(January 28, 2016 at 9:45 am)Emjay Wrote: Say you have a cube with nine 'panels' on each face, like a Rubik's cube, and each of those panels represents a pressure sensor and has a corresponding neuron in the input layer, then it would be interesting to see what representations and expectations form in the network given simulations of what could happen to it... such as just laying on a table (ie nine panels active)... being pushed off a table (nine panels + one on an adjoining face... representing pushing it along with a finger... then some of the nine opposite from where the pressure is turning off as the cube hangs over the edge before falling... then the random way it bounces, rolls, and comes to rest at the bottom, meaning 3 panels each from two adjoining faces if it bounces on an edge, and 1 each from three faces if it bounces on a corner etc... no pattern in the which faces/edges/corners hit first after the fall but you would still expect a general pattern of it's movement to emerge... the rolling motion. Then to 'evolve' this thing you would just add panels/mini-cubes to the surface and add corresponding neurons to the input layer. I'm just curious to see what sort of representations would form and if it could come to represent its own shape. I'm just thinking that about the question of what could be the simplest hypothetical organism to model where you could see, scaled down massively, the principles at work in the human brain.
Sounds a lot like this to me: