Communications of the ACM
Google Puts Its Virtual Brain Technology to Work
Google Fellow Jeff Dean
Credit: Ariel Zambelich/Wired
Google's recently developed learning software, which can identify specific objects by watching YouTube videos, is now being used to make Google's products smarter.
Google's learning software is based on simulating groups of connected brain cells that communicate and influence one another. When this type of neural network is exposed to data, the relationships between different neurons can change, giving the network the ability to react in certain ways to incoming data of a particular kind. Google now is using the neural network to improve speech-recognition technology. "We got between 20 and 25 percent improvement in terms of words that are wrong," says Google researcher Vincent Vanhoucke.
In addition, Google's image-search tools will benefit from the neural network. Google's self-driving car and Google Goggles also will improve from software that can make better sense of more real-world data. Meanwhile, Google researchers are testing models that understand both images and text together. "We are seeing better than human-level performance in some visual tasks," notes Google researcher Jeff Dean.
From Technology Review
View Full Article
Abstracts Copyright © 2012 Information Inc., Bethesda, Maryland, USA
No entries found