Jump to content

Google starts implementing highly-distributed neural networks in its products


Recommended Posts

Posted

The linked article even contains a decent layman explanation of neural networks.

 

I found it really interesting that Google's NNs are self-taught:

 

Google's engineers have found ways to put more computing power behind the approach than was previously possible, creating neural networks that can learn without human assistance and are robust enough to be used commercially, not just as research demonstrations.

 

Apparently Google's voice recognition software is a lot more accurate with this approach, and they plan to include this highly-distributed NN approach in their other products.

 

"We got between 20 and 25 percent improvement in terms of words that are wrong," says Vincent Vanhoucke, a leader of Google's speech-recognition efforts. "That means that many more people will have a perfect experience without errors." The neural net is so far only working on U.S. English, and Vanhoucke says similar improvements should be possible when it is introduced for other dialects and languages.

 

http://www.technologyreview.com/news/429442/google-puts-its-virtual-brain-technology-to-work/?a=f

 

From memory, the enteric nervous system (our gut brain) is about 100 million neurons, which would make our thinking brains around 1 billion neurons (IIRC the enteric nervous system contains one tenth the number of neurons as the brain). I wonder how many neurons Google's NN has?

 

Hmm, I was wrong by an order of magnitude. The human brain has around 20 billion neurons (and the enteric nervous system has 200 to 600 million, which is similar to the number of neurons in the spinal cord neural network as well): http://en.wikipedia.org/wiki/List_of_animals_by_number_of_neurons#Cerebral_cortex

 

The roundworm has only around 300 neurons in its ENTIRE nervous system! I'd love to see that network. Surely neuroscientists have pulled it apart?

 

Even the Drosophila has 'only' around 100,000 in its entire system, and yet it is capable of performing extremely complex ballistic predictions and aerodynamic computations (think of take-off and landing), as well as obviously image and smell recognition.

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...