[fluxus] [OT] fastbreeder

[dave]

On Sun, 2006-02-19 at 17:49 +0100, Artem Baguinski wrote:
> hi all
> 
> excuse my networking, eh, crossposting.
> 
> On 13 Feb 2006, at 11:54, Dave Griffiths wrote:
> 
> > something vaguely related to fluxus:
> > http://www.pawfal.org/Software/fastbreeder/
> >
> > I have an idea we could evolve turtle builder commands, and send  
> > them over
> > osc to fluxus.
> 
> what would the criteria for selection be then?

purely the operator's ears... I'm afraid it's really very trivial. it
could be taken much further in a number of directions - but I was mainly
interested in the code generation rather than fitness functions (which
are much much harder)

also with fastbreeder as it stands it's more about exploration rather
than self organisation.

but, yep, with the right fitness function you might be able to score
results against a sample wav file, and evolve similar sounds - maybe. 

> in gerald's fluidiom, if i get it right, the criterion is how nice  
> the thing can crawl or walk. gerald?
> 
> and ever since i worked with martin i'm interested in formalising the  
> criterion for the output being less boring :-)

well, that's the challenge. it's even harder if you can't constrain the
problems to a specific set of parameters (ie what makes a sound
interesting?) systems like gerald's are so succesful because you can
score against tangible things like speed (if it's done in the same way
as sim's work?) 

can you use a meta fitness function to evolve your fitness functions? at
some point about there, it all gets beyond me, and I just want to make
something make a sound/picture/taste... :)

cheers,

dave