The Powder Toy

feed forward Neural Net w/ simple learn by Paul_31415

Paul_31415
Paul_31415
9 / 0
6th Feb 2018
6th Feb 2018
Neural Net with 4:8:8:8:1 neurons. The teaching system is an oversimplified version of back-propagation. Taught by buttons. Reset code: for i=0,90001 do if tpt.get_property("type",i) == 66 then tpt.set_property("temp",math.random()*200+173.15,i) end end
artificial inteligence neuralnet backpropagation

Comments

  • Paul_31415
    Paul_31415     6th Feb 2018
    I taught it using only the existing teaching system. It took a long time, so for smaller nets, I would recommend just editing the temps directly.
  • R33sesK1ng
    R33sesK1ng     6th Feb 2018
    OMg amazing +1
  • Paul_31415
    Paul_31415     6th Feb 2018
    Theory of operation: Each neuron has a few axons. Axons are vats of GLOW with a temperature that functions as the axon weight (weight ranges from -100 to 100). Each time a neuron fires, it copies one pixel of GLOW to each of the neurons in the next layer. Neurons fire when the GLOW in their input (which is being cooled constantly) is above 0C.
  • Paul_31415
    Paul_31415     6th Feb 2018
    How to make it work: press the inputs then press "compute." Wait for it to finish then, to teach it, press "prep learn" (optionally press "learn mode 2" before) then press the correct answer. Then press "press before compute" and repeat from the top.