JP6781957B2 - Notification device, notification method, and its program - Google Patents

Description

The present invention relates to a notification device, a notification method, and a program thereof.

Conventionally, research on artificial intelligence technology such as neural networks (hereinafter referred to as "AI technology") has been widely conducted (see, for example, Patent Document 1). In particular, with the rise of AI technology called deep learning, products and services that implement AI technology have come to be seen in recent years, and are expected to be applied to a wider range of fields.

Japanese Patent No. 5816771

In machine learning such as deep learning, it is possible to train a machine to acquire a predetermined ability. The ability acquired as a result of learning can be used on another device via a storage medium or communication means by converting it into data as information on the structure and parameters of the learned neural network, for example. ..

However, when the learning device that performs learning and the device that uses the ability obtained as a result of learning are separated, the user of the learning result is made to perform what kind of learning using what kind of learning data. It is difficult to grasp whether it is the learning result obtained by this. In the future, as AI technology becomes widespread, when a wide variety of machine learning is performed in various places, a large number of learning results output by the learning device will automatically spread over a wide area without human intervention. There is a possibility, but it is expected that it will become more and more difficult to find new learning results that have acquired the abilities that users need.

Therefore, an object of the present invention is to provide a technique for detecting that a new ability has been acquired as a result of learning by machine learning and notifying a user who needs the ability.

The notification device according to one aspect of the present invention is acquired by a reception unit that accepts conditions specified by a client, an acquisition unit that acquires learning results obtained by performing predetermined learning by machine learning, and an acquisition unit. A judgment unit that determines whether or not the acquired ability of the learned learning result satisfies the conditions accepted by the reception unit, and a notification that notifies the requester when necessary based on the judgment result. It has a part. According to this aspect, it is possible to detect that a new ability has been acquired as a result of learning by machine learning and notify a user who needs the ability. Further, among the learning results output as the learning result in the learning device, when a predetermined condition is satisfied, the requester is notified, so that the amount of data to be communicated can be reduced.

The determination unit may determine whether or not the applicable range of the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit. The determination unit may determine whether or not the learning target of the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit. According to these aspects, when the acquired learning result deviates from the applicable range specified by the requester, it can be seen that notification is not required without further determination processing, so that the processing speed of the determination process is improved. At the same time, the processing process is shortened, so that the load on the processing device such as the CPU is reduced.

The judgment unit is based on whether or not the ability acquired by the learning result acquired by the acquisition unit is equivalent to the ability acquired by the predetermined learning result from the viewpoint of the conditions specified by the client. The determination may be made. The determination unit may determine whether or not the ability acquired by the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit. According to these aspects, since the amount of data requiring comparison and determination can be suppressed, the processing speed of the determination process is improved, and the processing process is shortened, so that the load on the processing device such as the CPU is reduced.

The notification device may be connected to a database that stores a plurality of learning results. At this time, the determination unit determines whether or not the acquired ability of the learning result acquired by the acquisition unit satisfies the condition specified by the client in comparison with the ability acquired by each of the plurality of learning results stored in the database. May be further determined. According to this aspect, since the requester is notified when a predetermined condition is satisfied as compared with the learning result existing in the past, the amount of data to be communicated can be reduced.

The inspection machine according to one aspect of the present invention can apply a learning result which is a result of a predetermined learning performed in order to acquire a predetermined ability. In addition, an inspection machine, a reception unit that accepts conditions specified by the client, an acquisition unit that acquires learning results obtained by performing predetermined learning by machine learning, and an acquisition unit that acquires learning results acquired by the acquisition unit. It is provided with a determination unit for determining whether or not the obtained ability satisfies the condition accepted by the reception unit, and a notification unit for notifying the requester when necessary based on the determination result. According to this aspect, the inspection machine itself detects that a new ability applicable to the inspection machine has been acquired as a result of learning by machine learning, and notifies the user who needs the ability. Becomes possible. Further, among the learning results output as the learning result in the learning device, when a predetermined condition is satisfied, the requester is notified, so that the amount of data to be communicated can be reduced.

The notification method according to one aspect of the present invention includes a step of accepting a condition specified by a client, a step of acquiring a learning result obtained by performing predetermined learning by machine learning, and a step of acquiring the acquired learning result. A step of determining whether or not the acquired ability satisfies the accepted condition and a step of notifying the requester when necessary based on the result of the determination are executed.

The program according to one aspect of the present invention is a means for accepting a condition specified by a client, a means for acquiring a learning result obtained by performing predetermined learning by machine learning, and a means for acquiring the acquired learning result. It functions as a means for determining whether or not the acquired ability satisfies the accepted condition, and as a means for notifying the requester when necessary based on the result of the determination.

According to the present invention, it is possible to provide a technique for detecting that a new ability has been acquired as a result of learning by machine learning and notifying a user who needs the ability.

