JPH07271747A - Information processor - Google Patents

Abstract

PURPOSE:To enable smooth relearning when input data or the output from a neuro circuit network shows an abnormal value by providing a switching device which switches and inputs input data stored in a storage device and external input data to the neuro circuit network. CONSTITUTION:This information processor consists of the storage device 2 which stores the external input data and the switching device which selects and outputs one of the external input data and the data taken out of the storage device 2 to the neural network 3. The storage device 2 uses, for example, a RAM. The storage capacity of the RAM is not specially prescribed, but the device is used in FIFO mode so that when the device is full of data, the oldest data are discarded and new data are overwritten. A data selector is used as the switching device 1. Namely, the input data are stored and when an error occurs to the neural network 3, the input data and the data from the storage device 2 are switched to the input data to relearn the neural network 3 with the input data in the storage device 2.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a neurocomputer that imitates a nerve cell network and can be applied to waveform recognition, motion control, temperature control and the like.

[0002]

2. Description of the Related Art A nerve cell network is constructed by using nerve cell elements that imitate the nerve cells of a living body and connecting them in layers or nets. Such a nerve cell network can be applied to various controls such as image and voice recognition, position control of robots, stock price and earthquake prediction, and the like.

In the conventional learning of a nerve cell network, teacher data is set in advance for each learning data to be presented,
The weight coefficient between nerve cell elements was changed so as to reduce the error between the output value of the nerve cell network and the teacher data. Therefore, the learning was only for the input data prepared in advance to the neural network. Or
As disclosed in Japanese Patent Laid-Open No. 03-11458, an apparatus for learning by adaptively updating a teacher signal from learning data when a target output value of input data to a neural network cannot be determined. Are known.

[0004]

However, in the case of using the nerve cell network, learning is performed using typical data prepared in advance, and unlearned data is input to the nerve cell network after learning to The above-mentioned control and recognition are performed by the output of the cell network. In particular, in the case of control or recognition that requires such work to be processed in real time, typical input data during learning due to unexpected data in the input data or changes in the conditions for sampling the data. There may be a gap between and. When such a deviation occurs, the learned neural cell network cannot cope. Therefore, it is an object of the present invention to provide an information processing device that can smoothly perform relearning when input data or an output from a nerve cell network has an abnormal value.

[0005]

According to a first aspect of the present invention, a nerve cell network capable of inputting data, a storage device for storing input data from the outside, an input data stored in the storage device, and an external input data. A switching device adapted to switch input data and input the data to the nerve cell network.

According to the second aspect of the present invention, the nerve cell network capable of inputting data, the storage device for storing the input data from the outside, the input data stored in the storage device and the input data from the outside are switched to the nerve. A switching device for inputting to the cell network, a monitoring device for monitoring the output from the nerve cell network, and a management device for managing the nerve cell network, the storage device and the switching device.

According to another aspect of the present invention, a nerve cell network capable of inputting data, a storage device for storing input data from the outside, an input data stored in the storage device and an input data from the outside are switched to enable the nerve. A switching device for inputting to the cell network, a monitoring device for monitoring the input data, a nerve cell network, a storage device, and a management device for managing the switching device.

According to another aspect of the present invention, a nerve cell network capable of inputting data, a storage device for storing input data from the outside, and a nerve by switching the input data stored in the storage device and the input data from the outside. A switching device for inputting to the cell network, a first monitoring device for monitoring the input data, a second monitoring device for monitoring the output from the neural cell network, a switching between the neural network and the storage device. It is configured by a management device that manages the device.

According to a fifth aspect of the present invention, in these inventions, when the monitoring device for monitoring the output from the nerve cell network notifies the abnormality, it has a storage device for storing the input data at that time.

According to a sixth aspect of the present invention, in these inventions, when the monitoring device for monitoring the input data reports an abnormality, the storage device stores the input data at that time.

According to a seventh aspect of the invention, in the fourth aspect of the invention, when the first monitoring device for monitoring the input data reports an abnormality, or the second monitoring for monitoring the output from the nerve cell network. When the device reports an abnormality, it has a storage device for storing the input data at that time.

According to an eighth aspect of the present invention, in these inventions, a storage device is provided for storing the input data and the target output value when the monitoring device reports an abnormality.

