END OF AN ERA, FRACTALFORUMS.COM IS CONTINUED ON FRACTALFORUMS.ORG

i was just randomly remembering reading an article like this one in recent times about training a neural net to play super mario

http://thenextweb.com/insider/2015/06/14/watch-this-learning-neural-network-annihilate-super-mario-world-with-ease/

and wondered if it could potentially be at all worthwhile to try a neural net sort of setup on something like zooming in on the mandelbrot set?  feed it any useful info like iterations and distance to the boundary, that sort of thing, give it negative feedback for doing things like zooming into the set or blank areas, give it positive feedback for doing favorable things like staying on the boundary and keeping iterations low so its path can be rendered in a reasonable amount of time, etc.  probably take ages to get anything interesting out of it, maybe a dumb idea altogether?

hm, though I love both the idea of fractals and neural networking like deepdream, I don't think "training a computer to zoom" will lead to anything useful.
what should it look for?
it would need to know what it looks for.
spirals? minibrots? dragon-dogshapes with 144 horns?
whatever you search for, you will find, given you have enough time and cpu power.

If you don't know what you should look for, everything will be just random, within the realms of the parameters you mentioned. like "staying alive and keep iteration count managable"
no special zooms just randomly going through each and every possibility (which of course are endless) within your set parameters.

i think, of course it would be possible. it could zoom by itself. and even find stuff the human programmer deems worthy.
look for spirals-->endless spiralzooms..
look randomly-->random zooms  
look for dogs, and it will find dogs http://www.fractalforums.com/kalles-fraktaler-gallery/dogs/   might take a little longer to find these deep&complex shapes.
in the end it would be the programmer, setting the rules. just like kf already does with the "find minibrot"-tool.


the special beauty in mandelbrot zooms and deeper pictures is that the decisions you take where to zoom will form the future path (as I tried to show here).
and our decisions are driven by beauty and curiousity. at least that I would like to think  
how could you teach a computer to recognize this? and without, as said, the results would just be random..
us humans, we haven't learned what we find beautiful or interesting from the m-set, but from the world, our life. and we zoom where our life experience drives us towards..
imho, zooming could be seen as a highly psychological process, unique to everyone. 

a neural network without external output, let loose on the m-set.. I don't think it would do anything "useful", just endless randomness.

then again, most of my zooms are pretty useless too..


-ok, I admit, the golden means might be an interesting starting point-