18 Aug
2001
18 Aug
'01
2:07 p.m.
It was thus said that the Great Iggy Drougge once stated:
Tony Duell skrev:
But what if they can't? Here's an article about a researcher who applied
genetic programming [1] techniques to building a circuit in an FPGA and
ended up with a 32 gate result (to distinquich between two tones), with five
gates seemingly unconnected yet crutial to the operation. The researcher
doesn't fully understand *why* it works:
http://www.newscientist.com/hottopics/ai/primordial.jsp
-spc (Is uneasy with this but yet sees this as becoming more and more
common ... )
[1] A method of evolving the best program to solve a particular
problem. It takes several *solutions* and breeds the most
promising ones to produce better offspring that get a better
answer to the problem.
I've seen hardware 'designers' do
something similar with FPGAs (and other
technologies that are easy to modify). Things like 'maybe it'll work if I
change this AND gate to an OR gate' or 'I'll try inverting that clock
signal'. Or 'Maybe I need one more state in that counter'. No real idea
as to what they should be doing, and why.
As long as you analyse it afterwards and find out what made it work/not work,
it's all right by me.