According to the ninth aspect of the invention, the learning error of the nerve cell network is recorded so that the coupling coefficient of the nerve cell network can be set arbitrarily.

[0014]

According to the first aspect of the present invention, the input data is stored in the storage device such that the input data is stored and the input data and the data from the storage device are switched when an error occurs in the nerve cell network. Can relearn the neural network.

According to the present invention, the input data from the outside is stored, the output from the nerve cell network is monitored, and when the output is not desired, the input data from the outside is stored in the storage device. By managing to switch to the input data and input to the neural network,
You can relearn the neural network.

According to the third aspect of the present invention, input data from the outside is stored, the input data is monitored, and when the input data has not been input so far, the input data from the outside is stored. The neural network can be relearned by switching to the input data stored in the device and managing so as to input to the neural network.

According to the fourth aspect of the present invention, when the input data from the outside is stored and the input data and the output from the nerve cell network are monitored, the output of the nerve cell network is not desirable or until now. When the input data has never been input to the neural network, the input data stored in the storage device and the input data from the outside are switched and managed so as to be input to the neural network. Can learn.

According to the fifth aspect of the present invention, when the monitoring device for monitoring the output from the nerve cell network notifies the abnormality, the input data is stored in the storage device. When there is an abnormality between the output value and the target output value, the data at that time is stored, so it is possible to know the input data that could not be learned until now and use it for re-learning. . Also, the number of stored data can be reduced.

According to the sixth aspect of the invention, when the monitoring device for monitoring the input data notifies the abnormality, the input data is stored in the storage device. Since the data of is stored, it is possible to know the input data that could not be learned until now and use it for re-learning. Also, the number of stored data can be reduced.

According to another aspect of the invention, when the monitoring device for monitoring the input data reports an abnormality, or when the monitoring device for monitoring the output from the nerve cell network reports the abnormality, the data is input to the storage device. By storing data, there is an abnormality between the output value of the nerve cell network and the target output value, or the data when there is an abnormality in the input data is stored. The input data that could not be obtained can be known and can be used for re-learning. Also, the number of stored data can be reduced.

According to the eighth aspect of the present invention, since the target output value is stored together with the input data when the monitoring device notifies the abnormality, it is not necessary to input the target output value from the outside at the time of re-learning. It was

According to the ninth aspect of the invention, the learning error of the neural network is recorded and the coupling coefficient of the neural network can be set arbitrarily, so that the change of the learning state can be known. At the same time, since the connection weight of the nerve cell network can be initialized, it is possible to get out of the unlearnable state due to the inappropriate connection weight.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the invention described in claim 1 will be described with reference to FIG. The information processing apparatus of the present embodiment selects either the storage device 2 that stores input data from the outside or the input data from the outside or the data extracted from the storage device 2 and outputs it to the neural cell network 3. It is constituted by the switching device 1. Here, the storage device 2 is, for example, R
It is also possible to use AM. Although the storage capacity of this RAM is not specified, a FIFO (first in first) is used so that when this storage capacity is full, the oldest stored data is discarded and new data is overwritten.
out) may be used. For the switching device 1, for example, a data selector may be used.

An embodiment of the present invention will be described with reference to FIG. The information processing apparatus according to the present embodiment stores a data input from the outside, and outputs a storage content data designated by the management apparatus 5 to the switching apparatus 1.
A switching device 1 that selects either external input data or data from the storage device 2 according to a command from the management device 5 and outputs it to the neural network 3, and a nerve that processes the input data and outputs the result. The circuit 3, the monitor device 4 that monitors the output result from the nerve cell network 3 and the target output signal, and outputs the monitor result to the management device 5, and the nerve cell circuit 3 is switched by the output of the monitor device 4. It is composed of a device 1 and a management device 5 that specifies the operation of the storage device 2. Here, the storage device 2 may use, for example, a RAM. The storage capacity of this RAM is not specified, but when this storage capacity becomes full, the FIFO (firsts) is used so that the oldest stored data is discarded and new data is overwritten.
Tin first out) may be used. For the switching device 1, for example, a data selector may be used. The monitoring device 4 calculates, for example, the mean square error between the target output value and the output value from the nerve cell network, and when the value becomes 0.1 or more, it notifies the management device that the above has occurred. To do so. Here, the error calculation is not limited to the mean square error, and the allowable value of the error is not limited to 0.1, and any value may be set.

