Neural network processing apparatus for identifying an unknown image pattern as one of a plurality of instruction image patterns

Claim 2

A neural network processing apparatus, comprising: a plurality of neurons functioning as a group to generate an output signal by performing an operating function on an input signal, the operating function having a gain; a network for interconnecting the plurality of neurons and for setting weights according to the strengths of connections between the plurality of neurons; activator means for activating each neuron of the plurality of neurons to receive as a neuron input signal the weighted sum of the output signals of interconnected other neurons of the plurality of neurons and corresponding weight between the plurality of neurons, and to generate the output signals; calculation means for repeatedly calculating an average of the absolute values of differences between the output values of all of the plurality of neurons and a center value, and for calculating a difference between a current value of the average and a previous value of the average; function control means for determining the gain of the operating function according to the difference between the current value of the average and a threshold, and according to the difference between the current value of the average and the previous value of the average; and function set means for setting the gain of each neuron using the gain of the operating function determined by the function control means for each neuron.

Claim 3

The neural network processing apparatus according to claim 1, wherein the center value is an average of the output values of all of the plurality of neurons.

Claim 4

The neural network processing apparatus according to claim 2, wherein the threshold is an average of the absolute value of the difference between the output values of all of the plurality of neurons and the center value for the instruction image pattern.

Claim 5

The neural network processing apparatus according to claim 3, wherein the operating function is represented as f(λu) and wherein the function control means calculates a new value of λ, λ n ; where: u=the input value of the neuron λ n ←(1+K 1 (a-X)-K 2 D)λ n-1 K 1 , K 2 =positive constants a=the threshold X=the average of the absolute values of differences between the output values of all of the plurality of neurons and the center value. D=the difference between the average and the previous average.

Claim 6

The neural network processing apparatus according to claim 4, wherein the function control means restricts a rate of change of the gain K 1 (a-x)-K 2 D within predetermined limits.

Claim 7

The neural network processing apparatus according to claim 1, wherein the center value is an average of a maximum output value and a minimum output value of the plurality of neurons.

Claim 8

The neural network processing apparatus according to claim 1, wherein the network means determines the weights according to an instruction image pattern.

Claim 9

The neural network processing apparatus according to claim 1, wherein the activator means begins activating the network means according to an initial value, when output values of an unknown pattern are supplied to corresponding neurons as the initial value.

Claim 10

The neural network processing apparatus according to claim 1, wherein the calculation means calculates the average whenever output values of all of the plurality of neurons change.

Claim 11

The neural network processing apparatus according to claim 9, wherein the function control means decreases the gain of the operating function when the average is larger than the threshold.

Claim 12

The neural network processing apparatus according to claim 9, wherein the function control means increases the gain of the operating function when the average is smaller than the threshold.

Claim 13

The neural network processing apparatus according to claim 9, wherein the function control means decreases the gain of the operating function when the average is larger than the previous average.

Claim 14

The neural network processing apparatus according to claim 9, wherein the function control means increases the gain of the operating function when the average is smaller than the previous average.

Claim 15

The neural network processing apparatus according to claim 1, wherein the activator means reactivates the network whenever the operating functions of the plurality of neurons are set by the function set means.

Claim 16

The neural network processing apparatus according to claim 14, further comprising examine means for determining that the plurality of neurons are in a stable state based upon a determination that the output values of all of the plurality of neurons are unchanging for a predetermined time.

Claim 17

The neural network processing apparatus according to claim 15, further comprising extract means for extracting the output values of all of the plurality of neurons, which correspond to the unknown pattern, when the plurality of neurons are determined to be in the stable state by the examine means.

Claim 18

A neural network processing apparatus, comprising: a plurality of neurons functioning as a group to generate an output signal by performing an operating function on an input signal, the operating function having a gain; a network for interconnecting the plurality of neurons and for setting weights according to the strengths of connections between the plurality of neurons; activator means for activating each neuron of the plurality of neurons to receive as a neuron input signal the weighted sum of the output signals of interconnected other neurons of the plurality of neurons and corresponding weight between the plurality of neurons, and to generate the output signals; calculation means for calculating an average of the absolute values of differences between the output values of all of the plurality of neurons and a center value whenever the output values of all of the plurality of neurons change, and for calculating a difference between a current value of the average and a previous value of the average; function control means for determining the gain of the operating function by decreasing the gain of the operating function when the current value of the average is larger than a threshold or the previous value of the average, and by increasing the gain of the operating function when the current value of the average is smaller than the threshold or the previous value of the average; and function set means for setting the gain of each neuron using the gain of the operating function determined by the function control means for each neuron.

Claim 19

A neural network processing apparatus, comprising: a plurality of neurons functioning as a group to generate an output signal by performing an operating function on an input signal, the operating function having a gain; a network for interconnecting the plurality of neurons and for setting weights according to the strengths of connections between the plurality of neurons; activator means for activating each neuron of the plurality of neurons to receive as a neuron input signal the weighted sum of the output signals of interconnected other neurons of the plurality of neurons and corresponding weight between the plurality of neurons, and to generate the output signals; calculation means for calculating an average of the absolute values of differences between the output values of all of the plurality of neurons and a center value whenever the output values of all of the plurality of neurons change, and for calculating a difference between a current value of the average and a previous value of the average; function control means for determining the gain of the operating function by decreasing the gain of the operating function when the current value of the average is larger than a threshold or the previous value of the average, and by increasing the gain of the operating function when the current value of the average is smaller than the threshold or the previous value of the average; and function set means for setting the gain of each neuron using the gain of the operating function determined by the function control means for each neuron.