It is a figure which shows the outline of the whole learning system 100 in this embodiment. It is a figure which conceptually shows the flow of processing by a learning system 100. It is a functional block diagram of the learning data preparation device 21 in this embodiment. It is a functional block diagram of the learning request device 22 in this embodiment. It is a functional block diagram of the learning result utilization apparatus 23 in this embodiment. It is a block diagram which shows the structure of the learning database 11 in this embodiment. It is a functional block diagram of the learning request reception device 12 in this embodiment. It is a functional block diagram of the learning apparatus 13 in this embodiment. It is a functional block diagram of the ability determination device 15 in this embodiment. It is a block diagram which shows an example of the hardware composition of each apparatus which constitutes a learning system 100. It is a flowchart of the ability determination process in the ability determination apparatus 15. It is a block diagram which shows an example of the structure of an inspection machine.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The same elements are designated by the same reference numerals, and duplicate description will be omitted. In addition, the following embodiments are examples for explaining the present invention, and the present invention is not intended to be limited only to the embodiments. Furthermore, the present invention can be modified in various ways as long as it does not deviate from the gist thereof.

<1. System overview>
The outline of the system in this embodiment will be described with reference to FIGS. 1 and 2.

FIG. 1 is a diagram showing an outline of the entire learning system 100 in the present embodiment. The learning system 100 includes a learning service providing system 1 and a learning request system 2, and both are connected to each other via a network N0 such as the Internet. The configuration of this system is not limited to the one shown in the figure. For example, the learning service providing system 1 and the learning request system 2 may be physically or logically integrated.

As shown in FIG. 1, the learning service providing system 1 has a learning database 11, a learning request receiving device 12, one or more learning devices 13, and an ability determination device 15, which form a local network N1. They are connected to each other via. Further, the learning request system 2 has a learning data preparation device 21, a learning request device 22, and one or more learning result utilization devices 23, which are connected to each other via a local network N2. In addition, each device is configured to be able to receive input from a learning requester or the like via an input device (not shown).

FIG. 2 is a diagram conceptually showing the flow of processing from the learning request to the use of the learning result by the learning system 100 shown in FIG. First, the client requests learning by designating a learning target (for example, identifying a specific object) to the learning requesting device 22 via the input device. When the learning request device 22 receives a learning request from the requester, the learning request device 22 accesses the learning service providing system 1 via the network N0, and provides the learning request information including the information necessary for performing machine learning to the learning service providing system 1. Send to. (S1). The learning request information includes, for example, a learning goal, information for identifying a device that uses the learning result, learning data, and the like.

When the learning request receiving device 12 of the learning service providing system 1 receives the learning request information, the learning request receiving device 12 instructs the learning device 13 to execute machine learning based on the learning request information (S2). The learning device 13 executes machine learning based on the learning request information. As a result, for example, a learning result such as a trained neural network that has acquired a predetermined ability can be obtained. The learning result is converted into replicable packaged data (S3). The data-converted learning result is transmitted from the learning device 13 to the learning result utilization device 23 of the learning request system 2 (S4). By using the learning result in the learning result utilization device 23, a predetermined ability is exhibited.

The learning result obtained by the learning device 13 is further stored in the learning database 11. As a result, any user of the learning system 100 can utilize the past learning results. In the present embodiment, when a new learning result is obtained by the learning device 13, the newly obtained learning result is compared with the past learning result stored in the learning database 11, and the learning obtained this time is compared. The ability determination device 15 detects whether or not the result has acquired a new ability as compared with the past learning result. When it is determined that a new ability has been acquired as a result of the detection, the ability determination device 15 notifies the user who needs the ability. As a result, the user can quickly grasp that new ability has been acquired as a result of machine learning.

<2. Configuration of each device>
<2-1. Learning request system 2>
Next, the configuration of the learning request system 2 will be described with reference to FIGS. 3 to 5. The configuration of each device included in the learning request system 2 is not limited to the configuration described below, and each device may be modified so as to have an arbitrary function provided by the other device, if necessary.

FIG. 3 is a functional block diagram of the learning data preparation device 21 according to the present embodiment. The learning data preparation device 21 has a function of preparing data (learning data) necessary for training a learning module (for example, a neural network 233 described later). As shown in FIG. 3, the learning data preparation device 21 has an operation unit 211, a learning data acquisition unit 212, a learning data storage unit 213, and a data acquisition control unit 214 as functional units.

For example, the operation unit 211 receives an operation from a user of the learning request system 2 (hereinafter, also referred to as a “requester”). The learning data acquisition unit 212 acquires data necessary for creating learning data from an arbitrary input device such as a camera, a sensor, a network terminal, or a sensor of a self-propelled robot, and stores the data in the learning data storage unit 213. The data acquisition control unit 214 controls the operation unit 211, the learning data acquisition unit 212, and the learning data storage unit 213 to prepare data necessary for learning. The communication unit 216 connects to the local network N2 of the learning request system 2 and transmits the data required for learning created by the data acquisition control unit 214 to another device.

The learning data preparation device 21 may be constructed by using the same device as the learning result utilization device 23 described later. In this case, the learning data preparation device 21 can be constructed as an input device externally connected to the learning result utilization device 23.

FIG. 4 is a functional block diagram of the learning request device 22 according to the present embodiment. The learning request device 22 has a function of transmitting learning request information to the learning service providing system 1. As shown in FIG. 4, the learning request receiving device 12 has a learning request unit 221, a learning request content storage unit 222, a learning data storage unit 223, and a communication unit 224.

The learning request unit 221 receives a learning request from a requester via a learning data preparation device 21 or an input device (not shown), creates learning request information, and provides learning request information via a communication unit 224 as a learning service providing system. Send to 1. The learning request information transmitted at this time is stored in the learning request content storage unit 222. In addition, the learning data required for performing the learning requested by the client is acquired from the learning data preparation device 21 and transmitted to the learning service providing system 1. The learning data transmitted to the learning service providing system 1 is also stored in the learning data storage unit 223.