In such a configuration, first, the management device 5 commands the switching device 1 to output the input data as it is. The management device 5 commands the storage device 2 to store the input data from the outside. Further, the management device 5 instructs the nerve cell network 3 to process and output the input data supplied from the switching device 1. The monitoring device 4
For example, it is monitored whether or not the error between the target output signal and the output of the nerve cell network exceeds a preset range, and if it exceeds, the management device 5 is notified of the abnormality.

When there is no notification of abnormality from the monitoring device 4, the management device 5 commands the neural cell network 3 to process the data input as it is. When the management device 5 is notified of an abnormality from the monitoring device 4, the management device 5 sequentially retrieves the data stored in the storage device 2 and sends the data to the switching device 1, and the switching device 1 sends the data to the neural cell network 3 Command to input. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is repeated until there is no notification of abnormality from the monitoring device 4.

An embodiment of the invention described in claim 3 will be described with reference to FIG. The information processing apparatus according to the present embodiment stores a data input from the outside, and outputs a storage content data designated by the management apparatus 5 to the switching apparatus 1.
A switching device 1 that selects either external input data or data from the storage device 2 according to a command from the management device 5 and outputs it to the neural network 3, and a nerve that processes the input data and outputs the result. The circuit network 3, the monitoring device 6 that monitors the input data and outputs the monitoring result to the management device 5, and the management that specifies the operations of the nerve cell network 3, the switching device 1 and the storage device 2 by the output of the monitoring device 6. And device 5. Here, the storage device 2 may use, for example, a RAM. The storage capacity of this RAM is not specified, but when this storage capacity is full, the oldest stored data is discarded and the new data is overwritten.
It may be used in FO (first in first out). For the switching device 1, for example, a data selector may be used. The monitoring device 6 may be a comparator, for example.

In such a configuration, first, the management device 5 commands the switching device 1 to output the input data as it is. The management device 5 commands the storage device 2 to store the input data from the outside. Further, the management device 5 instructs the nerve cell network 3 to process and output the input data supplied from the switching device 1. The monitoring device 6
The input data is monitored. For example, it is monitored whether the input data exceeds a preset range, and if it exceeds, the management device 5 is notified of the abnormality.

When there is no notification of abnormality from the monitoring device 4, the management device 5 gives an instruction to process the data input to the nerve cell network 3 as it is. When the management device 5 is notified of an abnormality from the monitoring device 4, the management device 5 sequentially retrieves the data stored in the storage device 2 and sends the data to the switching device 1, and the switching device 1 sends the data to the neural cell network 3 Command to input. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is performed a preset number of times.

An embodiment of the invention described in claim 4 will be described with reference to FIG. The information processing apparatus according to the present embodiment stores a data input from the outside, and outputs a storage content data designated by the management apparatus 5 to the switching apparatus 1.
A switching device 1 that selects either external input data or data from the storage device 2 according to a command from the management device 5 and outputs it to the neural network 3, and a nerve that processes the input data and outputs the result. A circuit network 3, a first monitoring device 15 for monitoring input data and outputting a monitoring result to the management device 5,
The second monitoring device 14 that monitors the output result from the nerve cell network 3 and the target output signal and outputs the monitoring result to the management device 5, and the outputs of the first monitoring device 15 and the second monitoring device 14. The nerve cell network 3, the switching device 1 and the storage device 2
And the management device 5 that specifies the operation of.
Here, the storage device 2 may use, for example, a RAM. The storage capacity of this RAM is not specified in particular, but when this storage capacity becomes full, the oldest stored data is discarded and new data is overwritten by the FIFO.
It may be used in (first in first out). For the switching device 1, for example, a data selector may be used. A comparator may be used for the first monitoring device 15, for example. In the second monitoring device 14, for example, the mean square error between the target output value and the output value from the nerve cell network is calculated, and when the value becomes 0.1 or more, the above occurs in the management device. Let them know. Here, the error calculation is not limited to the mean square error, and the allowable value of the error is not limited to 0.1, and any value may be set.

