WO2018168908A1 - Notification apparatus, notification method and computer program therefor - Google Patents

Abstract

Provided is a technology for detecting that a new capability has been procured as a result of having performed learning by machine learning, and notifying a user who requires the capability. A notification apparatus is provided with an acceptance unit that accepts a condition designated by a requester, an acquisition unit that acquires a learning result obtained due to predetermined learning being performed by machine learning, a determination unit that determines whether a procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit, and a notification unit that notifies the requester if necessary, based on a result of the determination.

Description

NOTIFICATION APPARATUS, NOTIFICATION METHOD AND COMPUTER PROGRAM THEREFOR

The present invention relates to a notification apparatus, a notification method, and a computer program therefor.
CROSS-REFERENCES TO RELATED APPLICATIONS
This application claims priority to Japanese Patent Application No. 2017-049141 filed March 14, 2017, the entire contents of which are incorporated herein by reference.

Heretofore, research relating to artificial intelligence technologies (hereinafter, "AI technologies") such as neural networks has been widely carried out (e.g., see JP 5816771). In particular, goods and services that implement AI technologies have become prominent in recent years due to the rise of an AI technology known as deep learning, and application in a broad range of fields is expected.

JP 5816771 is an example of background art.

With machine learning such as deep learning, a machine can be trained to procure a predetermined capability. The capability procured as a result of learning can, for example, be utilized on another apparatus via a storage medium or a communication means, by being digitized as information relating to the structure and parameters of a trained neural network.

However, when the learning apparatus that performs learning and the apparatus that utilizes the capability obtained as a result of the learning are respectively different, it is difficult for a user of the learning result to grasp what data was used and what learning was performed to obtain the learning result. In the future, when a diverse variety of machine learning will be performed at various locations following the spread of AI technologies, it is possible that large numbers of learning results output by learning apparatuses will be automatically distributed without human intervention, although this is predicted to make it more and more difficult to find a new learning result that has procured a capability that is required by a user from among those learning results.

In view of this, an object of the present invention is to provide a technology for detecting that a new capability has been procured as a result of having performed learning by machine learning, and notifying a user who requires that capability.

A notification apparatus according to one aspect of the present invention is provided with an acceptance unit that accepts a condition designated by a requester, an acquisition unit that acquires a learning result obtained due to predetermined learning being performed by machine learning, a determination unit that determines whether a procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit, and a notification unit that notifies the requester if necessary, based on a result of the determination. According to this aspect, it becomes possible to detect that a new capability has been procured as a result of having performed learning by machine learning, and notify a user who requires that capability. Also, since the requester is notified in the case where a predetermined condition is satisfied, among learning results that are output as a result of learning performed in a learning apparatus, the volume of data that is communicated can be reduced.

The determination unit may also be configured to determine whether an application range of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit. The determination unit may also be configured to determine whether a learning object of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit. According to these aspects, in the case where the acquired learning result is outside the range designated by the requester, it is understood that notification is not required since further determination processing will not be performed, thus allowing the processing speed of determination processing to be improved, and the load on processing apparatuses such as the CPU to be reduced since the processing process is shortened.

The determination unit may also be configured to perform the determination based on whether the procured capability of the learning result acquired by the acquisition unit is equivalent from a viewpoint of the condition, through comparison with a capability procured by a predetermined learning result. The determination unit may also be configured to determine whether the procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit. According to these aspects, the volume of data that requires comparison determination can be suppressed, thus allowing the processing speed of the determination processing to be improved, and the load on processing apparatuses such as the CPU to be reduced since the processing process is shortened.

The notification apparatus may be connected to a database that stores a plurality of learning results. At this time, the determination unit may further determine whether the procured capability of the learning result acquired by the acquisition unit satisfies the condition designated by the requester, through comparison with a capability procured by each of the plurality of learning results stored in the database. According to this aspect, the requester is notified in the case where a predetermined condition is satisfied on comparison with learning results from the past, thus enabling the volume of data that is communicated to be reduced.

An inspection device according to one aspect of the present invention is able to apply a learning result that is a result of predetermined learning performed in order to procure a predetermined capability. Also, the inspection device is provided with an acceptance unit that accepts a condition designated by a requester, an acquisition unit that acquires a learning result obtained due to predetermined learning being performed by machine learning, a determination unit that determines whether a procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit, and a notification unit that notifies the requester if necessary, based on a result of the determination. According to this aspect, it becomes possible for an inspection device itself to detect that a new capability applicable to the inspection device has been procured as a result of having performed learning by machine learning, and notify a user who requires the capability. Also, the requester is notified in the case where a predetermined condition is satisfied, among the learning results that are output as a result of learning performed by the learning apparatus, thus enabling the volume of data that is communicated to be reduced.