Further, the learning request unit 221 receives the notification request specified by the requester from the input device, creates the notification request information, and transmits the notification request information to the learning service providing system 1 via the communication unit 224. The notification request information includes, for example, at least one of the conditions such as the ability required by the requester, the scope of application in which the requester uses the learning result, the learning target, and the output requirement. The notification request information transmitted at this time is stored in the learning request content storage unit 222. For example, when the client requests notification of the ability to classify agricultural products, the notification request information includes information that the condition of the ability required by the client is the classification of agricultural products. In addition, the notification request information may include classifying tomato grades by further narrowing down the learning target or the scope of application. As a required ability, classification of grades into 5 grades may be included in the notification request information, or as an output requirement, a condition of notification that the classification accuracy is improved by a predetermined point or more as compared with the existing ability. Good.

Further, when the learning result is given the identification information, the learning request device 22 refers to the identification information given to the learning result and determines whether or not the input learning result is suitable for the purpose of use. You may. When the learning result is suitable for the purpose of use, the learning request device 22 incorporates the learning result into the learning result utilization device 23 and makes the learning result used. On the other hand, when the learning result does not match the purpose of use, the learning requesting device 22 can notify the learning service providing system 1 of the nonconformity. At this time, the learning requesting device 22 may request learning again under different conditions.

FIG. 5 is a functional block diagram of the learning result utilization device 23 in the present embodiment. The learning result utilization device 23 has a function of providing a predetermined ability to the user by utilizing the learning result. As shown in FIG. 5, the learning result utilization device 23 has a learning result input unit 231, a neural network setting unit 232, a neural network 233, a control unit 234, an input unit 235, and a communication unit 236 as functional units. And a data acquisition unit 237 and an output unit 238.

The learning result input unit 231 accepts the input of the learning result. At this time, the neural network setting unit 232 sets the neural network 233 according to the purpose of use. Further, the control unit 234 controls the data acquisition unit 237 and the input unit 235 to input the data necessary for using the learning result into the neural network 233 and use the learning result. The result of using the learning result is output from the output unit 238.

<2-2. Learning service provision system 1>
The function of the learning service providing system 1 will be described with reference to FIGS. 6 to 10. The learning service providing system 1 can be implemented using a data center or a cloud. In this case, each device of the learning service providing system 1 can be constructed by using a PC server or a blade PC. By constructing each device of the learning service providing system 1 with a plurality of PCs, it is possible to shorten the processing time when executing repetitive operations such as deep learning. The learning service providing system 1 may have a configuration constructed by one PC or a configuration implemented by an embedded device.

FIG. 6 is a block diagram showing the configuration of the learning database 11 in the present embodiment. The learning database 11 stores various information required for learning. As shown in FIG. 6, the learning database 11 has a learning data DB 111, a learning request DB 112, a learning result usage history DB 113, a learning result DB 114, a learning program DB 115, and a learning target DB 116.

The learning data DB 111 stores learning data used for learning. For example, the learning data DB 111 can store the learning data in association with the learning requirements such as the target of the learning data, the breakdown of the learning data, the range of the learning data, and the purpose of learning.

The learning request DB 112 stores the learning request information and the content of learning to be performed when the learning request information is received from the learning request device 22. For example, the learning request DB 112 can store information about a learning requester, a target of learning data, a breakdown of learning data, a range of learning data, a purpose of learning, and other learning request requirements in association with the learning request information.

The learning result usage history DB 113 stores the learning result usage history. For example, the learning result usage history DB 113 can store information about the result of classification by using the acquired classification ability when the classification ability is acquired as a learning result. Further, the learning result usage history DB 113 is related to the use of the learning result in relation to the information about the user of the learning result, the target of the learning data, the breakdown of the learning data, the range of the learning data, the purpose of learning, and the like. Information can be stored. Further, the learning result usage history DB 113 preferably includes usage identification information. The usage identification information is information that can identify the use of the learning result, and includes, for example, information related to settings (basic factors, influencing factors, etc., which will be described later) that affect the ID and ability of the device to be used. Since the acquired ability may change depending on the device to be used and the environment in which it is used, the usage history can be recorded in more detail by including the usage identification information in the learning usage history DB 133.

The learning result DB 114 stores the learning results output by the learning device 13, and stores a plurality of past learning results. The learning result stored in the learning result DB 114 includes the learning result digitized. Further, when the identification information is given to the learning result, the information stored in the learning result DB 114 includes the identification information of the learning result.

The learning program DB 115 stores a learning program for performing learning. For example, the learning program DB 115 can store a learning program in association with learning requirements such as a learning target, contents of learning data, and learning goals. It is preferable that a large number of learning programs can be registered in the learning program DB 115. In this case, the learning device 13, which will be described later, can identify the learning program from the learning program DB 115 and make it executable by designating the learning requirements.

The learning target DB 116 stores information about the learning target. The learning target may include, for example, an object identification device for identifying the type of object, an image processing device, a product management device, a robot, a sensor signal prediction device, an in-vehicle device, and the like.

