ARM about to create connectionist's paradise

ARM is developing a new kind of architecture which is supposed to run a simulation of the human brain. We're not talking about an artificial intelligence here, the project is about  the reconstruction of the neuronal topology. In other words, they could prove that connectionist paradigm is right, that is, the brains cause the minds.

SpiNNaker, that's the name of the system, is based on a chip developed by Steve Furber, the father of ARM's 32-bit RISC processor. Every chip will contain 20 cores on a single die; one of them is designated as a "head" and the others run the "neuron simulation".

All cores are interconnected on the die itselft, and all processors (thousands in the final system) are interconnected through a "Globally Asynchronous Locally Synchronous" architecture. The simulated neurons can "fire" to any other neuron in about 1ms, which is practically the same of humans.

Obviously, Furber and his team are far from recreating the 90billion neurons topology of human brain; according to the well known Moore's law, in a couple of decades a full simulation would be possible but, meanwhile, better models of the brain could be studied and exploited.

Fascinating complexity

Here we go, another little step towards Master Degree! "Complex systems" is achieved and with full score, thanks to a crowd simulation based on Craig Reynolds' BOIDS and rendered by my own 3D engine "PLT".

I love emerging models - as one can easily foresee by reading the name of this blog - and BOIDS is no exception. A few rules and parameters, two constraints and a bit of randomness and we have a hell of algorithm, almost chaotic and showing a lot of unexpected and non-designed behaviours: formation of queues, spontaneous groups, and so on. Michele and I did a validation study, basing on the previous works of Crystals' Project, a joint effort in crowds simulation. Our model worked well, mathing the expected timings and behaviours, at least with reasonable agents' density; being BOIDS a simple reactive multi-agent system, our pedestrian do not embedded sufficient "intelligence" and can easily fall in many subtle cul-de-sac. It's a good result, anyway, and leaves a lot of space for further investigation.