US 5,350,953 AGranted

Digitally weighted neuron for artificial neural network

OwnerRichard M. Swenson, John C. Cole, III +2inventors

InventorsRichard M. Swenson, John C. Cole, III, Steve L. Holmes, Paul B. Ekas

Life of this patent

Estimating the term…

Loading patent…

Abstract

A neuron for an artificial neural network provides digital weighting of input signals at a common portion of the neuron rather than at each synapse. The neuron is adapted for use of differential signals, and the weighting may be provided by field effect transistors of different widths, by subtracting a plurality of differential signal components from an opposite most significant component, or by subtracting one half of a differential signal component from the opposite next most significant component. The neuron may provide binary sign selection and digit selection by switching input and reference signals at each synapse.

Claims12

Dependent claims

A neuron for an artificial neural network receiving a plurality of input signals representing values in a predetermined range, the neuron weighting each of the input signals by weight digits corresponding to numerals at digit positions in a predetermined number system and comprising:

a plurality of synapses, each of said synapses receiving one of said input signals and having a plurality of synapse digit weighting means, which correspond individually to said digit positions and which store single digit weight values corresponding to said numerals, for generating synapse digit position signals individually corresponding to the products of each of the values of single digit weight values and said one input signal
a plurality of digit summing means, which correspond individually to said digit positions and which receive the synapse digit position signals generated by the synapse digit weighting means corresponding to the same digit positions as the digit summing means, for providing digit position sum signals representing the sums of the synapse digit position signals corresponding to each of said digit positions;
and
digit position weighting means for receiving said digit position sum signals and for generating therefrom a neuron sum signal corresponding to the sum of the products of each of said digit position sum signals and the positional value in said number system of the digit position corresponding to the digit position sum signal,
whereby said neuron sum signal represents the sum of the products of the values of each of said input signals and a weight represented in said number system by the weight digits of the synapse receiving the input signal.

2 more elements

A neuron for an artificial neural network receiving a plurality of input signals representing values in a predetermined range, the neuron weighting each of the input signals by binary weight digits in a predetermined number of binary digit positions and comprising:

a plurality of synapses, each of said synapses receiving one of said input signals and having a plurality of synapse digit weighting means, which correspond individually to said binary digit positions, for generating synapse digit position signals having a predetermined zero value when a weight digit at a corresponding binary digit position is "zero" and having a value corresponding to the value of said one of said input signals when said weight digit is "one";
a plurality of digit summing means, which correspond individually to said digit positions and which receive the synapse digit position signals generated by the synapse digit weighting means corresponding to the same digit positions as the digit summing means, for generating digit position sum signals representing the sums of the synapse digit position signals corresponding to each of said digit positions;
and
digit position weighting means for receiving said digit position sum signals and for generating therefrom a neuron sum signal corresponding to the sum of the products of each of said digit position sum signals and the positional values of the binary digit positions corresponding to the digit position sum signals,
so that said neuron sum signal corresponds to the sum of the products of the values of each of said input signals and weights represented by the binary weights digits of the synapses receiving the input signals.

2 more elements

Dependencies are inferred from the claim text.