Good luck! (Score 1)

Neurones are remarkably complicated things, not simple at all. To model one neurone properly, you have to model each chemical reaction in it, and how these interact with other reactions, no mean feat. My lab works on the learning rules for these neurones in visual processing and, even with a monster like Blue Brain, we couldn't accurately model a single neurone. We rely on short cuts. We use mathmatical models which give fairly similar RESULTS, but using completely different METHODS. The daddy of all short cuts is vector summation and multiplication. We simulate the weights of different synapses as vectors and then calculate a model output vector. As a result we are drooling over the parallel vector processing units in the Cell. A vector is nothing like a synapse, one is mathematical, the other biological. The art is getting accurate mathematical models of the biological. I would not call our models accurate or sophisticated, but a simple bit of learning (eg. to recognise a cube from all angles) takes all night on our PCs, high-end linux x86's, for a few hundred neurones. I wish we could do 10,000. I wish we could do ONE well. A good brain simulation requires many billions of accurately modelled neurones. Such a thing does not exist, and probably never will with current technology. What it needs is huge clusters of truely parrallel processors running off the same cache. Why they have even bothered building this thing just before Cell processors hit the market is completely beyond me. I guess they will find out something, but their model will be surpassed by the first generation of Cell clusters doing the same thing.