A notification method according to one aspect of the present invention executes the steps of accepting a condition designated by a requester, acquiring a learning result obtained due to predetermined learning being performed by machine learning, determining whether a capability procured by the acquired learning result satisfies the accepted condition, and notifying the requester if necessary, based on a result of the determination.

A computer program according to one aspect of the present invention causes a computer to function as a unit that accepts a condition designated by a requester, a unit that acquires a learning result obtained due to predetermined learning being performed by machine learning, a unit that determines whether a capability procured by the acquired learning result satisfies the accepted condition, and a unit that notifies the requester if necessary, based on a result of the determination.

According to the present invention, a technology for detecting that a new capability has been procured as a result of having performed learning by machine learning and notifying a user who requires that capability can be provided.

Fig. 1 is a diagram schematically showing the overall configuration of a learning system 100 in an embodiment. Fig. 2 is a diagram conceptually showing the flow of processing by the learning system 100. Fig. 3 is a functional block diagram of a learning data preparation apparatus 21 in the embodiment. Fig. 4 is a functional block diagram of a learning request apparatus 22 in the embodiment. Fig. 5 is a functional block diagram of a learning result utilization apparatus 23 in the embodiment. Fig. 6 is a block diagram showing the configuration of a learning database 11 in the embodiment. Fig. 7 is a functional block diagram of a learning request acceptance apparatus 12 in the embodiment. Fig. 8 is a functional block diagram of a learning apparatus 13 in the embodiment. Fig. 9 is a functional block diagram of a capability determination apparatus 15 in the embodiment. Fig. 10 is a block diagram showing an example of a hardware configuration of each apparatus constituting the learning system 100. Fig. 11 is a flowchart of capability determination processing in the capability determination apparatus 15. Fig. 12 is a block diagram showing an example of the configuration of an inspection device.

Hereinafter, embodiments of the present invention will be described in detail, with reference to the drawings. Note that the same reference signs are given to elements that are the same and overlapping description will be omitted. Also, the following embodiments are illustrative embodiments for describing the present invention, and it is not intended to limit the present invention to only these embodiments. Furthermore, various modifications can be made to the present invention, without departing from the gist of the invention.

1. System Outline
An outline of the system in the present embodiment will be described with reference to Figs. 1 and 2.

Fig. 1 is a diagram schematically showing the overall configuration of a learning system 100 in the present embodiment. This learning system 100 is constituted to include a learning service provision system 1 and a learning request system 2, which are connected to each other via a network N0 such as the Internet. Note that the configuration of this system is not limited to the illustrated configuration, and may, for example, be constituted as a system in which the learning service provision system 1 and the learning request system 2 are physically or logically integrated.

As shown in Fig. 1, the learning service provision system 1 has a learning database 11, a learning request acceptance apparatus 12, one or a plurality of learning apparatuses 13 and a capability determination apparatus 15, which are connected to each other via a local network N1. Also, the learning request system 2 has a learning data preparation apparatus 21, a learning request apparatus 22 and one or a plurality of learning result utilization apparatuses 23, which are connected to each other via a local network N2. Also, these apparatuses are configured to accept inputs from a learning requester or the like, via an input apparatus (not shown).

Fig. 2 is a diagram conceptually showing the flow of processing by the learning system 100 shown in Fig. 1 from a learning request until utilization of a learning result. First, a requester requests the learning request apparatus 22 to perform learning, via the input apparatus, by designating an objective of learning (e.g., identifying a specific object, etc.). The learning request apparatus 22, upon acceptance of a learning request from the requester, accesses the learning service provision system 1 via the network N0, and transmits learning request information including information required in order to perform machine learning to the learning service provision system 1 (S1). The learning request information includes, for example, the objective of learning, information identifying an apparatus that will utilize the learning result, and learning data.

The learning request acceptance apparatus 12 of the learning service provision system 1, upon accepting learning request information, instructs the learning apparatus 13 to execute machine learning that is based on the learning request information (S2). The learning apparatus 13 executes machine learning, based on the learning request information. For example, the learning result of a trained neural network or the like that has procured a predetermined capability is thereby obtained. The learning result is converted into reproducible packaged data (S3). The digitized learning result is transmitted from the learning apparatus 13 to the learning result utilization apparatus 23 of the learning request system 2 (S4). In the learning result utilization apparatus 23, the predetermined capability is exhibited due to the learning result being utilized.

