Model of homeostatic artificial neuron
Martin Růžek, Tomáš Brandejský · 2010
This paper presents a model of homeostatic neuron that is able to find its state of equilibrium by observing the others neurons weights. This method is based on measuring the weights that the other neurons of the neural network assign to the output of the reference neuron, and on improving the parameters of the reference neuron in order to maximize the weights of the other neurons. The basic presumption is that the neuron is trying to maximize its importance in the whole network, which means that it is trying to maximize the values of the weights of the other neurons. The neuron is changing its own input weights and is measuring the reaction of the other neurons. Several types of learning are presented, depending on the way in which the importance of the weights is evaluated.