FIG. 7 is a functional block diagram of the learning request receiving device 12 according to the present embodiment. The learning request receiving device 12 has a function of receiving learning request information from the learning request device 22 and transmitting the learning request information to the learning device 13. As shown in FIG. 7, the learning request receiving device 12 includes, for example, a learning request receiving unit 121, a learning data storage unit 123, a learning request content storage unit 124, and a communication unit 125.

The learning request reception unit 121 receives a learning request from the learning request device 22. The learning request reception unit 121 registers the learning request information included in the received learning request in the learning request DB 112. At this time, the learning request receiving unit 121 transmits a notification to the effect that the learning request has been accepted to the learning management device 14 via the communication unit 125. Further, the learning request receiving device 12 can temporarily store the received learning request in the learning data storage unit 123 or the learning request content storage unit 124.

FIG. 8 is a functional block diagram of the learning device 13 according to the present embodiment. The learning device 13 has a function of performing learning based on learning request information and acquiring a predetermined ability as a learning result. As shown in FIG. 8, the learning device 13 has a learning control unit 131, a neural network 132, a learning result extraction unit 133, a communication unit 134, and a learning result output unit 135 as functional units. ..

In the learning device 13, the learning control unit 131 can control the neural network 132 and perform learning based on the learning request information. The learning result is extracted by the learning result extraction unit 133, and is output by the learning result output unit 135 via the communication unit 134. When learning is performed a plurality of times in the learning device 13, a plurality of learning results can be obtained. In each learning, if the requirements that influence the learning result such as learning data, learning program, learning time, and learning goal are different, the abilities obtained by multiple learning results may not be the same. When the identification information is given to the learning result, if there is a possibility that the abilities obtained by the plurality of learning results are not the same, different identification information may be given and managed as a variation.

The learning device 13 can also make a learning request by itself. In this case, the learning device 13 enables autonomous learning.

FIG. 9 is a functional block diagram of the capability determination device 15 according to the present embodiment. The ability determination device 15 has a function of determining whether the acquired ability of the learning result output as a result of learning in the learning device 13 satisfies a predetermined condition. As shown in FIG. 9, the ability determination device 15 includes an ability determination control unit 151, a comparison target selection unit 152, a learning result comparison unit 153, an identification information generation unit 154, and a communication unit 155. ..

The ability determination control unit 151 controls each unit included in the ability determination device 15 to perform the ability determination process described later. The ability determination control unit 151 further includes a reception unit 1511, an acquisition unit 1512, a determination unit 1513, and a notification unit 1514. The reception unit 1511 receives the notification request information including the conditions specified by the requester from the learning request device 22. The acquisition unit 1512 acquires the learning result obtained by performing the predetermined learning from the learning device 13. The determination unit 1513 uses the comparison target selection unit 152 and the learning result comparison unit 153 to determine whether the ability acquired by the learning result acquired by the acquisition unit 1512 is a new ability that satisfies the conditions included in the notification request information. Judge whether or not. When it is determined that a new ability has been acquired based on the conditions included in the notification request information, the notification unit 1514 notifies the requester via the communication unit 155. The comparison target selection unit 152 selects a learning result to be compared with the learning result obtained this time from a plurality of learning results stored in the learning database 11. When the notification request information includes a condition regarding the scope of application and a condition regarding the learning target, the selection process is performed based on these conditions.

The learning result comparison unit 153 compares the learning result acquired from the learning device 13 with one or more learning results selected by the comparison target selection unit 152, and whether or not a new ability is acquired by the learning result obtained this time. Is determined. When the notification request information includes a condition related to the output requirement, the determination process is performed based on this condition.

The identification information generation unit 154 assigns identification information to a learning result that has acquired a predetermined ability by learning. Here, adding identification information to the learning result is a concept including associating the learning result with the identification information. For example, adding the identification information to the learning result or transmitting the learning result and the identification information to the storage device. Includes associating and storing. In addition to directly associating the learning result with the identification information, it also includes indirectly associating these with each other.

<2-3. Hardware configuration of each device>
FIG. 10 is a block diagram showing an example of the hardware configuration of each device constituting the learning system 100. As shown in the figure, each device such as the capability determination device 15 includes a memory such as a CPU 1010, a ROM 1020 or a RAM 1030, a storage device 1040 for storing various information, an input / output unit 1050, a communication unit 1060, and a network connecting them. A general-purpose or dedicated computer equipped with a bus can be applied.

The functions implemented in each device are not limited to the configuration realized by the CPU executing a predetermined program stored in the memory or the storage device. Any function included in each device may be configured to be realized by hardware. For example, the neural networks 132 and 233 described later may be configured by electronic circuits such as a custom LSI (Large-Scale Integration) and an FPGA (Field-Programmable Gate Array). Further, a part of each device may be configured to be realized by a physical key, a biochemical method such as DNA, or an identification means by an optical method such as a hologram. The program can be installed or loaded on a computer by downloading it through various recording media such as an optical disk such as a CD-ROM, a magnetic disk, or a semiconductor memory, or via a communication network or the like.

<3. Processing flow>
Next, the processing flow of the system according to the present embodiment will be described with reference to FIG.

FIG. 11 is a flowchart of the capacity determination process in the capacity determination device 15. First, the reception unit 1511 of the ability determination control unit 151 receives the notification request information of the requester from the learning request device 22 (S1110). The received notification request information is stored in the storage device in association with information such as the notification destination of the requester.