The learning result obtained with the learning apparatus 13 is, furthermore, stored in the learning database 11. An arbitrary user of the learning system 100 thereby becomes able to make use of past learning results. In the present embodiment, when a new learning result is obtained with the learning apparatus 13, the newly obtained learning result is compared with past learning results stored in the learning database 11, and the capability determination apparatus 15 detects whether a new capability has been procured through comparison of the learning result obtained this time with the past learning results. In the case where it is determined that a new capability has been procured, based on a result of the detection, the capability determination apparatus 15 notifies the user who requires that capability. The user is thereby able to quickly grasp that a new capability has been procured as a result of machine learning.

2. Apparatus Configurations
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. Note that the configurations of the apparatuses that are included in the learning request system 2 are not limited to the configurations described below, and may be modified such that the apparatuses are provided with arbitrary functions that other apparatuses are provided with if necessary.

Fig. 3 is a functional block diagram of the learning data preparation apparatus 21 in the present embodiment. The learning data preparation apparatus 21 has a function of preparing data (learning data) required in order to train a learning module (e.g., neural network 233 discussed later). As shown in Fig. 3, the learning data preparation apparatus 21 has, as functional units, an operation unit 211, a learning data acquisition unit 212, a learning data storage unit 213 and a data acquisition control unit 214.

For example, the operation unit 211 accepts operations from a user (hereinafter, also "requester") of the learning request system 2. The learning data acquisition unit 212 acquires data required in order to create learning data from an arbitrary input apparatus such as a camera, a sensor, a network terminal or a sensor of an automated robot, and stores the acquired 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 required in learning. A communication unit 216 connects to the local network N2 of the learning request system 2, and transmits data required in learning created by the data acquisition control unit 214 to another apparatus.

Note that the learning data preparation apparatus 21 may be configured to be built using the same apparatus as the learning result utilization apparatus 23 discussed later. In this case, the learning data preparation apparatus 21 can also be built as an input apparatus that is externally connected to the learning result utilization apparatus 23.

Fig. 4 is a functional block diagram of the learning request apparatus 22 in the present embodiment. The learning request apparatus 22 has a function of transmitting learning request information to the learning service provision system 1. As shown in Fig. 4, the learning request acceptance apparatus 12 has a learning request unit 221, a learning request contents storage unit 222, a learning data storage unit 223 and a communication unit 224.

The learning request unit 221 accepts a learning request from a requester via the learning data preparation apparatus 21 or an input apparatus (not shown), creates learning request information, and transmits the learning request information to the learning service provision system 1 via the communication unit 224. The learning request information that is transmitted at this time is stored in the learning request contents storage unit 222. Also, learning data required in order to perform the learning requested by the requester is acquired from the learning data preparation apparatus 21, and transmitted to the learning service provision system 1. The learning data that is transmitted to the learning service provision system 1 is also stored in the learning data storage unit 223.

Also, the learning request unit 221 accepts a notification request designated by the requester from the input apparatus, creates notification request information, and transmits the notification request information to the learning service provision system 1 via the communication unit 224. The notification request information includes, for example, at least one of a capability that is required by the requester, an application range in which the requester will utilize the learning result, a learning object, and an output requirement. The notification request information that is transmitted at this time is stored in the learning request contents storage unit 222. For example, when the requester requests notification of a capability for classifying grades of agricultural products, information indicating that the condition of the capability required by the requester is grade classification of agricultural products is included as notification request information. Also, the learning object or the application range may be further narrowed down, and classification of grades of tomatoes may be included as notification request information. As the capability that is required, classification of the grades into five levels may be included as notification request information, and the fact that the classification accuracy was improved by a predetermined number of points or more compared with the existing capability may be notified as an output requirement.

Furthermore, the learning request apparatus 22 may, in the case where identification information is assigned to the learning result, refer to the identification information that is assigned to the learning result, and determine whether the input learning result is compatible with the utilization purpose. In the case where the learning result is compatible with the utilization purpose, the learning request apparatus 22 causes the learning result to be utilized by embedding the learning result in the learning result utilization apparatus 23. On the other hand, in the case where the learning result is not compatible with the utilization purpose, the learning request apparatus 22 is able to notify that the learning result is incompatible to the learning service provision system 1. At this time, the learning request apparatus 22 may change the conditions and request learning again.

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

The learning result input unit 231 accepts input of a learning result. At this time, the neural network setting unit 232 configures the settings of the neural network 233 according to the utilization purpose. Furthermore, the control unit 234 controls the data acquisition unit 237 and the input unit 235 to input data required in utilization of the learning result to the neural network 233, and carries out utilization of the learning result. Note that the result of having utilized the learning result is output from the output unit 238.

