Facebook is sharing some of its most prized tools for developing deep learning projects, in an effort to establish itself as a pioneering player in artificial intelligence research. Potential benefits from AI innovations could be developed with the use of tools.
Facebook AI Research (FAIR) has announced that it is releasing as open-source its modules optimized for Torch.
Torch is a computing platform that specializes in deep learning and convolutional networks, which are used commonly for image- and video-recognition technologies, as well as networks specializing in natural language processing, which is used to create algorithms that train themselves to learn new information without a substantial amount of human assistance.
"These modules are significantly faster than the default ones in Torch and have accelerated research projects by allowing us to train larger neural nets in less time," said Soumith Chintala, AI researcher and software engineer at Facebook.
Chintala claimed that Facebook's modules ensure that the neural networks, which are based on processes in the human brain, that are at the center of deep learning systems can train up to 23.5 times faster than the fastest default modules publicly available today.
This will allow researchers to solve problems in a faster, more efficient manner than ever before, citing as an example a photo-recognition tool created by Facebook engineers that can tell whether people are standing, sitting, lying down or positioned in other ways in a photo.
"We benchmarked our code, and these are the fastest open source implementations out there," Chintala told Wired. "People didn't explore certain areas because they didn't think it was possible and now they are."
Torch is widely used in several academic laboratories as well as at other companies making their foray into AI and deep learning research, including Google, Twitter, Intel and NVIDIA.
Facebook has been using deep learning technologies on its platform for quite a while now. For instance, the News Feed uses deep learning to guess what posts from friends and pages the user would like to see. The feature that recognizes the faces in photos uploaded by users is also powered by Facebook's deep learning tools.
Facebook hopes to develop deep learning to power things such as a personal digital assistant that can distinguish if the user is posting a drunken photo of himself that he would rather not post in his sober state.
In 2013, Facebook cemented its stake in AI and deep learning by hiring Yann LeCun and Rob Fergus, two heavyweight AI researchers, to lead Facebook's AI efforts.