[MURG] Neuropeptide Function?

[Yan King Yin]

From: Eugen Leitl <eugen at leitl.org>

>> Statistical methods can do the trick.
>
>Not really, because corellation goes haywire if causality is very delayed.
>A lot of stuff happens, which was caused by preceding events?
>
>Moreover, you assume the box will leak complete info over operation time.
>This is wildly optimistic for less than geologic time scales. You can be 
>sure you nail all emergent effects if you work up from lowest level of 
>theory there is. You can fall up several abstraction layers by pure machine
>learning that way, until you arrive at optimal representation for given
>hardware (you have to co-evolve both for optimal result, actually).
>
>To obtain a compact representation you need a full model to know what to
>abstract.

The second approach is exactly what I'm proposing. Work up from very
low levels and use data mining methods to help create compact models.
We don't need to go ridiculously low level, no quantum stuff with
microtubules etc. This seems very useful. The only remaining problem
is experimental data.

(I agree the first approach, finding statistical patterns from neural
recordings, is kind of hopeless for uploading purposes.)



____________________________________________________________
Get advanced SPAM filtering on Webmail or POP Mail ... Get Lycos Mail!
http://login.mail.lycos.com/r/referral?aid=27005