Evolutionary Computing

I''ve never heard of it myself, but it sounds awesome. The downside, of course, is that it also sounds insanely complicated to code something like that.

Are you talking about genetic algorithms?

If so, then programming them is pretty easy, but coming up with accurate and very very precise requirements is the touch part. I recall seeing a video of a virtual humanoid "learning" how to walk, and before the rules were tweaked correctly, it ended up dragging itself around by its arms, which accomplished its goal, just in the wrong way.

I''ve heard that it''s already being applied to electrical designs and all sorts of things where optimal efficiency is required. It''s definitely an interesting field.

As I understand it, GA''s certainly have their uses, but they simply cannot live up to the hype that was built up about them a few years ago (eg "they will solve all our problems and now we don''t even need to think for ourselves anymore because neural nets will save the day").

I think they would be much more usefull if computers were built to solve large parallel problems quickly or were otherwise designed specifically to run neural nets and GA''s.

quote:

---

Original post by Man_with_a_Plan I was told they did this to make an Analog to Digital Converter and what the computer spit out baffled them. They just couldn''t figure how it worked but it worked really really well.

---

I heard they used a GA to make a circuit whose target design was a regular oscillating voltage of any frequency. They spent AGES trying to work out how this random collection of components generated a signal, an impressively regular signal... an impressively regular 50Hz signal...

********


[Cleese]: I''m pink, therefore I''m spam.

Evolutionary Computing consists mainly of

Genetic Algorithms (chromosomes as fixed length strings)

Genetic Programming (chromosomes as trees)

you can do "real" stuff with it like letting the weights of a ANN evolve over time or finding local optima in functions and fun stuff like writing a little program that evolves pictures over time (that's what i do at the moment) and that's called Evolutionary Art.


[edited by - Honey on May 11, 2004 9:20:27 AM]

I''m an CS/AI student, and a few months ago we had a guest lecturer speak on this subject. He personally had a GA solve a problem which he cannot understand, and he (this is apocryphal, so take it with a pinch of salt) knows of others who used GAs to evolve circuits which are not connected, they use the electromagnetism caused by the current in one to induce the current in another area.

If you like that sort of stuff, read about Karl Simm''s work. We were shown some a video of some artificial creatures that he evolved swimming around. Some evolved to be very lifelike, sea snakes, caterpillar style movement, while others looked completely alien. It''s pretty amazing.

Here''s what I''m talking about.

Another interesting thing (depends if you like music or not really) is what I''m hoping to replicate, GenJam. It was a GA that took a set of chord progressions and started improvising jazz solos from a set of scales, based on that progression. The fitness function was a human in this case (so it wasn''t a typical GA, more an IGA), although a neural network was constructed which attempted to train the fitness function, but failed.

You can read about GenJam here.

Interchip communication is being developed by Sun, and it would make sense for them to use GAs to do it. Two chips are placed close together and the electromagnetic induction allows them to pass signals back and forth without needing solid connections between the circuits. Pretty cool stuff.

For my Master's thesis in Computer Science I used evolutionary computation to design CPU components. The ultimate goal was to design entire chips using the process I had refined.

Evolutionary computation consists of genetic algorithms, genetic programming, evolutionary strategies and evolutionary programming. Note that although these four domains sound like they are the same thing, they do have they're differences. That being said, there's considerable overlap, which is why I prefer the term "evolutionary computation."

There's a wealth of information out there on the Internet, most of it pretty bad. I doubt that you'd be interested in the volumes of scientific research on the topic, so I'll point you to one of my favourite "popular" articles. It's a piece for Discover magazine about a researcher who uses GAs and reconfigurable chips to solve problems. (It was the original inspiration for my MSc work.) Here's the information about the article:

------------------------------
Evolving A Conscious Machine
Some computer scientists think that by letting chips build themselves, the chips will turn out to be stunningly efficient, complex, effective, and weird—-kind of like our brains.
By Gary Taubes
DISCOVER Vol. 19 No. 06 | June 1998 | Technology
------------------------------

[edited by - Graylien on May 11, 2004 12:45:42 PM]

quote:

---

Original post by MDI
... he (this is apocryphal, so take it with a pinch of salt) knows of others who used GAs to evolve circuits which are not connected, they use the electromagnetism caused by the current in one to induce the current in another area.

---

Sounds like Adrian Thompson, the researcher profiled in the article I mentioned previously (see above post).