In such a configuration, first, the management device 5 instructs the switching device 1 to output the input data as it is. The management device 5 commands the storage device 2 to store the input data from the outside. Further, the management device 5 instructs the nerve cell network 3 to process and output the input data supplied from the switching device 1. The first monitoring device 15 monitors the input data and, for example, monitors whether the input data exceeds a preset range, and when it exceeds, notifies the management device 5 of an abnormality. Second monitoring device 14
Monitors the target output signal and the output from the nerve cell network 3, for example, monitors whether the error between the target output signal and the output of the nerve cell network exceeds a preset range,
When it exceeds, the management device 5 is notified of the abnormality.

The management device 5 includes a first monitoring device 15 and a second monitoring device 15.
When there is no notification of abnormality from the monitoring device 14, the instruction is given to cause the data input to the nerve cell network 3 to be processed as it is. When the management device 5 is notified of an abnormality from the first monitoring device 14, the management device 5 takes out the data stored in the storage device 2 one after another and sends the data to the switching device 1, and at the same time sends the data from the switching device 1 to the nerve cell. Instruct to input to the network 3. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is performed a preset number of times. Management device 5
When the second monitoring device 14 notifies the abnormality, the data stored in the storage device 2 are sequentially taken out and sent to the switching device 1, and the switching device 1 sends the data to the nerve cell network 3 Command to input.
In addition, the management device 5 controls the switching device 1 for the nerve cell network 3.
It is instructed to learn with the data input through and the target output signal. This learning is repeated until there is no notification of abnormality from the second monitoring device 14.

An embodiment of the invention described in claim 5 will be described with reference to FIG. The information processing apparatus according to the present embodiment stores externally input data designated by the management apparatus 5, and stores the data of the storage content designated by the management apparatus 5 in the switching apparatus 1.
The storage device 2 for outputting to the neuron circuit network 3, and either the input data from the outside or the data from the storage device 2 is selected by a command from the management device 5 and output to the nerve cell network 3. The neural network 3 for processing the results and outputting the results, and monitoring the input data and managing the monitoring results by the management device 5.
To the first monitoring device 15, which outputs the output result from the neuron network 3 and the target output signal, and outputs the monitoring result to the management device 5, and the first monitoring device 15. It comprises a nerve cell network 3, a switching device 1 and a management device 5 which designates the operation of the storage device 2 by the outputs of the device 15 and the second monitoring device 14. Here, the storage device 2
May use a RAM, for example. Although the storage capacity of this RAM is not specified, a FIFO (first-in-first-out) method is used so that when this storage capacity becomes full, the oldest stored data is discarded and new data is overwritten.
It may be used in the first out). Switching device 1
For example, a data selector may be used. A comparator may be used for the first monitoring device 15, for example. In the second monitoring device 14, for example,
The mean square error between the target output value and the output value from the nerve cell network is calculated, and when the value becomes 0.1 or more, the management device is notified of the occurrence of the above. Here, the error calculation is not limited to the mean square error, and the allowable value of the error is not limited to 0.1, and any value may be set.

In such a configuration, first, the management device 5 commands the switching device 1 to output the input data as it is. Further, the management device 5 instructs the nerve cell network 3 to process and output the input data supplied from the switching device 1. The first monitoring device 15 monitors the input data and, for example, monitors whether the input data exceeds a preset range, and when it exceeds, notifies the management device 5 of an abnormality. The second monitoring device 14 monitors the target output signal and the output from the nerve cell network 3, and checks whether the error between the target output signal and the output of the nerve cell network exceeds a preset range, for example. Monitor and monitor device 5 if exceeded
Notify to abnormal.

The management device 5 includes a first monitoring device 15 and a second monitoring device 15.
When there is no notification of abnormality from the monitoring device 14, the instruction is given to cause the data input to the nerve cell network 3 to be processed as it is. When the management device 5 is notified of an abnormality from the first monitoring device 1, the management device 5 sequentially retrieves the data stored in the storage device 2 and sends the data to the switching device 1, and at the same time outputs the data from the switching device 1 to the nerve cell. Circuit network 3
Command to input. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is performed a preset number of times. Management device 5
When the second monitoring device 14 notifies the abnormality, the storage device 2 instructs the storage device 2 to store the abnormal input data, and the data stored in the storage device 2 is taken out one after another. The data is sent to the switching device 1, and the switching device 1 instructs the data to be input to the neural cell network 3. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is repeated until there is no notification of abnormality from the second monitoring device 14.