After that, when the learning device 13 outputs a new learning result based on the learning request from the requester or another requester, the acquisition unit 1512 acquires the learning result output from the learning device 13 (S1120). Then, the determination unit 1513 uses the comparison target selection unit 152 and the learning result comparison unit 153 to determine whether or not the ability acquired by the acquired learning result satisfies the condition included in the notification request information, that is, whether or not the requester's ability is satisfied. It is determined whether or not the required new ability has been acquired (S1130).

Specifically, the determination unit 1513 executes a determination process composed of steps S1131 to S1134 described below. It should be noted that it is not always necessary to execute all the processes of steps S1131 to S1134, and the necessary processes may be selectively executed according to the conditions included in the notification request information. Further, the order of processing is not limited to the order described below, and can be arbitrarily changed as long as the processing contents do not conflict with each other.

First, the comparison target selection unit 152 performs a range limiting process on the learning result acquired in step S1120 or a plurality of learning results stored in the learning database 11 (S1131). The ability determination device 15 may be able to compare and determine abilities within a range that can be used by each client (hereinafter, also referred to as "applicable range"). For example, even with the same type of capability, the range of use will differ if the input or output configuration (number of inputs / outputs, types of input / output data, etc.) is different. Alternatively, it is not necessary to determine whether the learning result that is not available to the client is a new ability. The requester can specify the applicable range under arbitrary conditions when requesting notification. For example, the applicable range may be set for each group of learning programs that can correspond to the same learning target. Further, the applicable range may be set for a plurality of learnings using data of the same requester and the same device for different periods. When the learning result satisfies the applicable range, the specific process of step S1132 is executed. On the other hand, if the learning result does not satisfy the applicable range, the determination process is terminated.

To specify the applicable range, the applicable range specifying condition is used. The applicable range specifying condition is a condition for specifying whether or not it is included in the applicable range, and can be set for, for example, a target of the ability to be acquired. The applicable range specifying condition is stored in the applicable range specifying condition storage unit. As an example of the applicable range specifying condition, the applicable range identification information can be used. For example, identification information that can identify the input / output configuration and the goal of the ability to be acquired may be given to the learning result, and the learning result to which the same identification information is given may be specified to have the same applicable range. .. As another example of the applicable range specifying condition, the range equivalence judgment logic for judging the equivalence of the applicable range is stored, and whether or not the learning result is included in the applicable range is specified by using the logic. You may let me. The specific implementation example of the applicable range specifying condition is not limited to these, and any other condition can be used.

Next, the comparison target selection unit 152 performs a learning target identification process on the learning result selected by the application range limiting process (S1132). Even if it is included in the scope of application, if the learning target is different, even if the learning result is different, it is not the new ability required by the client. Therefore, it is necessary to determine and specify whether the learning target is the learning target for which the acquisition of new ability is to be detected. When the learning result satisfies the condition of the learning target, the determination process of step S1133 is executed. On the other hand, if the learning result does not satisfy the condition of the learning target, the determination process is terminated.

The learning results can be compared for each learning target. For example, a learning device that classifies images and a device that predicts sensor signals have different learning results because the targets are different. It is possible to determine whether it is a new ability between the two, but it is less useful from the viewpoint of system utilization. The processing speed can be improved by excluding the learning results obtained for different objects from the comparison targets and comparing the learning results obtained for the same object.

The learning target specifying condition is used to specify the learning target. The learning target specific condition can be set for a predetermined learning set. The learning target specific condition is stored in the learning target specific condition storage unit. As an example of the learning target specifying condition, the equivalence of the learning target may be specified by adding identification information to the learning target and designating it in advance. In addition, the equivalence of the learning target may be specified by using the logic for determining the equivalence of the learning target.

By specifying the learning target, it is possible to compare the learning results performed on the same learning target. In this case, it is possible to compare a plurality of learning results having the same learning target and compare how different the learning results are. For example, by giving equivalent inputs to two learning results with the same learning target and seeing how the outputs differ, whether they have the same ability or a significant difference. It is possible to determine whether or not it has a certain ability. Further, when a plurality of learnings are performed on the same learning target, a plurality of learning results can be obtained. After that, when new learning is performed on the same learning target, a new ability that is not the same as any of the previous learning results and has a significant difference from any of the learning results. May be obtained.

When determining the equivalence of the learning targets, the learning results may be specified not only if the learning targets are the same but also if the learning targets are similar. As a result, it is possible to cope with the case where the learning result of the same learning target is small or the range of the learning target can be expanded.

Here, an example of the applicable range and the learning target will be specifically described. In the present embodiment, the applicable range and the learning target can be set for each learning program. Learning targets include things and things that are targeted by the abilities acquired as a result of learning. For example, if the learning result is the ability to classify agricultural products, then the learning target is agricultural products. At this time, the ability to classify the grades of agricultural products to be learned may differ depending on the learning program and the range of learning data used for learning. For example, a learning result dedicated to cucumber or tomato, or a learning result applicable to a plurality of kinds of fruit vegetables may be output, but in this case, the applicable range is cucumber, tomato, and fruit vegetables, respectively. That is, the applicable range includes the range to which the learning result can be applied. The scope of application may be defined hierarchically.

