Pulsed para-neural network (PPNN) synthesis in a 3-D cellular automata space

This paper deals with a synthesis of pulsed para-neural networks (PPNN) in a 3D cellular automata space. In its essence, PPNN is a set of simple processing units that change their states only in certain discrete moments of time. A given unit may send a pulse to and only to its nearest neighbors. There are three kinds of processing units: red cells, yellow cells, and blue cells. A red cell cumulates excitation/inhibition in both time and space. It emits a pulse if and only if the value of its counter at t+1 gets equal to or greater than 2, where the state of the counter for t+1 is the state for t plus the weighted sum of pulses incoming in t. Each weight is associated with one and only one inlet to the cell and may be equal to 1, 0 or -1. The counter is zeroed after every pulse emission. A yellow cell represents a formal neuron whose inputs provide a pre-synaptic inhibition to all other inputs. A string of adjacent blue cells constitutes an axon. A number of useful devices has been created in its framework. We present: (1) an associative memory to be filled via reinforcement learning (what is remembered is a set of phases of pulses circulating in closed loops made of cells), (2) a spiking neuron that non-linearly cumulates excitation and responds with a changeable frequency of produced pulses, (3) an adjustable timer, and (4) an object location recognizer. We also present tools and methods used for PPNN synthesis.