Moderationism: Self-Organization Algorithm for Neural Networks by Making Excitation of Every Neuron to Moderate

摘要

We propose a noble self-organization mechanism of neural networks, named "Moder-atism", that is, every neuron and synapse try to get moderate level of excitation. By this algorithm, the input-to-output relationship of neural networks is automatically formed so as to adapt to the environment, even if there are feedback loops inside and/or outside. By applying this organization mechanism, it is shown that the networks acquire the reflex to avoid the dangerous stimulus. Also the networks start self-oscillation by themselves to get some amount of stimulus when there is no external signal.