An embodiment of the invention described in claim 6 will be described with reference to FIG. The information processing apparatus according to the present embodiment stores externally input data designated by the management apparatus 5, and stores the data of the storage content designated by the management apparatus 5 in the switching apparatus 1.
The storage device 2 for outputting to the neuron circuit network 3, and either the input data from the outside or the data from the storage device 2 is selected by a command from the management device 5 and output to the nerve cell network 3. The neural network 3 for processing the results and outputting the results, and monitoring the input data and managing the monitoring results by the management device 5.
To the first monitoring device 15, which outputs the output result from the neuron network 3 and the target output signal, and outputs the monitoring result to the management device 5, and the first monitoring device 15. It comprises a nerve cell network 3, a switching device 1 and a management device 5 which designates the operation of the storage device 2 by the outputs of the device 15 and the second monitoring device 14. Here, the storage device 2
May use a RAM, for example. Although the storage capacity of this RAM is not specified, a FIFO (first-in-first-out) method is used so that when this storage capacity becomes full, the oldest stored data is discarded and new data is overwritten.
It may be used in the first out). Switching device 1
For example, a data selector may be used. A comparator may be used for the first monitoring device 15, for example. In the second monitoring device 14, for example,
The mean square error between the target output value and the output value from the nerve cell network is calculated, and when the value becomes 0.1 or more, the management device is notified of the occurrence of the above. Here, the error calculation is not limited to the mean square error, and the allowable value of the error is not limited to 0.1, and any value may be set.

In such a configuration, first, the management device 5 commands the switching device 1 to output the input data as it is. Further, the management device 5 instructs the nerve cell network 3 to process and output the input data supplied from the switching device 1. The first monitoring device 15 monitors the input data and, for example, monitors whether the input data exceeds a preset range, and when it exceeds, notifies the management device 5 of an abnormality. The second monitoring device 14 monitors the target output signal and the output from the nerve cell network 3, and checks whether the error between the target output signal and the output of the nerve cell network exceeds a preset range, for example. Monitor and monitor device 5 if exceeded
Notify to abnormal.

The management device 5 includes a first monitoring device 15 and a second monitoring device 15.
When there is no notification of abnormality from the monitoring device 14, the instruction is given to cause the data input to the nerve cell network 3 to be processed as it is. When the management device 5 is informed of an abnormality from the first monitoring device 15, the management device 5 instructs the storage device 2 to store the abnormal input data, and the data stored in the storage device 2 is successively stored. And sends it to the switching device 1, and commands the switching device 1 to input the data to the neural cell network 3.
In addition, the management device 5 controls the switching device 1 for the nerve cell network 3.
It is instructed to perform learning with the data input through and the target output signal. This learning is performed a preset number of times. When the management device 5 is informed of an abnormality from the second monitoring device 14, the management device 5 takes out the data stored in the storage device 2 one after another and sends the data to the switching device 1, and at the same time sends the data from the switching device 1 to the nerve cell. Instruct to input to the network 3. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is repeated until there is no notification of abnormality from the second monitoring device 14.

An embodiment of the invention described in claim 7 will be described with reference to FIG. The information processing apparatus according to the present embodiment stores externally input data designated by the management apparatus 5, and stores the data of the storage content designated by the management apparatus 5 in the switching apparatus 1.
The storage device 2 for outputting to the neuron circuit network 3, and either the input data from the outside or the data from the storage device 2 is selected by a command from the management device 5 and output to the nerve cell network 3. The neural network 3 for processing the results and outputting the results, and monitoring the input data and managing the monitoring results by the management device 5.
To the first monitoring device 15, which outputs the output result from the neuron network 3 and the target output signal, and outputs the monitoring result to the management device 5, and the first monitoring device 15. It comprises a nerve cell network 3, a switching device 1 and a management device 5 which designates the operation of the storage device 2 by the outputs of the device 15 and the second monitoring device 14. Here, the storage device 2
May use a RAM, for example. Although the storage capacity of this RAM is not specified, a FIFO (first-in-first-out) method is used so that when this storage capacity becomes full, the oldest stored data is discarded and new data is overwritten.
It may be used in the first out). Switching device 1
For example, a data selector may be used. A comparator may be used for the first monitoring device 15, for example. In the second monitoring device 14, for example,
The mean square error between the target output value and the output value from the nerve cell network is calculated, and when the value becomes 0.1 or more, the management device is notified of the occurrence of the above. Here, the error calculation is not limited to the mean square error, and the allowable value of the error is not limited to 0.1, and any value may be set.

