Web hotspot news

<   Back to posts

Google has just created what is closest to the brain in any case, bringing together 16 000 processors in a "neural network". This neural network "intelligent" machines  consists of 16,000 processors connected to the Internet and where he quenches his thirst for knowledge from images taken by million in YouTube videos.

Thus, according to The New York Times, this "brain" has already taught himself to recognize a cat. Why a cat? Because there are thousands of online videos of cats probably, even if the system was also able to recognize a human body. As explained in this document (PDF), this "computer", with its simulation software for learning and the enormous amount of data available online, has recognized the cat among nearly 20 000 distinct objects, with an accuracy approximately 15.8% (an improvement of 70% compared to previous experiments).

To achieve this, researchers at Google and Stanford have chosen a different path. The idea is that instead of ensuring that researchers are trying to define the limits, you provide tons of data to the algorithm, you let the data speak for themselves and let the software learn itself, automatically. This software allows a computer to teach himself to recognize a cat in "inventing the concept of cat" from data in the manner of our "visual cortex".

A long way to go

As impressive as these results, all that remains for the moment very weak compared to the capabilities of the human brain. However, as said by researcher the modeling of the entire human visual cortex could be within reach before the end of the decade. "Researchers at Google have them more cautious. "It would be fantastic if it turned out that we only have to use the current algorithm but on a larger scale, but instinctively something tells me that we still do not have the proper algorithm," and stated DR Andrew Y. Ng.

Joomla SEF URLs by Artio