are used for so-called convolutional neural networks, which are inspired by biological processes to recognize images: the algorithms must therefore be able to detect patterns in images and videos.

Facebook promises that it will provide the tools with updates in the near future. The social networking site probably already uses such software itself to analyze content posted by users.