Phil Stearns has constructed a 45 "neuron" network of electronic parts which responds to lights and tones with a (rather cute) squealing sound. A picture of the components for this strange device:
Each "neuron" consisted of analog electronics corresponding to each of 6 functions: Input, Summing, Threshold, "Offset," "Output," and "Structure" (not sure about those latter three). The connectivity was determined by hand.
Phil states that the sculpture is not intelligent, but rather "some kind of squid baby."
Neural networks have great potential for contributing to the arts. For example, JP Thivierge used results from the cascade correlation learning algorithm to visualize random output values from a neural network:
Here's some art from a portion of rat brain placed into a petri dish:
And, of course, there's Jonathan McCabe's fantastic network art:
Jon McCabe's process is perhaps the most interesting. The color of each pixel corresponds to the behavior of the network at a particular combination of two parameters, indicate by the X and Y coordinates of that pixel. (I get the feeling there is substantial artistic license taken with the results - for example, one interview indicates that each network generates up to 6000 images, and the best are selected by McCabe himself).
Update: jrandom from reddit has also done something similar, by randomly weighting all connections in a three-layer feedforward network with two inputs (x, y coordinates in these pictures) and three outputs (r, g, and b values for each pixel). Here's an example of the result:
It sounds like none of these networks were trained to produce good art - but they seem to have some "natural" talent regardless!
Humanities & Social Science
Technology
Comments
Actually, it appears that the picture you posted is not of the device, but of the pieces which were put together to form the device. At your link there are a couple of pics of the constructed creature (the one with it hanging in front of a light source makes it look a bit squid-baby like, actually!).
Posted by: Anon | January 31, 2008 12:52 PM
Anatomically accurate knitted and quilted brains: http://neuroscienceart.com/
The disclaimer is the best part: "While our artists make every effort to insure accuracy, we cannot accept responsibility for the consequences of using fabric brain art as a guide for functional magnetic resonance imaging, trans-cranial magnetic stimulation, neurosurgery, or single-neuron recording."
Enjoy!
Posted by: HRS | January 31, 2008 9:06 PM
Very nice! In a similar fashion, I remember using GeneticArc some time ago -- with some heavy training, it could produce pictures thay might qualify as "art". I don't have the url at hand, but it should be rather easy to find with google.
Posted by: Christophe | February 2, 2008 6:53 PM
As an undergrad I experimented with some similar ideas, using the networks as attractors that learned to follow the various parameters output by a Wacom tablet (position, orientation, pressure): http://rpi.edu/~mcdonk/imitativeart/
Posted by: Kyle McDonald | February 3, 2008 11:55 PM
Yowzer, that's not a pretty circuit.
About 10 years ago I designed an op-amp based analog neural network card with my friend and colleague Dr. John Gallagher (at the time, of SUNY Institute of Technology outside Utica, currently at Wright State in Ohio). Actually, John gave me the specs and I designed and built the unit. Each card was one neuron and there was a card cage that fit eight cards as I recall. It was much, much cleaner than the unit shown above. I mean, printed circuits aren't that hard to make. The unit above looks like the guts of a 1950's tube amplifier without the tubes. 45 neurons is pretty cool though.
I also recall doing a SPICE simulation of a three neuron connection during prototyping. The simulation was pretty slow (of course, it was actually a simulation of a simulator, so...). John had plans on altering my manual controls (replacing the potentiometers with digital pots)and ultimately creating an IC for the neuron. He went off to Wright State before that could be realized. I don't know if he's still pursuing it or not. He's published considerable material over the past dozen years on analog neural networks.
Posted by: JimFiore | February 4, 2008 6:04 PM
The image for this post may be a bit misleading. It's not of the final piece but of the completed individual neurons before they were assembled into the final sculpture. The final install image is here:
http://www.art-rash.com/pixelform/installation/AANN/images/AANN-soundwalk2007-01.jpg
I'll be showing this piece again in LA (Week of Feb 17th www.soundinspace.org) and Minneapolis (Feb 26 - March 2 www.sparkfestival.org). Hopefully I'll have some better photo and video documentation. I plan on posting a more technical report on this project in the coming weeks.
Posted by: Phil Stearns | February 4, 2008 6:52 PM
the pictures produce the the NN's are spectacular. this is a unique experiment...how about converting the output to sound waves? has anyone tried it before?
Posted by: zouze | February 17, 2008 10:03 AM