2-2. Learning Service Provision System 1
The function of the learning service provision system 1 will be described, with reference to Figs. 6 to 10. Note that the learning service provision system 1 can be implemented using a data center or a cloud. In this case, each apparatus of the learning service provision system 1 can be built using a PC server or a blade PC. By building the apparatuses of the learning service provision system 1 with a plurality of PCs, processing time can be shortened in the case of performing repetitive operations such as deep learning. Note that the learning service provision system 1 may have a configuration built with one PC or have a configuration implemented by embedded apparatuses.

Fig. 6 is a block diagram showing the configuration of the learning database 11 in the present embodiment. The learning database 11 stores various types of information that are required when learning is performed. As shown in Fig. 6, the learning database 11 has a learning data DB 111, a learning request DB 112, a learning result utilization history DB 113, a learning result DB 114, a learning program DB 115 and a learning object DB 116.

The learning data DB 111 stores learning data that is used in learning. For example, the learning data DB 111 is able to store the requirements for learning such as the object of learning data, the breakdown of learning data, the range of learning data and the purpose of learning in association with learning data.

The learning request DB 112 stores learning request information and the contents of learning that is carried out in the case where a learning request information is accepted from the learning request apparatus 22. For example, the learning request DB 112 is able to store information relating to the learning requester and requirements for a learning request such as the object of learning data, the breakdown of learning data, the range of learning data and the purpose of learning in association with learning request information.

The learning result utilization history DB 113 stores the utilization history of learning results. For example, the learning result utilization history DB 113, in the case where a classification capability is procured as a learning result, is able to store information relating to the result of having performed classification utilizing the procured classification capability. Furthermore, the learning result utilization history DB 113 is able to store information related to utilization of a learning result in association with utilization of learning such as information relating to users of a learning result, the object of learning data, the breakdown of learning data, the range of learning data and the purpose of learning. Furthermore, the learning result utilization history DB 113 preferably includes utilization identification information. Utilization identification information is information that is able to identify utilization of a learning result, and includes the ID of an apparatus that is utilized and information relating to settings (basic factors, effect factors, etc. discussed below) that affect capability, for example. Since the procured capability may change according to factors such as the apparatus that is utilized and the utilization environment, a more detailed utilization history can be recorded, as a result of the learning utilization history DB 113 including utilization identification information.

The learning result DB 114 is for storing learning results output by the learning apparatus 13, and a plurality of past learning results are stored. The learning results stored in the learning result DB 114 includes digitized learning results. Also, when identification information is assigned to a learning result, the identification information of the learning result is included in the information stored in the learning result DB 114.

The learning program DB 115 stores learning programs for performing learning. For example, the learning program DB 115 is able to store the learning programs in association with the requirements for learning such as the object of learning, the contents of learning data and the objective of learning. Note that it is preferable to be able to register a large number of learning programs in the learning program DB 115. In this case, the learning apparatus 13, which will be discussed later, is able to specify a learning program from the learning program DB 115 and make the learning program executable, by designating the requirements for learning.

The learning object DB 116 stores information relating to the object of learning. Examples of learning objects include an object identification apparatus that identifies the type of object, an image processing apparatus, a product management apparatus, a robot, a sensor signal prediction apparatus, and an in-vehicle apparatus.

Fig. 7 is a functional block diagram of the learning request acceptance apparatus 12 in the present embodiment. The learning request acceptance apparatus 12 has a function of accepting learning request information from the learning request apparatus 22, and transmitting the learning request information to the learning apparatus 13. As shown in Fig. 7, the learning request acceptance apparatus 12 is, for example, constituted to include a learning request acceptance unit 121, a learning data storage unit 123, a learning request contents storage unit 124 and a communication unit 125.

The learning request acceptance unit 121 accepts a learning request from the learning request apparatus 22. The learning request acceptance unit 121 registers learning request information that is included in the accepted learning request in the learning request DB 112. At this time, the learning request acceptance unit 121 transmits a notification indicating that a learning request has been accepted to the learning management apparatus 14, via the communication unit 125. Also, the learning request acceptance apparatus 12 is able to temporarily save the accepted learning request in the learning data storage unit 123 or the learning request contents storage unit 124.

Fig. 8 is a functional block diagram of the learning apparatus 13 in the present embodiment. The learning apparatus 13 has a function of performing learning based on learning request information, and acquiring a predetermined capability as a learning result. As shown in Fig. 8, the learning apparatus 13 has, as functional units, 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.

In the learning apparatus 13, the learning control unit 131 is able to control the neural network 132 to perform learning based on learning request information. A learning result is extracted by the learning result extraction unit 133, and output by the learning result output unit 135, via the communication unit 134.
In the learning apparatus 13, a plurality of learning results are obtained when learning is performed a plurality of times. When requirements that affect the learning result, such as the learning data, the learning program, the learning time and the objective of learning, are dissimilar each time learning is performed, there is a possibility that the capabilities obtained by the plurality of learning results will not be the same. In the case of assigning identification information to learning results, different identification information may be assigned and managed as a variation in the case where there is a possibility that the capabilities obtained by a plurality of learning results will not be the same.