It is preferable that the scope of application and the definition of the learning target are set in common in the entire learning program because it is easy to find new abilities, but it is not always necessary to set the definition in common in the entire learning program. If the user of the learning result is specified or limited, such as learning for a specific customer, the scope of application and the definition of the learning target may be set individually within the range. For example, in the learning program A, the learning target is set to agricultural products and the applicable range is fruit vegetables, while in the learning program B, the learning target is fruit vegetables and the applicable range is tomatoes according to the data used for learning. Then, the learning result of fruit vegetables and the applicable range of cucumber may be output. Further, the learning target may be set to tomato, and the learning range may be set to tomato.

As another example, if the learning result is the ability to create a summary of a news article, then the learning target is a news article. In this case, for example, learning results dedicated to domestic news, learning results dedicated to economic news, and learning results covering all news that can be applied to any field are assumed. Further, when the learning result is the control of the assembly device for a specific customer, the following cases can be assumed. For example, the learning target is the control of the assembly device of a specific product, and the production lines are 1 to 5. At this time, when the learning result 1 corresponds only to the production line 1, the applicable range of the learning result 1 is the production line 1. Further, when the learning result 2 corresponds to the production lines 2 to 5, the applicable range of the learning result 2 is the production lines 2 to 5.

Next, the learning result comparison unit 153 determines the equivalence of the learning results (S1133). That is, it is determined whether or not it is equivalent to the learning result selected by the comparison target selection unit 152. If it is determined that they are equivalent, the output requirement determination process of step S1134 is executed. On the other hand, if it is determined that they are not equivalent, the determination process is terminated.

In general, when machine learning is performed, if the learning conditions are slightly different, the learning results will be slightly different. Therefore, when new machine learning is performed, it is determined whether or not it is equivalent to the previously performed learning, and when it is determined that the learning is not equivalent, the result of the current machine learning is new. It may be considered that the ability has been obtained. In addition, when the learning program uses random numbers internally, differences may occur in the learning results when learning is performed multiple times under exactly the same conditions except for random numbers. At this time, the slightly different abilities can be determined as new abilities. However, in that case, the value of the information that it is a new ability may be greatly impaired because most of the cases are almost the same when notified of the new ability and confirmed.

In order to avoid such inconvenience, a process of comparing equivalence, that is, comparing whether or not a plurality of learning results are equivalent is performed. For example, the equivalence is compared by comparing the learning results obtained in the current learning with the past learning results narrowed down based on the applicable range and the learning target.

Here, the determination of the equivalence of the learning results is performed using the learning result equivalence determination conditions. The learning result equivalence determination condition is stored in the learning result equivalence determination condition storage unit. The learning result equivalence judgment conditions will be described below.

First, the conditions or criteria for comparing equivalence (hereinafter, also simply referred to as "comparison conditions") differ depending on the type of ability. Equivalence comparison conditions are registered in the storage device in advance for each type of ability. Specifically, for example, comparison conditions are set for each learning program or for each group of similar learning programs. It may be done according to the learning requester.

If the requirements expected as a result of learning are included as notification conditions, it is determined whether or not the requirements are met. When the degree of ability is included as a notification condition, for example, it is determined whether or not the degree of improvement of the item selected under the predetermined condition is equal to or higher than the predetermined condition.

When comparing a plurality of learning results, if the types of learning data used for learning are different, it is possible to determine the degree of difference between the learning results by using the internal information of the learning program. When the types of training data are the same but the data used for training are different, it is possible to determine the degree of difference in each learning result by comparing the difference in ability according to the difference in training data. it can. When the output of learning results using the same learning data is different, it is possible to determine the degree of difference in each learning result by setting comparison conditions according to the difference for each different content. As a standard for comparing learning results, for example, abilities such as classification ability, prediction ability, control ability, and language ability obtained as a result of learning can be compared.

Next, as a result of the determination of equivalence, when it is determined that the learning result acquired this time is not equivalent to the past learning result, the learning result comparison unit 153 performs the output requirement determination process (S1134). That is, in order to be notified of the ability expected by the requester, a process for determining whether or not to notify the requester is performed according to a predetermined requirement different from the condition for determining the equivalence of learning results. If all the learning results that are judged to be not equivalent according to the equivalence judgment conditions are notified to the outside, there is a possibility that the ability not expected by the requester will be notified, and if many unnecessary notifications are output, the requester This is not desirable because it may impose an unnecessary burden on the user.

In the output requirement determination process, whether or not the learning result satisfies the output requirement is performed using the output requirement specified in advance by the client. The output requirements included in the notification request information are stored in the output requirement judgment condition storage unit. Output requirements include, for example, constraint requirements related to the learning result utilization device such as the scale of the neural network and response time when incorporated into the learning result utilization device, and input constraints such as camera resolution, sensor response cycle, and sensor accuracy. Requirements, output constraint requirements such as the number of classified outputs and output cycle, and target constraint requirements can be set.

As a result of the determination process by the determination unit 1513, when it is determined that the new ability required by the requester has been acquired, the notification unit 1514 notifies the requester to that effect (S1140).

The content of the notification includes information for identifying the learning result. For example, the notification content may include information indicating the content of learning, such as information on the learning target, information on the scope of application, information on the judgment of equivalence, and information on the output requirement. In addition, the notification content may include identification information indicating a predetermined learning result. By prescribing the identification information to be notified when the learning result satisfying the predetermined conditions is obtained with the client, the new ability expected by the client can be acquired with less information. Can be notified promptly.

