X-Message-Number: 25806 From: Date: Wed, 16 Mar 2005 02:28:59 EST Subject: Uploading technology (1). Uploading technology (1). I come back on this contreversial subject. The aim here is to set the basis for a logical thinking about that possibility. There are three questions: - What are we uploading ? ie: What is a neural network in a biological brain ? - How to model it into a mathematical framework ? - What are the technologies able to implement this model ? In this first part, I define what I understand as a "generic neuron", if some reader think I have some gross misconception about it, any comment would be welcome. There may not be in any brain a "generic neuron", but I think the pyramidal ones could be near that definition. Outside very special cases, a neuron has 3 elements : An input domain in the form of a dendritic tree, a body or soma and an exit system formed by a single axon. The axon divides into small terminal elements near the dendritic spine of another neuron. The gap between them is the synapse. A run of the mill pyramidal neuron has on the order of 8 000 dendritic spines or input elements. In the cerebellum, Purkinje neurons implied in memory, have more than 100 000 such entries. In a synapse, the electrochemical signal or action potential, at the axon end command the release of a neurotransmitter storred in small vesicles. A single action potential releases one or some vesicles, up to 6 or 7. Each vesicle contains something as 3000 to 10 000 molecules. On the dendritic side of the synapse there are at most some hundreds of receptors. This system works as an one way channel, from axon terminal to the dendrite spine and an amplifier. A single signal in the axon produce tens or more receptor hits in the dendrite. This system is quantized on the axon side by the number of vesicles released and on the dendrite spine side by the receptor number. "Quantized" must be understood here as defined in discrete units, this is not quantum mechanics. The transmission time in a synapse is on the order of some milliseconds, 3 to 5 in most case, up to 20 sometime. Here are other receptors on the spine side with far longer effects in the seconds or minutes range or more. Another kind of communication is the electrical gap junction between cells. These are channels going from a cell to another. A junction channel allows the exchange of small ions or molecules. In most case the link is bidirectional, there is no amplification. Gap channels seem most common in glial cell and in neurons of the brain fetus or newborn. Here, the main function seems to be linked to the neural differentiation process. In the slow processes, some can work backward from the dendrite to axon side. This is the case of the nitrogen monoxyde gas, NO for example. What to modelize ? First, here are a lot of function linked to the "technology" used, ie biochemistry. Most gap juctions for example seem to fall in this domain. Another technology, electronics for example, will need not to bother with it, but may need it own technology defined support fuctions. There is no significant signal shorter than 2 ms. From the Shanon theorem, it is then sufficient to sample the processus at two time that frequency to recover all the information. A brain model can then be time quantized at the millisecond level. The vesicle release in the synapse can too be quantized. It need 3 bits, from 000 for no vesicle release up to 111 for seven vesicles. On the dendridic side, there is too a quantification process defined by the number of channel receptors. Up to 10 bits may be requested here to define 1023 elements. This is for a single channel kind, a spine may have 3 or 4 receptor famillies. In a spine head, the information is processed chemically in a very non-linear way. After that it is transfered as an electrochemical potential in a more "simpler" form. Simpler, is quite relative here. At the junction between the spine and a dendrite branch, there is more processing as at each dendrite branch junction. At the soma, the signal acts on the biochemical factory of the cell by releasing calcium ions. At the soma membrane level or in the largest dendrite branchs,the action potential sent to the axon is generated. Each element : The synapse, the dendritic receptors, the dendrite head spine, the spine body, the spine junction, the dendrite segments, the dendrite junctions, the electrochemical soma, the biochemical soma, the axon first segment, the generic axon segments, the axon terminal buttons at synapses must all be modelized. This is a single "ordinary" neuron, not a brain. Far simpler "neurons" can be modelized with interesting properties for electronics neural networks, but that can't simulate a true biological neuron. On the political side, to get some financing, a project must display some "progress". This one is most simply defined by the number of neurons simulated. There is so a political pressure to use very simplified neurons in large numbers to get a good announce effect and get money. This bring the reseach away from the solution of the uploading problem where the full complexity of real neurons must be taken into account. Yvan Bozzonetti. Content-Type: text/html; charset="US-ASCII" [ AUTOMATICALLY SKIPPING HTML ENCODING! ] Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=25806