In such a configuration, first, the management device 5 commands the switching device 1 to output the input data as it is. Further, the management device 5 instructs the nerve cell network 3 to process and output the input data supplied from the switching device 1. The first monitoring device 15 monitors the input data and, for example, monitors whether the input data exceeds a preset range, and when it exceeds, notifies the management device 5 of an abnormality. The second monitoring device 14 monitors the target output signal and the output from the nerve cell network 3, and checks whether the error between the target output signal and the output of the nerve cell network exceeds a preset range, for example. Monitor and monitor device 5 if exceeded
Notify to abnormal.

The management device 5 includes a first monitoring device 15 and a second monitoring device 15.
When there is no notification of abnormality from the monitoring device 14, the instruction is given to cause the data input to the nerve cell network 3 to be processed as it is. When the management device 5 is informed of an abnormality from the first monitoring device 15, the management device 5 instructs the storage device 2 to store the abnormal input data, and the data stored in the storage device 2 is successively stored. And sends it to the switching device 1, and instructs the switching device 1 to input the data to the neural cell network 3. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is performed a preset number of times. When the management device 5 is informed of an abnormality from the second monitoring device 14, the management device 5 instructs the storage device 2 to store the abnormal input data, and the data stored in the storage device 2 is successively stored. And sends it to the switching device 1, and instructs the switching device 1 to input the data to the neural cell network 3. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the switching device 1 and the target output signal. This learning is repeated until there is no notification of abnormality from the second monitoring device 14.

An embodiment of the present invention will be described with reference to FIG. The information processing apparatus according to the present embodiment stores the input data designated by the management apparatus 5, and outputs the data of the storage content designated by the management apparatus 5 to the first switching apparatus 10 as a first storage apparatus 12. , A first switching device 10 that selects either input data from the outside or data from the first storage device 12 according to a command from the management device 5 and outputs it to the neural cell network 3, and the management device 5 A second target device which stores a target output signal designated by the control device 5 and outputs the data of the storage content designated by the management device 5 to the second switching device 11;
Second storage device 13 and a second output device that selects either a target output signal from the outside or data from the second storage device 13 according to a command from the management device 5 and outputs the selected signal to the neural cell network 3.
Switching device 11, a neural cell network 3 that processes input data and outputs the result, a first monitoring device 15 that monitors the input data and outputs the monitoring result to the management device 5, and a neural cell network. Monitor the output result from 3 and the target output signal,
The second monitoring device 14 that outputs the monitoring result to the management device 5.
And the outputs of the first monitoring device 15 and the second monitoring device 14, the neural network 3, the first switching device 10, the first storage device 12, the second switching device 11, and the second storage device. 1
3 and the management device 5 for designating the operation. Here, the first storage device 12 and the second storage device 1
For example, RAM 3 may use RAM. Although the storage capacity of the RAM is not specified, a FIFO (first i) is used so that when the storage capacity is full, the oldest stored data is discarded and new data is overwritten.
n first out). For the first switching device 10 and the second switching device 11, for example, a data selector may be used. A comparator may be used for the first monitoring device 15, for example. In the second monitoring device 14, for example, the mean square error between the target output value and the output value from the nerve cell network is calculated, and when the value becomes 0.1 or more, the above occurs in the management device. Let them know. Here, the error calculation is not limited to the mean square error, and the allowable value of the error is 0.1
However, any value may be set.

In such a configuration, first, the management device 5 commands the first switching device 10 to output the input data as it is. The management device 5 commands the second switching device 11 to output the target output signal as it is. or,
The management device 5 commands the neural cell network 3 to process and output the input data supplied from the first switching device 10. The first monitoring device 15 monitors the input data and, for example, monitors whether the input data exceeds a preset range, and when it exceeds, notifies the management device 5 of an abnormality. The second monitoring device 14 monitors the target output signal from the second switching device 14 and the output from the nerve cell network 3,
For example, it is monitored whether or not the error between the target output signal and the output of the nerve cell network exceeds a preset range, and if it exceeds, the management device 5 is notified of the abnormality.

