From bipartite graphs to neuro-like structures. Ver.3.

Implemented 2-layers network prototype. First layer consists of Sensors. 

There’re 2 types of sensors.
Type 1 : neuron is activated on condition {input bits} & {input(activation) mask} = {input(activation) mask} mask}
  Type 0 : sensor is activated on condition {input bits} ^ {activation mask} = {0}  
Sesors are generated. Number of sensors depends on learning set and unlimited.

Second layer consists of Indicators.

Indicator is activated on condition Ni + … Nk +.. NL >= Threshold Yi.
Number of indicators is ammount of bits '1' in second elements of learning set pairs.

Learning sample is considered as reflection of sensors (its mediators) to indicators, like set of pairs.
It looks like {‘011000101101111000’ => ‘11000001010’}.
First set of bits represents input for sensors, second set of bits – activated indicators.
Conditions:
- set of samples must not contain holes, i.e. S1 || S2 ... || S_last = {11..111}, all bits must be '1'

It's limitation as for neuron/sensor set as for indicator set..
- number of indicators and number of bits in input sensor mask is limited by sizeof(long long)*8.

This prototype implements only one rule : synthesized neural network should not produce errors while
trying learning sample, i.e. after network has been learned, it is able to reproduce learning sample
( input to sensors =>activated indicators) with no errors for all pairs .
In other words, the prototype realizes the principle “Not prohibited - means allowed”.