Google Creates Learning Brain, Turns It Loose On The Internet
Michael Harper for redOrbit.com
Google is alive. Rather, they’ve connected 16,000 computers in a neural network, brought it to life, and has let it loose on the internet to “learn.”
Brought to you by the same secretive lab which brought us things like self-driving cars and Project Glass, this new project is designed to create a working simulation of the human brain, which means it will watch as it learns.
Of course, when you create a giant learning brain, the first thing you do is sit it in front of YouTube for days to study videos of cats. Now, the neural computer is able to recognize the face of a cat, and can even recall what a cat looks like in the form of a very creepy, mix-mashed kind of image.
The researchers responsible for this learning giant brain will take to Edinburgh, Scotland this week to present their creepy cat face, as well as their findings.
Though the internet already resembles a crazy cat ladies house, the team said they were surprised to see how well their “brain” performed.
According to a New York Times story, the team said the neural computer nearly doubled the accuracy with which it drew the cat face.
“We never told it during the training, ‘This is a cat,’ ” said Dr. Dean, the developer of the software used by this learning computer brain.
“It basically invented the concept of a cat. We probably have other ones that are side views of cats.”
The videos shown to the neural computer of more than 16,000 processors were selected at random.
The software used by these computers is also designed to mimic the design of the brain in that specific neurons can detect significant objects. Therefore, the fact that it was able to detect and redraw an image of a cat is surprising to the team. Normally, these types of experiments would be supervised by a team, pointing the learning computers in a specific direction. With their research, they were able to let the brain come up with the cat image all on its own.
“The idea is that instead of having teams of researchers trying to find out how to find edges, you instead throw a ton of data at the algorithm and you let the data speak and have the software automatically learn from the data,” Dr. Ng, a Computer scientist from Stanford who led the research, told John Markoff of the New York Times.
By pulling each of its neuron-like processors together, the computer than drew the image of the cat after being exposed to millions of cat images. The computer had learned. Shocked, the scientists said they had developed a “cybernetic cousin” to the actual human brain, closely replicating how our brain views and reconstructs images.
Referring to a “grandmother neuron,” the scientists have discussed the possibility of cells in the brain being trained to recognize a certain image or face when they are exposed to it long enough. In other words, “You learn to identify a friend through repetition,” said Gary Bradski, a neuroscientist in Palo Alto, California.
The learning brain computer was also able to learn and draw similarly creepy images of a human face and other body parts. While surprised at their accomplishments, Dr. Ng remains cautious about comparing software to biology.
“A loose and frankly awful analogy is that our numerical parameters correspond to synapses,” said Dr. Ng, saying despite their having 16,000 processors to do the heavy lifting, a brain has many many more connections inside.
“It is worth noting that our network is still tiny compared to the human visual cortex, which is a million times larger in terms of the number of neurons and synapses.”
Maybe, for their next trick, Google will release a Google Maps Street View of the connections in the human brain?