초록 ▼

A method and apparatus are disclosed that modify [ies]and generalize [s]the use in artificial neural networks of the error backpropagation algorithm. Each neuron unit first divides a plurality of weighted inputs into more than one group, then sums up weighted inputs in each group to provide each gro

A method and apparatus are disclosed that modify [ies]and generalize [s]the use in artificial neural networks of the error backpropagation algorithm. Each neuron unit first divides a plurality of weighted inputs into more than one group, then sums up weighted inputs in each group to provide each group\s intermediate outputs, and finally processes the intermediate outputs to produce an output of the neuron unit. Since the method uses, when modifying each weight, a partial differential coefficient generated by partially-differentiating the output of the neuron unit by each weighted input, the weight can be properly modified even if the output of a neuron unit as a function of intermediate outputs has a plurality of variables corresponding to the number of groups. Since the conventional method uses only one differential coefficient, that is, the differential coefficient of the output of a neuron unit differentiated by the sum of all weighted inputs in a neuron unit, for all weights in a neuron unit, it may be said that the method according to the present invention generalizes the conventional method. The present invention is especially useful for pulse density neural networks which express data as an ON-bit density of a bit string.

대표청구항 ▼

A neuron network comprising: an input layer; an output layer; and at least one hidden layer located between said input layer and output layer, wherein the input layer, hidden layer and output layer each respectively includes at least one neuron unit, neuron units arranged at subsequent layers being

A neuron network comprising: an input layer; an output layer; and at least one hidden layer located between said input layer and output layer, wherein the input layer, hidden layer and output layer each respectively includes at least one neuron unit, neuron units arranged at subsequent layers being interconnected to each other via a synapse, an input signal to be supplied to each neuron unit in hidden layers and output layer, said input signal being weighted by a weight of a corresponding synapse whereby each neuron unit receives weighted input signals, and wherein each neuron unit comprises: output signal generating means for generating an output signal from a plurality of weighted input signals, said means including further means for separating the plurality of weighted input signals into more than one group, for summing up weighted input signals in each group to provide each group\s intermediate output signals, and for processing the intermediate output signals to produce an output signal of the neuron unit; internal error signal generating means for generating an internal error signal for each neuron unit in the output layer which is representative of a difference between a desired output signal and an actual output signal of the neuron unit at the output layer; internal error generating means for generating an internal error signal for each neuron unit in hidden layers by summing error signals backpropagated to the neuron unit from neuron units in other layers; sensitivity generating means for generating a sensitivity signal, said sensitivity signal being representative of a partial differential coefficient that is the result of successively partially differentiating the output signal of the neuron unit with respect to each weighted input signal; error signal generating means for generating error signals, by electronically processing the sensitivity signal for the synapse, the internal error signal, and the weight for the synapse; and back-propagation means, coupled to said error signal generating means, for back-propagating each of said error signals through each synapse; and weight modifying means, coupled to said internal error signal generating means and sensitivity signal generating means, for modifying each weight of a weighted input signal of said neuron unit, by electronically processing the internal error signal and sensitivity signal of said neuron unit.