Note that it is also possible for the learning apparatus 13 itself to make a learning request. In this case, it will be possible for the learning apparatus 13 to perform autonomous learning.

Fig. 9 is a functional block diagram of the capability determination apparatus 15 in the present embodiment. The capability determination apparatus 15 has a function of determining whether the procured capability of the learning result output as a result of learning performed in the learning apparatus 13 satisfies a predetermined condition. As shown in Fig. 9, the capability determination apparatus 15 has a capability determination control unit 151, a comparison object selection unit 152, a learning result comparison unit 153, an identification information generation unit 154 and a communication unit 155.

The capability determination control unit 151 controls the units included in the capability determination apparatus 15 to perform capability determination processing which will be discussed later. The capability determination control unit 151 is further provided with an acceptance unit 1511, an acquisition unit 1512, a determination unit 1513 and a notification unit 1514. The acceptance unit 1511 accepts notification request information that includes conditions designated by the requester from the learning request apparatus 22. The acquisition unit 1512 acquires a learning result obtained due to predetermined learning being performed from the learning apparatus 13. The determination unit 1513 determines whether the procured capability of the learning result acquired by the acquisition unit 1512 is a new capability that satisfies the conditions included in the notification request information, using the comparison object selection unit 152 or the learning result comparison unit 153. The notification unit 1514 notifies the requester via the communication unit 155, when it is determined that a new capability was procured based on the conditions included in the notification request information. The comparison object selection unit 152 selects learning results to compare with the learning result obtained this time, from the plurality of learning results stored in the learning database 11. When a condition relating to the application range and a condition relating to the learning object are included in the notification request information, selection processing is performed based on these conditions.

The learning result comparison unit 153 compares the learning result acquired from the learning apparatus 13 with one or a plurality of learning results selected by the comparison object selection unit 152, and determines whether a new capability was procured by the learning result obtained this time. When a condition relating to the output requirements is included in the notification request information, the determination processing is performed based on this condition.

The identification information generation unit 154 assigns identification information to a learning result that has procured a predetermined capability due to learning having been performed. Here, assigning identification information to learning results is a concept that includes associating learning results with identification information, and includes, for example, assigning identification information to learning results and storing learning results in a storage device in association with identification information. Also, in addition to directly associating learning results with identification information, learning results may be indirectly associated with identification information.

2-3. Hardware Configuration of Apparatuses
Fig. 10 is a block diagram showing an example of the hardware configuration of the apparatuses constituting the learning system 100. As the apparatuses such as the capability determination apparatus 15 and the like, a general purpose or dedicated computer that is provided with a CPU 1010, memories such as a ROM 1020 and a RAM 1030, a storage device 1040 that store various types of information, an input-output unit 1050, a communication unit 1060 and a network or a bus that connects these constituent elements can be applied, as shown in the diagram.

The functions that are implemented in the apparatuses are not limited to a configuration that is realized by a CPU executing a predetermined program stored in a memory or a storage device. Suitable functions that are included in the apparatuses may have a configuration realized by hardware. For example, the neural networks 132 and 233 which will be discussed later may be constituted by an electronic circuit such as a custom LSI (Large-Scale Integration) or FPGA (Field-Programmable Gate Array). Furthermore, some of the apparatuses may have a configuration realized by an identification means that uses a biochemical method such as a physical key or DNA or an optical method such as a hologram. Note that the programs can be installed or loaded in a computer through various types of recording media such as an optical disk like a CD-ROM, a magnetic disk or a semiconductor memory or by being downloaded via a communication network.

3. Processing Flow
Next, the processing flow of the system according to the present embodiment will be described with reference to Fig. 11.

Fig. 11 is a flowchart of capability determination processing in the capability determination apparatus 15. First, the acceptance unit 1511 of the capability determination control unit 151 accepts the notification request information of a requester from the learning request apparatus 22 (S1110). The accepted notification request information is stored in a storage device in association information such as the notification destination of the requester.

Thereafter, when the learning apparatus 13 outputs a new learning result based on a learning request from the same requester or a different requester, the acquisition unit 1512 acquires the learning result output from the learning apparatus 13 (S1120). The determination unit 1513 then determines whether the capability procured by the acquired learning result satisfies conditions that are included in the notification request information, that is, whether a new capability that is required by the requester has been procured, using the comparison object selection unit 152 or the learning result comparison unit 153 (S1130).

