General > Off Topic

Mechanical Neural Network

(1/3) > >>

Houshalter:
I'm not particularly fond of neural networks since their not universal turing, but in the real world they do have their uses. I was wondering if it would be possible to make a neural network out of paper. You might laugh, but I've seen some crazy things built out of paper inculding clocks and automata, plus I've seen old mechanical adding machines before. My problem is trying to find a simple way to do addition and multiplication for the weights and thresholds. Any ideas? Am I crazy?

happyhamsterchan:
I thought ANNs WERE turing complete... Where'd you find that out?

Houshalter:
Well what if you have 0 as an input. No matter how many times you multiply it by weights, you will never get one as an output.. You could make the thresholds dynamic, but that makes it even more complicated and I didn't think most ANNs did that anyways. They also can't do XOR supposedly, but I haven't tried that.

ashton15:
maybhe you could have like a vertical mechanical NN that's like an abaucus so you drop a bead down this tube and it does like see-saws and that kind of thing and can give outputs that way like binary.

Numsgil:
Clicky.  It says there are proofs that NN are turing complete.  It doesn't elaborate, though.  I would expect neural networks to be turing complete.  Just not terribly efficient at it.

Navigation

[0] Message Index

[#] Next page