Sign In

Communications of the ACM

ACM TechNews

Why Facebook Is Teaching Its Machines to Think Like Humans


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
Artist's rendition of a robot with a human brain.

Facebook is trying to teach computers how to act more like humans, with the ultimate goal of learning more about individual users.

Credit: Universitt Bielefeld

Facebook is turning to deep learning to teach computers to more closely imitate the human brain, with the goal of gaining greater insight into individual users.

Natural language processing is one area in which Facebook hopes to advance, as its Graph Search tool released earlier this year is expanding to make everything a user does on Facebook, including posts and comments, searchable.

"Humans differ in the way they use language because of differences in their cultural upbringing," says text analytics company Semantria CEO Oleg Rogynskyy. "We still need to teach machines these nuances."

This fall, Facebook launched a deep learning research group, as Google, Microsoft, IBM, and Baidu also have done. Deep learning relies on neural networks, which are multi-layered software systems modeled on the brain that collect information and build an understanding of objects and words. Because neural networks can learn on their own, human engineering is not as important as with previous machine-learning methods, but vast quantities of data on which to train are essential.

The next step in deep learning will be to create algorithms that can better understand opinion, sentiment, and emotion. This technology will ultimately enable Facebook and other companies to target individual users in a very precise way to improve user experience, enhance brand loyalty, and sell products.

From Wired News
View Full Article

 

Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA


 

No entries found