The target of the notification may include the requester who requested the notification, the person concerned with the learning target, the developer of the learning program, the learning requester, and the like. The notification target is stored in the notification target storage unit. In addition to the requester of the notification request, the person selected in the notification target selection unit is stored in the notification target storage unit as the notification target.

<4. Usage example>
An example of using the ability determination process of the learning system 100 will be described.

In the target identification device for identifying the type of target, for example, when the misjudgment rate is improved by 20% or more, when it becomes possible to cope with a new crop, or when it becomes possible to identify using a three-dimensional shape. , When the learning data is greatly increased and the reliability is improved, when it becomes possible to identify a new kind of animal, etc., the client is informed that new abilities have been acquired by the learning results acquired this time. Notice.

In an image processing device, for example, when it becomes possible to handle a larger image, the number of gradations that can be handled increases, the number of types to be classified increases, the number of objects that can be processed in real time increases, and the like. Notify the requester.

In product management devices such as visual inspection devices, for example, when the matching rate with a human sensory test is improved, or when it becomes possible to respond to a new input, a combination of three images from an inspection of only one image is used. Notify the requester when it becomes possible to inspect.

In robots, for example, when the ability to perform a new motion is acquired, when the ability is improved, when the motion speed is improved, when the motion accuracy is improved, when the motion evaluation result is improved, etc. Notify to.

In the prediction device, for example, when it becomes possible to predict the next data or state based on the input data up to the previous time, the requester is notified.

For in-vehicle devices, for example, when the number of types that can be identified by the environment sensor increases, when the misjudgment rate such as identification of tunnels and overpasses improves, or when the non-detection rate for events that may occur improves, etc. Notify the person.

In the machine player of the game, for example, when the experience value or skill is improved, or when it becomes possible to correspond to a new game, the requester is notified.

<5. Example of inspection machine>
An example in which the notification device according to the present embodiment is applied to the inspection machine will be described. For example, the manufacturer Z of an electronic circuit board uses an inspection machine to sort out non-defective products and defective products. In this example, the manufacturer Z is the notification requester, and the inspection machine acquires the learning result acquired by learning the ability to distinguish between good and defective products, and a new ability to satisfy the conditions included in the notification request. The case where the manufacturer Z is notified when it is determined that the item has been acquired and the learning result is applied to the inspection machine will be described.

First, the configuration of the inspection machine a will be described with reference to FIG. FIG. 12 is a block diagram showing an example of the configuration of the inspection machine a. As shown in FIG. 12, the inspection machine a includes a storage unit A1, an input unit A2, a discrimination unit A3, and an output unit A4. The storage unit A1 stores, for example, the learning result of learning performed by the learning service providing system 1. The input unit A2 has a function of acquiring information to be inspected, and is composed of, for example, a sensor such as a camera or a microphone. The discrimination unit A3 has a function of discriminating a predetermined inspection target such as a non-defective product or a defective product based on the information (for example, an image) of the inspection target input from the input unit A2. Further, the output unit A4 has a function of outputting the determination result of the determination unit A3. Note that outputting the judgment result here means not only presenting the judgment result to the user, such as displaying the judgment result on a display device or the like, but also an inspection machine for selecting a predetermined inspection target from other inspection targets. It includes the operation of. The inspection machine a further includes a reception means, an acquisition means, a determination means, and a notification means unit, in addition to the above-mentioned configuration as the inspection machine. The reception means, acquisition means, determination means, and notification means have the same functions as the reception unit 1511, the acquisition unit 1512, the determination unit 1513, and the notification unit 1514 in the above embodiment.

Next, a process of determining and notifying that the processing inspection machine a has acquired a new ability will be described. First, when the manufacturer Z requests the inspection machine a to notify the inspection machine a when a learning result satisfying a predetermined condition is acquired, the reception means of the inspection machine a receives the notification request information and stores it in the storage unit. Remember. For example, when the manufacturer Z determines a non-defective product or a defective product based on a plurality of images of an electronic circuit board captured by a camera provided in the inspection machine a, the manufacturer Z uses a learning model mounted on the inspection machine a. Also makes a notification request on condition that the correct answer rate improves by 10% or more.

The acquisition means of the inspection machine a acquires the learning result via the network on a regular or irregular basis. For example, the learning service providing system 1 is accessed, and the learning result and its identification information stored since the last access from the learning database 11 are acquired.

The determination means of the inspection machine a determines whether or not the acquired learning result satisfies the condition included in the notification request information, and if the condition is satisfied, the manufacturer Z is determined by the notification means. Notify to. In response to an instruction or the like from the manufacturer Z, the inspection machine a can replace the learning model applied to the inspection machine a with the learning result newly acquired from the currently applied one. As described above, by applying the notification device according to the present embodiment to the inspection machine, the inspection machine a can autonomously update the learning model to be mounted, and the reliability of the inspection machine is further improved.

The embodiment of the present invention has been described above. It should be noted that the present embodiment is for facilitating the understanding of the present invention, and is not for limiting and interpreting the present invention. Further, the present invention can be changed or improved without deviating from the gist thereof. Further, each step in the above-mentioned processing flow can omit a part of each step within a range that does not cause a contradiction in the processing contents, or can arbitrarily change the order of each processing step or execute them in parallel.

