JPH06195488A - Neural network for fuzzy inference - Google Patents

Abstract

PURPOSE:To make a cell composing a neural network output the limit output of the addition result of an input signal. CONSTITUTION:In a neural network where a cell outputting an input signal and plural intermediate cells with which the signals of the cells are connected via a prescribed coupling coefficient are connected with plural final output cells by having hierarchical constitutions of these intermediate cells, an intermediate cell (a) is provided with a function outputting the limit output of the addition result of a signal to be coupled with an input terminal.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a neural network (hereinafter referred to as NN) for fuzzy inference having cells that output a limit output of addition results of input signals.

[0002]

2. Description of the Related Art Neurocomputers focus on neurons (nerve cells), which are the basic elements of the brain,
It aims to achieve the same function as the brain, using the NN resulting from the combination of these as a hint. The feature of these NNs lies in the parallel information processing between the neurons and the ability of learning. The NN has a hierarchical structure consisting of several layers, each layer is composed of an appropriate number of cells, there is no coupling within the same layer, and coupling between layers is from the input layer (first layer) to the output layer ( It is a one-way connection toward the final layer). The cells of each layer except the input layer receive the weighted inputs from the cells of the previous layer, calculate the sum of the weighted inputs, and multiply them by an appropriate function f to output. The input / output functions of cells used in the NN include a threshold function, a piecewise linear function, a logistic function, an identity function, etc. (references, published by Sangyo Tosho, Hideki Aso, "Neural Network Information Processing" P.13).

[0003]

If a piecewise linear function is used as the input / output function of the cells forming the NN, the membership function consisting of straight lines can be created in the NN. And the membership function consisting of a straight line has an advantage that hardware can be easily realized. The present invention has been made in view of the above circumstances, and is an NN for fuzzy inference having a cell that outputs a limit output of an addition result of input signals.
Is intended to provide.

[0004]

In order to achieve the above object, in the NN of [Claim 1] of the present invention, a cell outputting an input signal is connected to a signal of the cell via a predetermined coupling coefficient. In a neural network in which a plurality of intermediate cells and the intermediate cells have a hierarchical structure and are connected to a plurality of final output cells, the intermediate cells output a limit output of addition results of signals coupled to input terminals. It was equipped with. In the learning method of [Claim 2], the error is not propagated in the limit region of the intermediate cell when the coupling coefficient is updated by using the error backpropagation method. According to the learning method of [Claim 3], when the coupling coefficient is updated using the error backpropagation method, the error is propagated at a predetermined ratio in the high limit, the low limit and the linear region of the intermediate cell.

[0005]

Embodiments will be described below with reference to the drawings. Figure 1
FIG. 1 is a configuration diagram of an embodiment of a NN according to the present invention. In FIG. 1, a is a cell that outputs a piecewise linear function, O1 to Ojm.
x is the output of the cell bj (j = 1 to mx) coupled to the input end of the cell a, and W1 to Wjmx are coupling coefficients. At this time, cell a
Outputs the value shown in the following equation according to the value of the total sum I of the inputs.

[Equation 1] The input / output function of this cell is shown in FIG. As shown in FIG. 3, three sections of the cell input I, I <x1, x1 ≦ I ≦
x2 and x2 <I are called low limit, linear, and high limit regions (when y1 ≦ y2, the low limit and the high limit are switched when y1> y2).

Next, a learning method using this cell will be described with reference to FIG. Where Wi (k-1), j (k) is k-
The coupling coefficient between the i-th cell in the first layer and the j-th cell in the k-layer is shown, and Ij (k) is the sum of the inputs to the j-th cell in the k-layer,
Oj (k) indicates the output of the j-th cell in the k-th layer. Coupling coefficient Wij
The following formula is used for updating.

[Equation 2] Ordinary cell (sum of input x to cell and output y of cell is
Differentiate the input / output function y = f (x),
We will call it a normal cell. Specifically, reference (1)
, And the cell back error propagation method according to the present invention will be described below.

[Equation 3] dj (k) is divided in the following cases.

(I) When the k layer is the output layer

[Equation 4]

(Ii) When the k layer is the intermediate layer

[Equation 5] Dm (k + 1) in the above equation is obtained in the same manner as dj (k) from the output side.

[Equation 6] Therefore, learning is possible even in the NN including the cell according to the present invention.

[0009]

As described above, according to the present invention, N
Since the function of outputting the limit output of the addition result of the input signals is added to the cells forming N, it is effective for allocating the neural network of software to the linear hardware.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of a NN according to the present invention.

FIG. 2 is a diagram illustrating the learning method of FIG.

FIG. 3 is a diagram showing an input / output function.

[Explanation of symbols]

a, b _{1 to} b _jmx cell O _{1 to} O _jmx output of cell b _j W _{1 to} W _jmx coupling coefficient

Claims (3)

[Claims] 1. A cell that outputs an input signal, a plurality of intermediate cells in which the signals of the cells are connected via a predetermined coupling coefficient, and these intermediate cells have a hierarchical structure to form a plurality of final output cells. In a connected neural network,
The neural network for fuzzy inference, wherein the intermediate cell has a function of outputting a limit output of an addition result of signals coupled to an input terminal. 2. The learning method according to claim 1, wherein the error is not propagated in the limit region of the intermediate cell when the coupling coefficient is updated using the error back propagation method. 3. The method according to claim 1, wherein when the coupling coefficient is updated using the error backpropagation method, the error is propagated at a determined ratio in the high limit, the low limit and the linear region of the intermediate cell. How to learn.