JPH0554012A - Neural circuit network automatic constituting device using expert system - Google Patents

Abstract

PURPOSE:To provide neural circuit network automatic constituting device using an expert systems which efficiently and automatically constitute neural circuit network most suitable for given learning data. CONSTITUTION:A learning data base 11 where learning data of a constitution object neural circuit network 10 is stored, a learning controller 12 which selects prescribed learning data from the learning data base 11 and teaches it to the constitution object neural circuit network 10, an expert system 14 which monitors the learning state of the constitution object neural circuit network 10 and executes a preliminarily stored rule by interpretation of the monitor result to infer the correction for convergence of the constitution object neural circuit network 10, and a structure corrector 16 which corrects the structure of the constitution object neural circuit network 10 in accordance with the inference result of the expert system 14 are provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a neural network automatic construction apparatus using an expert system for automatically constructing a desired neural network using an expert system.

[0002]

2. Description of the Related Art Generally, in order to construct a neural network, a model of the neural network and parameter values associated with the model must be determined by analyzing data to be processed. However, conventionally, the work of determining this neural network model and the parameter values associated with this model has been performed manually, which requires a lot of experience in constructing the neural network and requires a large amount of experience. It took a lot of effort and time.

Although devices for automatically constructing the neural network have also been proposed, these devices have a rule of thumb for efficiently learning the neural network and learning control means for the neural network. Was not separated, and the control program had to be recreated every time a new rule of thumb was obtained, which was not practical.

[0004]

As described above, since the work of constructing the neural network is conventionally performed manually, it requires a lot of experience to construct the neural network, and it requires a great deal of labor and effort. Devices that take time and automatically build a neural network have also been proposed. However, these devices use a control program every time an empirical rule for efficiently learning the neural network is obtained. It had to be recreated and was not practical.

Therefore, an object of the present invention is to provide an automatic neural network construction apparatus using an expert system capable of efficiently and automatically constructing an appropriate neural network for given learning data. To do.

[0006]

In order to achieve the above object, the present invention provides a learning database means for accumulating learning data of a neural network to be constructed and a learning state of the neural network to be constructed, By interpreting the result, and by the analysis result of the learning data recorded in the learning database means, an expert system for executing a pre-stored rule to infer a correction for converging the neural network to be constructed, Learning control means for selecting predetermined learning data from the learning database means in accordance with the inference result of the expert system and teaching it to the construction target neural network; and the construction corresponding to the inference result of the expert system. And a structure modifying means for modifying the structure of the target neural network.

[0007]

In the present invention, the neural network to be constructed is monitored by an expert system in which empirical rules and procedures for constructing the neural network to be constructed are described, and the neural network to be constructed corresponding to the monitoring result. The correction for converging the network is inferred, the structure correction means corrects the structure of the neural network to be constructed according to this inference, and the learning controller selects the data to be learned from the learning database means. This enables efficient automatic construction of neural networks.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of an automatic neural network construction system using an expert system according to the present invention will be described below with reference to the drawings.

Generally, in the back propagation network, in order to efficiently perform learning, the number of intermediate units (elements) is determined, the learning data teaching order, the coefficient of the correction amount of the load value between the units, and the like are set. Ingenuity is needed. Therefore, in the embodiment described below, the construction of a network (neural network) is automated by describing these in an expert system.

FIG. 1 shows the basic configuration of an automatic neural network construction apparatus using the expert system of the present invention. This device monitors a learning target state of the construction target neural network 10, a learning database 11 for accumulating learning data of the construction target neural network 10, and a learning state of the construction target neural circuit network 10, and An expert system 14 that interprets the results and, based on the analysis results of the learning data recorded in the learning database 11, executes a pre-stored rule and infers a correction for converging the neural network 10 to be constructed. A learning controller 12 that controls learning of the neural network 10 to be constructed by selecting predetermined learning data from the learning database 11 corresponding to the inference result of the expert system 14 and teaching the selected learning data to the neural network 10 to be constructed.
And a structure corrector 16 that corrects the structure of the neural network 10 to be constructed so as to converge the learning of the neural network 10 to be constructed according to the inference result of the expert system 14.

