Pages Menu
Categories Menu

Posted by on Dec 15, 2012 in Science, Technology |

Spaun, functional brain simulation

Spaun, functional brain simulation

The simulated brain is a fascinating area in which neuroscience and mix high-level computer. We regularly receive new astonishing news simulations. Often the information is confusing and it is important to put everything in its place. Spaun is a simulator that first simulates the functionality of a complex brain. That is, try to do what a real brain. A very promising path.

One of the first attempts was the project Blue Brain , which is ongoing but with little news lately. The idea is to simulate each of the tiles or bricks of the brain called neocortical columns (comprising about 10,000 neurons). It is a long term project whose results we will see in the future.

IBM has made most neural to date. 500 billion neurons and 100 trillion synapses. However, computing is a project not of neuroscience. No such thing as a simulated human behavior. The long-term goal is to build chips neurosynaptic novedos not Von Neumann architecture.

A far less ambitious but more realistic is OpenWorm . Try to simulate the behavior of C. Elegans, a worm that we know its connectome , its 302 neurons and their synapses. First things simple has its advantages.

Gather a bunch of nueronas and expect something interesting emerges not seem a plausible way to understand how something as sophisticated as the brain

Spaun, the new simulator is novel and interesting. Along with OpenWorm line is functional simulations. This means that it is intended to simulate actual tasks that humans do.

To this end, 2.5 million simulated neurons grouped in brain areas that really we humans. Subnets simulator anatomical areas correspond to physical and visual area in the occipital lobe, the motor area, executive functions in the frontal lobe and memory. It also simulates the basal ganglia, stewards of complex movement (affected in Parkinson’s).

The model receives a 28×28 pixels image as input, processes the signal and generates an output using a simulated arm. Individual neurons communicate by electrical signals (action potentials) and neurotransmitters. Though simplified , the model captures many aspects of neuroanatomy, neurophysiology and psychological behavior.

This is demonstrated in 8 different tasks. For example: copy a picture, recognize numbers, sorts, counts them, and even complete a list (2,3,4 …). A set of videos is available to learn more about Spaun. Some of the tests are taken from conventional intelligence tests such as Raven’s Matrices test. Spaun rate almost like a human.

Until now, a simulation run was obtained from human brain size. From now on, the challenge is to get the largest animal biological functions and behaviors. For now, Spaun is the winner.

The behavior is similar to humans. Plays similar errors and response times similar (more complex tasks are more complex execution times).

If we destroy parts of the model, we can see what behaviors fail. Or we can change the functioning of neurotransmitters and see how it correlates with behavior.

So you can see what happens in the elderly with cellular aging.

They are working on a program that does not have specific instructions but positive or negative reward. This is a classic example of human learning.

Although the yield is expected near the actual response time, now takes hours in one second simulated.

Spaun is a jump in the simulation. Dazzling supercomputing exercises are not enough. We simulate actual behaviors. Spaun is on the way.

Tags: , ,