Specifically, the determination unit 1513 executes determination processing that is constituted by steps S1131 to S1134 which will be discussed below. Note that the respective processing from step S1131 to S1134 need not all necessarily be executed, and required processing need only be selectively executed, according to the conditions that are included in the notification request information. The order of processing is also not limited to the order that will be described below, and suitable changes can be made as long as inconsistencies do not arise in the processing contents.

First, with respect to the learning result acquired at step S1120 or the plurality of learning results stored in the learning database 11, the comparison object selection unit 152 performs processing for limiting the application range (S1131). The capability determination apparatus 15 need only compare and determine the capabilities in an available range (hereinafter, a "range") for each requester. For example, even with capabilities of the same type, the available range differs if the input or output configurations (number of inputs and outputs, type of data that is output and input, etc.) are dissimilar. Alternatively, it is not necessary to determine whether a learning result that cannot be utilized by a requester is a new capability. The requester is able to designate the application range with suitable conditions when making a notification request. For example, the application range may be set for every group of learning programs that are able to deal with equivalent learning objects. Also, the application range may be set for multiple instances of learning using data from different periods for the same apparatus of the same requester. In the case where a learning result satisfies the application range, the specification processing of step S1132 is executed. On the other hand, in the case where a learning result does not satisfy the application range, the determination processing is ended.

Application range specification conditions are used in specification of the application range. The application range specification conditions are conditions for specifying whether a learning result is included in the application range, and can, for example, be set for an objective of the capability that is procured. The application range specification conditions are stored in an application range specification condition storage unit. As an example of application range specification conditions, identification information of the application range can be used. For example, identification information capable of identifying the input-output configuration or the objective of the capability that is procured may be assigned to the learning results, and learning results to which the same identification information is assigned may be specified as having equivalent application ranges. As another example of application range specification conditions, a range equivalency judgment logic for judging the equivalency of an application range may be stored, and this logic may be used to specify whether a learning result is included in the application range. Note that specific examples of the application range specification conditions are not limited thereto, and other suitable conditions can be used.

Next, the comparison object selection unit 152 performs processing for specifying the learning object on the learning result selected by the application range limiting processing (S1132). Even when the learning result is included in the application range, if the object of learning differs, the capability that is procured will not be a new capability that is sought by the requester, even when the learning result is dissimilar. Accordingly, it needs to be determined and specified whether the learning object is a learning object with respect to which it is desired to detect procurement of a new capability. If the learning result satisfies the conditions of the learning object, the determination processing of step S1133 is executed. On the other hand, if the learning result does not satisfy the conditions of the learning object, the determination processing is ended.

Comparison of learning results can be performed for every object of learning. For example, because a learning apparatus that performs classification of images and an apparatus that performs prediction of sensor signals have different objects, the learning results differ. Although it is possible to determine for both of these apparatuses whether a new capability was procured, usefulness is limited from the viewpoint of utilization of the system. An improvement in processing speed can be achieved, by removing learning results obtained for different objects from being objects for comparison, and comparing learning results obtained for the same object.

The learning object specification conditions are used in specification of a learning object. The learning object specification conditions can be set for a predetermined learning set. The learning object specification conditions are stored in a learning object specification condition storage unit. As an example of learning object specification conditions, the equivalency of learning objects may be specified, by learning objects being assigned identification information and designated in advance. Also, the equivalency of learning objects may be specified, using a logic for determining the equivalency of learning objects.

By specifying the learning object, learning results obtained from learning performed on the same learning object can be compared. In this case, a plurality of learning results whose learning objects are the same are compared, enabling the degree to which the respective learning results differ to be compared. For example, by providing equivalent inputs for two learning results whose learning objects are the same and looking at how the outputs are dissimilar, it can be determined whether both learning results have comparable capabilities or whether they have significantly different capabilities. Also, a plurality of learning results are obtained when learning is performed a plurality of times on the same learning object. Thereafter, when learning is newly performed on the same learning object, there are cases where the capability is comparable to one or more pervious learning results, and cases where a new capability is obtained that is significantly different and not comparable to any of the learning results.

Note that at the time of determining the equivalency of learning objects, learning results may be specified assuming that similar learning objects are equivalent, in addition to learning objects that are the same. It is thereby possible to deal with times when there are few learning results having the same learning object or cases where the range of learning objects can be expanded.

Here, an example of the application range and the learning object will be specifically described. In the present embodiment, the application range and the learning object can be respectively set for every learning program. The learning object includes objects and things targeted by the capability that is procured as a learning result. For example, if the learning result is a capability for classifying grades of agricultural products, the learning object is agricultural products. At this time, a capability for classifying grades of agricultural products which are the learning object can differ, depending on the learning program or the learning data that is used in learning. For example, learning results only for cucumbers or only for tomatoes and learning results applicable to multiple types of fruit and vegetables can be output, in which case, the application range will respectively be cucumbers, tomatoes, and fruit and vegetables. That is, the application range includes the range in which learning result can be applied. Note that the application range may be hierarchically defined.

