Title: experimenting with mandelbrot set and neural network Post by: ker2x on November 15, 2010, 11:36:28 PM i tried to program a neural network that try to guess if a complex point is in the mandelbrot set.
The result is... well... bad. (and it's one of my best result) :sad1: Anyone tried ? anyone know about NN ? (http://fractals.s3.amazonaws.com/nn1.jpg) Title: Re: experimenting with mandelbrot set and neural network Post by: cKleinhuis on November 16, 2010, 12:10:22 AM lol, funny application, just keep it running ;)
how many neurons do u use for that ? i think that a basic set up cannot lead to impressive results, but the overall shape is visible, so the system may has learned something, it will always be kinda "unsharp", if the method would be fast enough for a point to decide it is inside/outside, and any of those 2 can give valid results, the other variant can be calculated, perhaps you should train your nn for detecting points that are OUTSIDE ? regards Title: Re: experimenting with mandelbrot set and neural network Post by: Jesse on November 16, 2010, 12:19:34 AM A rough shape should work, of course a chaotic function that gives different results for very near points needs to much information storage. You need a good way to feed the network, maybe not just x and y, but a precomputed value... hmm like 10 iterations of the formula :dink: I already thought of a neural network as configurable formula for one iteration... so you could varify the weights and choose yourself if its getting better to find some interesting stuff... |