The workout then could to start with get started with a single neuron relating to the feedback and also the production and dynamically removing and adding them right up until being nearly some tolerance of effectiveness. A completely new part - once again with a single neuron - could then be generated choosing yet another metric. Maybe something like if one layer is a lot bigger than both of its neighbors. During the long-term I hope it can be much better to depart from the part construction. http://www.crazysales.com.au/ Particularly when moving in direction of fully commited hardware as an alternative to float-crunching on GPUs. Aside from that, a number of the more advanced themes might not exactly be also conceivable with a tidy coating system.