Date: Wed, 28 Aug 96 07:16:50 GMT
Subject: Re: darwin, speciation, optimization



> 
> Adam Rose wrote regarding Origin of Species:
> Date: Tue, 13 Aug 96 08:58:10 GMT
> Subject: Darwin ( dialectics ? ) 
> ... Also, it's clear that he could not have written the book without
> standing at the end of a long tradition of almost anti theoretical
> gentleman collectors.
> 
> Lisa:  What do you mean by "gentleman collectors" ?  I'd say he was
> much more than that.  
> 

Of course, he was much more than that.

What I was trying to describe by the term "gentleman collectors" , was
that layer of bourgeois men in the 19th century who collected and classified
rocks, fossils, and living organisms. Very often, these men were clerics, since
such an "occupation" gave them plenty of leisure time in a rural setting. Their
perspective was more often than not proving the diversity of god's creation, rather
than to do the opposite, as was Darwin's to start with. [ I forget his name, but
it was a cleric who taught Darwin his field craft.] This "collect and classify 
god's creation" tradition produced Darwin.

When Darwin puts forward some general law, he also provides a host of examples
to prove his point, and usually hints that he is aware of countless more.
Doubtless, he has used the "collect and classify" method to do his own research.
But he also implicitly or explicitly quotes other researchers, very often 
clerics of some sort. He stands at the apex of this tradition, and also marks
the beginning of its end, IMO.

> AR: His argument that natural selection advances at its fasttest when
> the sea level rises and falls, sometimes isolating islands, where
> specialisations can "mature" , and then allowing
> these specialisations to compete on a continental scale when the sea
> level falls, could show how to apply "genetic algorithms" to really
> large optimisation problems, which at the moment they don't solve
> effectively [ applicable to some stuff I do ].
> 
> Lisa:  I didn't know you were into that.  So was Steve Keen, formerly
> of M1, perhaps before your time there.  I don't know much of that
> application, but I know he was talking about a plain with mountains
> on it, mathmatically, in computer simulation.  
> 
> 
> What sort of optimization problems are you working on?
> 

My job is basically to write software that takes a large ( ish ) electronic
circuit, and to put it on a particular kind of chip ( an FPGA ). Any particular
layout has a "cost" associated with it, which says how "good" this particular
layout is.

We tend to use "Simulated Annealing" , an approximation to the way things seem to
"optimise" themselves when freezing from a fairly random looking liquid to a
regular, solid crystal. The concept of "temperature" in these algorithms is used
to get out of local minima - it controls the probability of accepting a modification
which is "bad", but which is neccessary to get from one local minimum via an
intermdiate "hill" to the global minimum.

"Genetic algorithms" don't tend to work for genuinely large problems such as this.
It occurred to me that the way Darwin is suggesting natural selection "coverges"
fasttest, is to have periods of isolation followed by periods of global connection.
The periods of isolation are neccessary so that some new adaption has time to
become "perfected" before a temporarily more advanced competitor wipes it out -
the periods of global competition are neccessary to see if these adaption is any
good. In effect, the "isolation" phase is like the "low temperature" phase of 
simulated annealing, the "global competition phase" similar to the "high
temperature" phase.

I have no experience with GA's. But people who work here said that the main
difficulty with them is defining the genetic "code" which encapsulates the
current solution, and the precise mechanism of combing two "parent" codes to
get a "child".

Actually, my guess is that a combination of the two approaches would be best,
but I have no time to prove it.

Adam.



Adam Rose
SWP
Manchester
UK


---------------------------------------------------------------



     --- from list marxism2-AT-lists.village.virginia.edu ---