The management device 5 includes the first monitoring device 15 and the second monitoring device 15.
When there is no notification of abnormality from the monitoring device 14, the instruction is given to cause the data input to the nerve cell network 3 to be processed as it is. When the management device 5 is notified of an abnormality from the first monitoring device 15, the first storage device 1
Instruct 2 to store the abnormal input data,
The target output signal at this time is stored in the second storage device 13, and the data stored in the first storage device 12 is taken out to the first switching device 10 and simultaneously to the second storage device 13. While taking out the stored data (target output signal) and sending it to the second switching device 11,
It is instructed to input those data to the nerve cell network 3 one after another. Further, the management device 5 commands the nerve cell network 3 to learn with the data input through the first switching device 10 and the target output signal input through the second switching device 11. This learning is performed a preset number of times. The management device 5 is the second monitoring device 14
When there is a notification of an abnormality from, the first storage device 12 is instructed to store the input data at this time,
The abnormal target output signal is stored in the second storage device 13, and the data stored in the first storage device 12 is taken out to the first switching device 10, and at the same time, the second storage device. The data (target output signal) stored in 13 is taken out and sent to the second switching device 11, and at the same time, the data is commanded to be input to the nerve cell network 3. In addition, the management device 5 is the nerve cell network 3
Is instructed to learn with the data input through the first switching device 10 and the target output signal input through the second switching device 11. This learning is repeated until there is no notification of abnormality from the second monitoring device 14.

An embodiment of the invention described in claim 9 will be described with reference to FIG. The ninth aspect of the invention (FIG. 6) is obtained by adding the recording device 16 to the eighth aspect of the invention, and the device having the same reference numeral is the same device as the above-described claim 8. The recording device 16 records the error information from the second monitoring device 14 and informs the management device 5 of the recorded contents. The management device 5 is claim 8.
If the information from the recording device 16 satisfies a certain condition in addition to the function in, the nerve cell network 3 is instructed to initialize the weight (coupling coefficient) of the nerve circuit connection to an arbitrary value. . This condition may be, for example, a case where the error information of the second monitoring device 14 continuously shows a large error.

Further, an expression device for notifying the monitoring result to the outside can be added to the monitoring device in the information processing apparatus according to the second to ninth aspects of the invention. As this expression device, for example, an LED, a display, a buzzer, a speaker sound source, or the like can be used to notify the outside of the information processing device by sound, light, a message, or the like.

[0047]

According to the first aspect of the present invention, since the storage device for storing the input data and the switching device for switching the input data and the data from the storage device are included, there is an error in the neural network. In this case, the neural network can be relearned by the input data stored in the storage device.

According to the second aspect of the invention, the storage device for storing the input data from the outside, and the input data stored in the storage device and the input data from the outside are switched to be input to the nerve cell network. When the output of the neural network is undesired, the switching device, the monitoring device that monitors the output from the neural network, and the management device that manages the neural network, the storage device, and the switching device are included. In addition, the neural network can be relearned by the input data stored in the storage device.

According to the third aspect of the invention, the storage device for storing the input data from the outside, and the input data stored in the storage device and the input data from the outside are switched to be input to the nerve cell network. The switching device, the monitoring device that monitors the input data, the neural network, the storage device, and the management device that manages the switching device. The neural network can be relearned by the input data stored in.

According to the fourth aspect of the invention, the storage device for storing the input data from the outside, and the input data stored in the storage device and the input data from the outside are switched and input to the neural cell network. The switching device, the first monitoring device that monitors the input data, the second monitoring device that monitors the output from the neural network, and the management device that manages the neural cell network, the storage device, and the switching device. Since it is configured, when the output of the neural network is undesired or the input data has never been input before, the neural network can be relearned by the input data stored in the storage device.

