Originally posted by Aspa
If you by computational units mean neurons, there's 10 billion of them, each with an average of 1000 connections to other neurons. And neurons work more like tiny computers (IIRC networks of tubulin (sp?) proteins functioning as cellular automata) than simple binary switches.
Brain has 10 ^ 11 neurons (100 billion, 1 billion = 10^9 unless i'm mistaken), 10^14 synapses, 10^-3secs cycle time, 10^14 bits/sec bandwidth and 10^14 neuron updates/sec.
IIRC the brains learning process is a strengthening of certain pathways between neurons, where the firing of said pathways is the result of sensory input.
I can;t remember offhand if the brains learning process is different from a computational neural network - in the latter case learning is performed by tuning a weight (upon the inputs to the 'neuron', also aka perceptron for non-feedback NNs), until a desired output is achieved (or rather, to move closer to that desired output). Albeit I'm not sure how similar the brains own learning process is to this, it's been about 4 months or so since I studied it.