When the definitions of the application range and the learning object are commonly set for the learning programs as a whole, it favorably becomes easier to find new capabilities, although the definitions need not necessary be commonly set for the learning programs as a whole. As long as the users of learning results are specified or limited, such as to specific customer-oriented learning, the definitions of the application range and the learning object may be set individually within the range. For example, with a learning program A, the learning object is set to agricultural products and the application range is set to fruit and vegetables, whereas with a learning program B, learning results in the case where the learning object is fruit and vegetables and the application range is tomatoes and learning results in the case where the learning object is fruit and vegetables and the application range is cucumbers may be output, according to the data used in learning. Also, a configuration may be adopted in which the learning object is set to tomatoes and the application range is also set to tomatoes.

As another example, if the learning result is the capability for creating a summary of news articles, the learning object is news articles. In this case, for example, learning results for only domestic news, learning results for only economic news, and learning results whose application range is news in general and that is able to deal with any of these fields are envisaged. Also, in the case where the learning result is control of a specific customer-oriented assembly apparatus, the following examples can be envisaged. For example, the learning object is control of the assembly apparatus for a specific product, and the production lines are 1 to 5. At this time, when a learning result 1 corresponds to only the production line 1, the application range of the learning result 1 is the production line 1. Also, when a learning result 2 corresponds to the production lines 2 to 5, the application range of the learning result 2 is the production lines 2 to 5.

Next, the learning result comparison unit 153 determines the equivalency of the learning results (S1133). That is, learning results selected by the selected comparison object selection unit 152 are compared and it is determined whether the learning results are equivalent. In the case where it is determined that the learning results are equivalent, output requirement determination processing of step S1134 is executed. On the other hand, the determination processing is ended in the case where it is determined that the learning results are not equivalent.

Generally, when machine learning is performed, the learning results are a little different, if the conditions of learning are even slightly different. Accordingly, when new machine learning has been performed, it may be determined whether the new machine learning is equivalent to learning performed previously, and when it is determined that the new machine learning is not equivalent learning, the result of machine learning this time may be regarded as having obtained a new capability. Also, in the case where the learning program uses random numbers therein, dissimilarities may occur in the respective learning results, when learning is performed a plurality of times under exactly the same conditions apart from the random numbers. At this time, each of capabilities that are slightly dissimilar can also be determined to be a new capability. However, in this case, the majority of these capabilities will be almost the same when checked upon receiving notification of a new capability, thus possibly resulting in the value of information indicating a new capability being greatly diminished.

In order to avoid such problems, processing for comparing equivalencies, that is, for comparing whether a plurality of learning results are equivalent, is performed. For example, equivalencies are compared, by contrasting the learning result obtained by learning this time with each of past learning results narrowed down based on the application range and the learning object.

Here, the determination of the equivalency of learning results is performed using learning result equivalency determination conditions. The learning result equivalency determination conditions are stored in a learning result equivalency determination condition storage unit. Hereinafter, the learning result equivalency determination conditions will be described.

First, the conditions or standards (hereinafter, simply "comparison conditions") used at the time of comparing equivalencies are different for every type of capability. The equivalency comparison conditions are registered in advance in a storage device, for every type of capability. Specifically, the comparison conditions are set for every learning program or for every group of similar learning programs, for example. A configuration may also be adopted in which settings are configured according to the learning requester.

In the case where a requirement that the requester is expecting as a result of learning is included as a notification condition, it is determined whether this requirement is satisfied. In the case where the degree of capability is included as a notification condition, it is determined, for example, whether the degree of improvement with regard to an item selected based on a predetermined condition is greater than or equal to a predetermined value.

In the case where the types of learning data used in learning are dissimilar when comparing a plurality of learning results, the degree to which the respective learning results are dissimilar can be determined, by using internal information of the learning program. In the case where the data used in learning is dissimilar even though the type of learning data is the same, the degree to which the respective learning results are dissimilar can be determined, by comparing differences in capabilities that depend on the dissimilarities in the learning data. In the case where the output of learning results that use the same learning data is dissimilar, the degree to which the respective learning results are dissimilar can be determined, by setting the comparison conditions according to the dissimilarities, for every content that is dissimilar. As standards that are used when comparing learning results, capabilities such as classification capabilities, predictive abilities, control capabilities and linguistic capabilities obtained as a result of learning can be compared, for example.