According to the fifth aspect of the present invention, when the monitoring device for monitoring the output from the nerve cell network notifies the abnormality, the input data is stored in the storage device. In addition to the effect of the invention described above, when there is an abnormality between the output value of the nerve cell network and the target output value, the data at that time is stored, so that the input data that could not be learned until now can be known. Can be used for re-learning. Moreover, the number of stored data can be reduced.

According to the sixth aspect of the invention, when the monitoring device for monitoring the input data notifies the abnormality, the input data is stored in the storage device.
In addition to the effects of the described invention, when the input data is abnormal, the data at that time is stored, so it is possible to know the input data that could not be learned until now, and use it for re-learning. You can Also, the number of stored data can be reduced.

According to the seventh aspect of the invention, when the monitoring device for monitoring the input data reports an abnormality, or when the monitoring device for monitoring the output from the nerve cell network reports the abnormality, By storing the input data, it is possible to obtain both the effects of the invention according to claims 5 and 6.

According to the eighth aspect of the present invention, the target output value is stored together with the input data when the monitoring device notifies the abnormality, so that it is necessary to input the target output value from the outside at the time of re-learning. lost.

According to the ninth aspect of the invention, the learning error of the neural network is recorded, and the coupling coefficient of the neural network can be set arbitrarily, so that the change of the learning state can be known. In addition, since the connection weight of the neural network can be initialized to an arbitrary value, it is possible to get out of the unlearnable state due to the inappropriate connection weight.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the invention described in claim 1.

FIG. 2 is a block diagram showing an embodiment of the invention described in claim 2.

FIG. 3 is a block diagram showing an embodiment of the invention according to claim 3;

FIG. 4 is a block diagram showing an embodiment of the invention described in claims 4, 5, 6, and 7.

FIG. 5 is a block diagram showing an embodiment of the invention described in claim 8;

FIG. 6 is a block diagram showing an embodiment of the invention according to claim 9;

[Explanation of symbols]

 1 ... Switching device, 2 ... Memory device, 3 ... Neuronal circuit network, 4, 6 ... Monitoring device, 5 ... Management device, 10 ... First switching device, 11 ... Second switching device Device, 12 ... First storage device, 13 ... Second storage device, 14 ... First monitoring device, 15 ... Second monitoring device, 16 ... Recording device.

Claims (9)

[Claims] 1. A nerve cell network capable of inputting data,
It is characterized by comprising a storage device for storing input data from the outside, and a switching device for switching the input data stored in the storage device and the input data from the outside to input to the neural cell network. Information processing equipment. 2. A nerve cell network capable of inputting data,
A storage device for storing input data from the outside, a switching device for switching the input data stored in the storage device and the input data from the outside to input to the neural cell network, and a switching device from the neural cell network An information processing device comprising: a monitoring device that monitors an output; and a management device that manages the nerve cell network, the storage device, and the switching device. 3. A nerve cell network capable of inputting data,
A storage device for storing input data from the outside, a switching device for switching the input data stored in the storage device and the input data from the outside to input to the neural cell network, and monitoring for monitoring the input data An information processing apparatus comprising: a device, a nerve cell network, a storage device, and a management device that manages the switching device. 4. A nerve cell network capable of inputting data,
A storage device for storing input data from the outside; a switching device for switching the input data stored in the storage device and the input data from the outside to input to the neural cell network; 1. An information processing system comprising: a monitoring device of No. 1; a second monitoring device for monitoring an output from a neural cell network; and a management device for managing the neural network, the storage device, and the switching device. apparatus. 5. The information according to claim 2 or 4, wherein when the monitoring device for monitoring the output from the nerve cell network reports an abnormality, the input data is stored in the storage device. Processing equipment. 6. The information processing apparatus according to claim 3, wherein the input data is stored in the storage device when the monitoring device for monitoring the input data reports an abnormality. 7. When the first monitoring device that monitors the input data reports an abnormality, or when the second monitoring device that monitors the output from the nerve cell network reports the abnormality, the input data is stored in the storage device. The information processing apparatus according to claim 4, wherein the information processing apparatus stores the information. 8. The target output value is stored together with the input data when the monitoring device reports an abnormality, and the target output value is also stored. The information processing apparatus according to claim 6 or 7. 9. An error in learning of a neural network is recorded,
The information processing apparatus according to claim 8, wherein the coupling coefficient of the nerve cell network can be arbitrarily set.