A model of neural network for spatiotemporal pattern recognition (Q1089279)

A general many-parameters neuron model is presented, which leads in special cases to Purkinje (P-) and basket (B-) cells. The cells are constituent parts of a network, structurally characterized by: (a) two sorts of inputs, receptory and heterosynaptic, the latter increasing the sensitivity of the neuron, (b) connection of every input to every P-cell, which sums up the postsynaptic potentials, (c) effect of B-cells, by means of which every P-cell output triggers subsequent inhibitory pulse trains to all P-cells. Due to this, the network is able to detect the cell which first reaches the threshold and thus represents the pattern to be recognized. With suitably adjusted heterosynaptic inputs the net is then able to recognize time-related events. While not given in all details, the model is supported by extensive simulation experiments and physiological evidence.