Facebook will provide the open-source Torch library with its own top notch tools intended for deep learning. Torch specializes in deep learning and neural networks, as well as natural language processing networks and is widely used in order to create self-taught and automatic algorithms.
Since the field of AI(Artificial Intelligence) is growing, Facebook thinks it's about time to uncover its own tools for the area of deep learning. But what exactly is deep learning? It's a complex system that involves neural networks and information pertaining to audio-visual inputs, a system that also relies on a response mechanism.
As an example, deep learning is responsible for photo recommended tagging on Facebook.
Torch is almost 13 years old and its framework has developed greatly in the last years, becoming more and more used. Being used by big companies like Twitter, Intel or Google, Torch is also used as a research tool by academic communities.
There are several modules that Facebook will release through Torch, an important one being a module with a convolutional neural layer aided by custom kernels, in order to put the IT engineering field up to speed to the "new stuff".
Coming from Facebook, this should be big, especially after last year's breakthrough with facial recognition, when their software was tested in differentiating two images.
Their computers were able to tell if the second picture was showing the same person as the first picture with very high accuracy, close to an actual human being.
Soumith Chintala, a Facebook AI researcher, said that Facebook presents these kernels and make them available as open-source because they are way faster than anything prior to them.
Chintala brags about Facebook's layer code being as 23.5 times faster than other similar systems.
Facebook is known to provide software with open-source code in order to aid the IT field, this specific field of Facebook's research dating back to only 2013, lead by Yann LeCun, later joined by another so-called genius of the field, Vladimir Vapnik.
Coming from Facebook and being open-source, this is kind of a big deal that should heat things up in the AI area, maybe even getting new researchers to join the initiative.
