[MURG] - emulation difficulties

[digfarenough]

I'd recommend a class (or book) in numerical methods rather than one 
in diff eq.


hm.. perhaps I'll check that out, although I'm under a pile of reading right now, not to mention some data analysis of LTP that I'm falling behind on..

Right: if the timestep is too large, the equations become unstable; 
if the timestep is sufficiently small, it stays stable.


alright, I'm partially convinced this is so, but of course I won't really believe it 'til I see it working out for myself.. I now have a different concern, but I have to think it through more--it may turn out to be nothing


That's error all right. But I see no reason to believe that this 
error is important in any way.


the error could possibly mean the difference between whether a neuron fires or not, not that that is necessarily important in the grand scheme of the simulation.. but if it happened often enough that cells weren't firing correctly, well, then I'd be a little concerned that my brain wasn't working the way it used to.. of course, this may not be a problem at all..


>No, basically all the universe can be described by different 
>equations, that doesn't mean it works on them.

Well, now we seem to be picking nits.


sort of :)

what I was getting at though is that you can use differential equations to describe the behavior of shooting a few thousand BB's out of a cannon, but you can also use the old fashioned kinematic equations to figure out where they each will go much more accurately (and more computationally expensively)... that might be a bad example though.. :)


Deterministic differential equations, yes. (And you can safely 
ignore Penrose; he's a reasonably good mathematician but a complete 
moron at neuroscience. He's looking to find God in quantum 
microtubules, and wants it badly enough to discredit himself in the 
attempt.)


well, I was referring to his five-fold symmetric crystals that he claimed showed quantum effects at the classical level... I keep meaning to look it up to see if there's any current work or new theories, but I never get around to it


Heh. Well, eventually the models would be built automatically, from 
the scanned-in information.


that'll take some pretty sophisticated technology, both AI-wise and imaging-wise, as you surely expect it to.. I'm trying to figure out right now how it'd work, to see if I have any critiques of that plan :)

it seems like the most important details are morphology and membrane protein type, number, and location... but I have a feeling there's a lot of information in the perikeryon you'd want too, like the current rates of transcription of different proteins, but I'm not sure how important that all is.. if you didn't include information about proteins currently being shipped about in the cell or transcribed, you'd probably have some cognitive problems for a while as an upload, while the algorithms make replacements..


Actually, I'm with you on that (i.e. I also used both and preferred 
GENESIS, though of course I preferred CONICAL even more). But good, 
you're way ahead of most people then. But if you've done simulations 
in these, then I would expect you'd have a good feeling for stability 
issues by now. Or perhaps not; both of these use advanced 
integration techniques and automatic adjustment of the timestep, 
which causes them to be always stable unless you go out of your way 
to make them unstable. So perhaps you've never seen a simulation 
blow up!


yeah, in using them I didn't have to pay much attention to the integration techniques.. basically you can just say use RK4 with a time step of oh.. whatever it was we used, and let it do its thing

I've never seen a simulation blow up, but I've seen properties in a simulation that I wouldn't expect in a real cell, for instance, after firing an action potential, and well after the refractory periods when the membrane potential had settled back to normal, there were tiny variations in the firing threshold caused by the gating variables oscillating slightly and occasionally lining up just right... I don't know if this is actually seen in cells, perhaps it's too noisy to tell


Instability would be a problem, except that it's easily solved, as 
tools like GENESIS show. Accumulated error is a problem for 
astronomers attempting to predict exactly where Jupiter is going to 
be a few hundred millenia from now. Accumulated error is not, 
however, a problem for us. It would be ludicrous to try to predict 
the exact state of any person years in the future. The brain has so 
many inputs, and is such a chaotic system, that any such error is 
meaningless.


but hypothetically, you could compare how a biological brain that had had the exact same inputs as a non-bio brain acts, and see if there's a difference.. there's a sense in which you could call the eventual difference an error, though I'm not saying it's correct to call it an error.. that'd be like myself as a 10 year old saying my brain now has errors because it acts differently than my old self expected it to

so basically I'm agreeing with you on that point :)


---------------------------------
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://askja.bu.edu/pipermail/murg/attachments/20030912/56e57913/attachment.htm