In the above-described embodiment, the learning request and the notification request have been described as separate ones, but the learning request may include the notification request. That is, the learning request device 22 creates learning request information and notification request information based on the request content specified by the requester, and transmits the learning request information and the notification request information to the learning service providing system 1. Subsequent processing is as described above, and the client can be notified after receiving the learning result based on the learning request, and when a new ability is acquired for the ability required by the client.

In the present specification, the "part" or "means" does not simply mean a physical configuration, but also includes a case where the function of the "part" is realized by software. Further, even if the function of one "part" or "means" or device is realized by two or more physical configurations or devices, the function of two or more "parts" or "means" or device is one. It may be realized by physical means or equipment.

In addition, some or all of the above embodiments may be described as in the following appendix, but are not limited to the following.
(Appendix 1)
With at least one hardware processor
The hardware processor
Accept the conditions specified by the requester,
Acquire the learning result obtained by performing the predetermined learning by machine learning,
It is determined whether or not the acquired ability of the acquired learning result satisfies the accepted condition.
A notification device comprising notifying the requester when necessary based on the result of the determination.
(Appendix 2)
By at least one or more hardware processors
Steps to accept the conditions specified by the requester and
The step of acquiring the learning result obtained by performing the predetermined learning by machine learning, and
A step of determining whether or not the acquired ability of the acquired learning result satisfies the accepted condition, and
Based on the result of the determination, if necessary, the step of notifying the requester and
Notification method to execute.

1 Learning service provision system, 2 Learning request system, 11 Learning database, 12 Learning request reception device, 13 Learning device, 14 Learning management device, 15 Ability judgment device, 21 Learning data preparation device, 22 Learning request device, 23 Learning result utilization Device, 100 learning system, 121 learning request reception unit, 123 learning data storage unit, 124 learning request content storage unit, 125 communication unit, 131 learning control unit, 132 neural network, 133 learning result extraction unit, 134 communication unit, 135 learning Result output unit, 151 ability judgment control unit, 1511 reception unit, 1512 acquisition unit, 1513 judgment unit, 1514 notification unit, 152 comparison target selection unit, 153 learning result comparison unit, 154 identification information generation unit, 155 communication unit, 211 operation Unit, 212 Learning data acquisition unit, 213 Learning data storage unit, 214 Data acquisition control unit, 216 Communication unit, 221 Learning request unit, 222 Learning request content storage unit, 223 Learning data storage unit, 224 Communication unit, 231 Learning result input 232 Neural network setting unit, 233 Neural network, 234 control unit, 235 input unit, 236 communication unit, 237 data acquisition unit, 238 output unit, 1010 CPU, 1020 ROM, 1030 RAM, 1050 storage device, 1050 input output unit 1060 Communication Department

Claims (8)

The reception department that accepts the conditions specified by the client,
An acquisition unit that acquires the learning results obtained by performing predetermined learning by machine learning,
A determination unit that determines whether or not the ability acquired by the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit.
Based on the result of the determination, a notification unit that notifies the requester when necessary,
Equipped with a,
The determination unit is a notification device that determines whether or not the applicable range of the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit . The determination unit determines whether or not the learning target of the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit.
The notification device according to claim 1. The determination unit determines based on whether or not the ability acquired by the learning result acquired by the acquisition unit is equivalent to the ability acquired by the predetermined learning result from the viewpoint of the conditions. I do,
The notification device according to claim 1 or 2 . The determination unit determines whether or not the acquired ability of the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit.
The notification device according to any one of claims 1 to 3 . Connected to a database that stores multiple learning results
The determination unit further determines whether or not the acquired ability of the learning result acquired by the acquisition unit satisfies the above condition by comparing with the ability acquired by each of the plurality of learning results stored in the database. To do
The notification device according to any one of claims 1 to 4 . It is an inspection machine that can apply the learning result that is the result of the predetermined learning performed to acquire the predetermined ability.
The reception department that accepts the conditions specified by the client,
An acquisition unit that acquires the learning results obtained by performing predetermined learning by machine learning,
A determination unit that determines whether or not the ability acquired by the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit.
Based on the result of the determination, a notification unit that notifies the requester when necessary,
Equipped with a,
The determination unit is an inspection machine that determines whether or not the applicable range of the learning result acquired by the acquisition unit satisfies the condition accepted by the reception unit . The computer
Steps to accept the conditions specified by the requester and
The step of acquiring the learning result obtained by performing the predetermined learning by machine learning, and
A step of determining whether or not the acquired ability of the acquired learning result satisfies the accepted condition, and
A step of determining whether or not the applicable range of the acquired learning result satisfies the accepted condition, and
Based on the result of the determination, if necessary, the step of notifying the requester and
Notification method to execute. Computer,
Means of accepting conditions specified by the requester,
A means of acquiring learning results obtained by performing predetermined learning by machine learning,
A means for determining whether or not the acquired ability of the acquired learning result satisfies the accepted condition, and a notification to the requester when necessary based on the result of the determination. means,
To function as,
The determination means is a program that determines whether or not the applicable range of the learning result acquired by the acquisition means satisfies the condition accepted by the accepting means .