That is, in this embodiment, an empirical rule for efficiently learning the neural network 10 to be constructed is described in the expert system 14, and by using this expert system 14, the neural network 10 to be constructed automatically. It shows a configuration for constructing.

In response to a command from the expert system 14, the structure modifier 16 determines the structure of the neural network 10 to be constructed, that is, the addition or deletion of elements, the presence or absence of connection between elements, and the load value of connection between elements. The learning parameter is further changed. Here, the structure corrector 16
When the construction target neural network 10 is a program installed on a computer, rewrites the contents of the storage area in the computer corresponding to the structure and learning parameters of the construction target neural network 10 to construct the construction target neural circuit. When the network 10 is composed of hardware such as a switch and a variable resistor, the switch is turned on and off, and the resistance value is changed.

The learning controller 12, in response to a command from the expert system 14, outputs appropriate data as learning data to be presented to the neural network 10 to be constructed.
Select from 1 and change the order in which the selected data is presented. The expert system 14 forms a command to be given to the structure corrector 16 and the learning controller 12 from the characteristics of the learning data and the convergence state of the neural network 10 to be constructed.

FIG. 2 is a block diagram showing an embodiment of a neural network automatic construction apparatus using the expert system of the present invention. In FIG. 2, reference numeral 10 denotes a construction target neural network which is automatically constructed in the apparatus of this embodiment, and learning data of the construction target neural network 10 is stored in the learning database 11. The system of this embodiment automatically constructs the neural network 10 to be constructed so that the data stored in the learning database 11 can be correctly recognized.

The learning controller 12 includes a learning database 11
Desired learning data is selected from the learning data stored in, and this is taught to the neural network 10 to be constructed, and various parameter values used during learning of the neural network 10 to be constructed are set. Selection of learning data from the learning database 11 and setting of parameter values are performed by a control signal from a control executor 15 described later.

The sensor interpreter 13 monitors the progress of learning of the neural network 10 to be constructed and data in which the learning is delayed, and the result is described by an expert system 14 which will be described later.
Is recorded in the working memory 142, which is a constituent element of That is, the input signal from the neural network 10 to be constructed to the sensor interpreter 13 includes numerical data representing the convergence of the learning data, and the sensor interpreter 13 sends this numerical data to the expert system 14 described later. A sensory data interpretation neural network 131 for converting into a logical expression that can be processed is incorporated.
The numerical data is converted into a logical expression by the sensory data interpretation neural network 131 and recorded in a working memory 142 of the expert system 14 described later.

The expert system 14 has an inference engine 141, a working memory 142, and a production memory 143. The production memory 143 includes:
An empirical rule for efficiently converging the learning of the target neural network 10 in various situations is described by rules of the IF ... THEN ... format. The inference engine 141 constantly monitors the items recorded in the working memory 142, searches the rules described in the production memory 143 for which the condition part is satisfied, and executes the rules.

The control controller 15 monitors the working memory 142 of the expert system 14, forms a signal corresponding to the logical expression when a predefined logical expression is recorded in the working memory 142, and outputs this signal. It is sent to the structure corrector 16 and the learning controller 12.

The structure corrector 16 is a neural network 1 to be constructed.
By editing the structure such as the number of elements forming 0 and the presence / absence of connection between each element, the learning of the target neural network 10 can be easily converged in accordance with the signal transmitted from the control controller 15. In this way, the structure of the neural network 10 to be constructed is edited and corrected.

The learning controller 12 also modifies the learning method of the construction-target neural network 10 so that the learning of the construction-target neural network 10 can be easily converged according to the signal sent from the control controller 15. To do.

Note that, in FIG. 2, arrows indicated by solid lines indicate the flow of control signals, and arrows indicated by dotted lines indicate the flow of data.

With such a configuration, the expert system 14 can give a qualitative interpretation to the numerical data obtained from the neural network 10 to be constructed and transmit the numerical data to the expert system 14. By describing the rules in the expert system 14, it becomes possible to automate the construction of a flexible and adaptive neural network.

