advertisement
advertisement

Facebook Makes Some Of Its Deep Learning Tools Available To Everyone

Facebook Makes Some Of Its Deep Learning Tools Available To Everyone
[Photo: Flickr user Robert Couse-Baker]

The tech giants are giving back to the community.

Not long after Samsung announced plans to help establish an open platform for the Internet of Things, Facebook has revealed its plans to share a selection of its groundbreaking Artificial Intelligence tools, related to “deep learning,” with anyone who wants access to them.

One of the most exciting and sought-after fields of AI, deep learning is a sub-category of what is called machine learning, using vast artificial brains called neural nets to improve things like speech recognition, computer vision, and natural language processing.

Facebook has been involved with some significant discoveries in deep learning, including last year’s facial recognition breakthrough, in which Facebook’s computers proved almost as accurate as a human in telling whether two different images showed the same person. It has also made some key appointments in the area, such as bringing on board former NYU professor Yann LeCun as its AI lab director.

Now Facebook’s Artificial Intelligence Research lab has introduced a set of modules for open source computing framework Torch, which represent considerable improvements on the online resource’s existing deep-learning algorithms in both speed and efficiency. The goal is to give researchers a way of grappling with bigger AI problems, which can benefit from state-of-the-art deep learning tools.

“We benchmarked our code, and these are the fastest open-source implementations out there,” Facebook artificial intelligence researcher and software engineer Soumith Chintala tells Wired. “People didn’t explore certain areas because they didn’t think it was possible, and now they are.”

Given how significant (and potentially life-changing) deep learning is as a tool in AI, hopefully more large companies will follow Facebook’s lead.

advertisement

[via Wired]

advertisement