Blue Brain Project, neocortex done.

Jeff7

Lifer
Jan 4, 2001
41,596
19
81
It'll be great when we have Windows AI, and it refuses to install updates because it insists that it's already perfect.


Maybe this is the ENIAC of AI systems capable of responding like an organic brain.
 

KingGheedora

Diamond Member
Jun 24, 2006
3,248
1
81
How did they accurately simulate 30 million connections? Or is it not a true to reality, neuron-for-neuron model?
 

Bill Brasky

Diamond Member
May 18, 2006
4,324
1
0
Weird. I heard a small lecture on Blue Brain this morning in class. Here are a few sentences from the power point for anyone that's interested. I've deleted the formatting so it looks like a paragraph, but the writing is choppy since it's supposed to be a talk.

"The Blue Column has ~10,000 neurons and ~5 x 106 synapses. The neurons have realistic geometries and are connected in a network that is itself determined computationally to optimize column calculations. Each neuron can have a unique shape and electrical properties. Each neuron is modeled as 10,000 compartments*, each with a combination of up to a dozen channels, also chosen computationally from a basic set of 20 ion channel genes. One way that the brain differs from a computer is that its basic elements are continually changing. Synaptic plasticity makes the brain ?more like a dynamically morphing computer?.

Rules for synaptic plasticity are included in the software that models neuronal activity. The simulations are run on a Blue Gene/L supercomputer with 65,000 CPU chips (each a 700 mHz processor) and a combined total of 32 TB of memory. It can operate at speeds of up to 360 teraFLOPS (1012 math operations/s). In a second phase of the project, applications developed in the first phase will be made freely available ?to allow a global and spontaneous effort to build software models of the brain?. "

*compartments = segments of dendrite separated by differences in ion channel mixture
 

NanoStuff

Banned
Mar 23, 2006
2,981
1
0
How did they accurately simulate 30 million connections? Or is it not a true to reality, neuron-for-neuron model?

What do you mean how? Big computer :)

I'm unsure how if at all they managed to get proper 'weight' of synapses, but progress is progress even if imperfect.