Abstract
|
In 1943, McCulloch and Pitts have created a computational model for neural networks. However, this model just only contains the spatial summation of neuron cell and threshold action. Because of this, all the neuron cells are provided with the same functions. In 2000, Tang created a neuron model based on The Koch-Poggio-Torre model whose mount of dendrites and shape are random, and synapses interact with each other. This model will eliminate the synapses which are needless, will rein-force others which are necessary, and even make synapses and dendrites` generation, so as to form a kind of dendrites holding a special function. However, this learning speed of the model is not so ideal. In this paper, the variable learning rate is introduced in order accelerate the learning speed.
|