FIG. 3 shows another embodiment of the present invention in which the neural network is incorporated in the inference process in the expert system 14 so that the flexible and adaptive interpretation in the expert system 14 can be performed more effectively. It is a thing. The embodiment shown in FIG. 3 is configured by adding a rule monitor 17 and a neural network 18 for inference control to the configuration of the embodiment shown in FIG.

The rule monitor 17 monitors the working memory 142 and the production memory 143 of the expert system 14, and transmits the number of the executed rule and the state of the working memory 142 to the inference control neural network 18. ..

The inference control neural network 18 has a plurality of elements corresponding to each rule described in the production memory 143, and the connection between the elements is strengthened according to the contents transmitted from the rule monitor 17. Change the height. Then, the items recorded in the working memory 142 can be simultaneously recorded in the rules described in the production memory 143.
When satisfying the condition parts of one or more rules, one of the rules satisfying the condition parts is selected and the selection result is transmitted to the inference engine 141. Inference engine 1
41 executes one rule transmitted from the neural network 18 for inference control.

Is a database for recording the contents transmitted by the rule monitor 17 to the inference control neural network 18.
When there are two or more rules that satisfy the condition part, the inference control neural network 18 selects one rule R, and the selected rule R is executed through the inference engine 141, resulting in the construction target neural network. It is assumed that the convergence state of 10 has deteriorated. In this case, the fact that the sensor interpreter 131 has deteriorated is recorded in the working memory 142 by a logical expression, and the recorded logical expression is transmitted to the inference control neural network 18 through the rule interpreter 17. Inference control neural network 18
Searches the situation when the rule R, which caused the convergence deterioration, is selected from the rule execution result recording database 19 so that the rule R is not selected even if the same situation occurs in the future. Modify the net 18.

With such a configuration, even if the rule of thumb for efficiently constructing the neural network recorded in the production memory 143 is incomplete, the neural network construction can be automated flexibly and adaptively. It will be possible.

FIG. 4 shows still another embodiment of the present invention in which the logic network 151 for logical expression interpretation is incorporated in the control executor 15 shown in FIG. In this embodiment, the logical formula recorded in the working memory 142 of the expert system 14 is interpreted by the neural network 151 for logical formula interpretation, and the structure modifier 16 and the learning controller 1 are analyzed.
2 to form a signal for controlling 2.

According to this structure, the structure corrector 16
And the flexible and adaptive control of the learning controller 12 is enabled.

[0030]

As described above, according to the present invention,
Since the empirical rules for constructing the neural network are described in the expert system and the structure of the neural network is automatically constructed by controlling the structure modification of the neural network and the selection of learning data by the command from this expert system. , It is possible to efficiently and automatically construct a neural network.

[Brief description of drawings]

FIG. 1 is a block diagram showing a basic configuration of a neural network automatic construction apparatus using an expert system of the present invention.

FIG. 2 is a block diagram showing an embodiment of an automatic neural network construction apparatus using the expert system of the present invention.

FIG. 3 is a block diagram showing another embodiment of a neural network automatic construction apparatus using the expert system of the present invention.

FIG. 4 is a block diagram showing still another embodiment of a neural network automatic construction apparatus using the expert system of the present invention.

[Explanation of symbols]

 10 Neural network to be constructed 11 Learning database 12 Learning controller 13 Sensor interpreter 14 Expert system 15 Control executor 16 Structure corrector 17 Rule monitor 18 Neural network for inference control 19 Rule execution result recording database 131 Sensor data Interpretation neural network 151 Logical expression interpretation neural network

Claims (1)

[Claims] 1. A learning database means for accumulating learning data of a neural network to be constructed, monitoring a learning state of the neural network to be constructed, interpreting the result, and recorded in the learning database means. An expert system that infers a correction for converging the target neural network by executing a pre-stored rule according to the analysis result of the learning data; and the learning database means corresponding to the inference result of the expert system. Learning control means for selecting a predetermined learning data from the selected learning data and teaching it to the neural network to be constructed, and structure modifying means for modifying the structure of the neural network to be constructed corresponding to the inference result of the expert system. An automatic neural network construction system using an expert system.