Next, when, as a result of the equivalency determination, the learning result acquired this time is determined to not be equivalent upon comparison with past learning results, the learning result comparison unit 153 performs processing for determining of the output requirements (S1134). That is, processing for determining whether to notify the requester is performed, based on predetermined requirements that differ from the conditions for determining the equivalency of the learning results, such that capabilities that the requester is expecting are notified. When learning results determined by the equivalency determining conditions to not be equivalent are all notified externally, there is a risk that capabilities that the requester does not expect will be notified, and when a large number of unnecessary notifications are output, there is the possibly of imposing an unnecessary burden on the requester, which is not desirable.

In the processing for determining the output requirements, whether a learning result satisfies the output requirements is determined using output requirements designated by the requester beforehand. The output requirements included in the notification request information are stored in an output requirement determination condition storage unit. Output requirements that can be set include, for example, restriction requirements relating to the learning result utilization apparatus such as the size of neural networks and the response time when a learning result is embedded in a learning result utilization apparatus, restriction requirements relating to inputs such as camera resolution, sensor response period and sensor accuracy, restriction requirements relating to output such as the number of classification outputs and the output period, and object-related restriction requirements.

When it is thus determined, as a result of the determination processing by the determination unit 1513, that a learning result has procured a new capability that is required by the requester, the requester is notified by the notification unit 1514 (S1140).

The contents of the notification include information for specifying the learning result. For example, a configuration can be adopted in which information indicating the contents of learning, such as information relating to the learning object, information relating to the application range, information relating to the equivalency determination and information relating to the output requirements, is included in the notification contents. Also, a configuration may be adopted in which identification information indicating a predetermined learning result is included in the notification contents. By establishing the identification information to be notified when a learning result that satisfies a predetermined condition is obtained in advance in cooperation with the requester, it becomes possible to quickly inform the requester that a new capability that he or she is expecting has been procured, with minimal information.

The object of notification can include, apart from the requester who requested notification, a person concerned with the learning object, the developer of the learning program, a learning requester, or the like. The object of notification is stored in a notification object storage unit. In addition to the requester of the notification request, a person selected with a notification object selection unit is stored as a notification object in the notification object storage unit.

4. Utilization Examples
Utilization examples of the capability determination processing of the learning system 100 will be described.

In an object identification apparatus for identifying the type of object, in cases such as, for example, when the misdetermination rate improves by 20% or more, when it becomes possible to deal with a new crop, when identification becomes possible using a three-dimensional shape, when reliability is enhanced due to the amount of learning data greatly increasing, or when it becomes possible to identify a new type of animal, the requester is notified of that a new capability has been procured by the learning result acquired this time.

In an image processing apparatus, in cases such as, for example, when it becomes possible to deal with larger images, when the number of gradations that can be handled increases, when the number of types that can be classified increases, or when the number of objects that can be processed in real-time increases, the requester is notified.

In a product management apparatus such as an inspection apparatus, in cases such as, for example, when the concordance rate with human sensory inspection improves, when it becomes possible to deal with a new input, or when it becomes possible to perform inspection by combining three images from inspection of a single image, the requester is notified.

In a robot, in cases such as, for example, when a capability for performing a new operation is procured, when a capability is improved, when operational speed improves, when operational accuracy improves, or when the evaluation result of an operation improves, the requester is notified.

In a prediction apparatus, in cases such as, for example, when it becomes possible to predict the next data or state based on input data up to this point, the requester is notified.

In an in-vehicle apparatus, in cases such as, for example, when the types that can be identified by an environmental sensor increase, when the misdetermination rate of identification of tunnels and bridges or the like improves, or when the non-detection rate relating to events whose occurrence is cause for concern improves, the requester is notified.

In a game machine player, in cases such as, for example, when an experience value or skill improves or when it becomes possible to deal with a new game, the requester is notified.

5. Working Example of Inspection Device
An example in the case where the notification apparatus according to the present embodiment is applied to an inspection device will be described. For example, a manufacturer Z of electronic circuit boards sorts non-defective product and defective products using an inspection device. In this example, a case will be described in which the manufacturer Z is a notification requester, and learning results that procure a capability for the inspection device to sort non-defective products and defective products are acquired, the manufacturer Z is notified if it is determined that a learning result has procured a new capability that satisfies conditions included in a notification request, and the learning result is applied in the inspection device.

First, the configuration of an inspection device a will be described using Fig. 12. Fig. 12 is a block diagram showing an example of the configuration of the inspection device a. As shown in Fig. 12, the inspection device a is provided with a storage unit A1, an input unit A2, a discrimination unit A3 and an output unit A4. The storage unit A1 stores learning results of learning performed by the learning service provision system 1, for example. The input unit A2 has a function of acquiring information of an inspection object, and is constituted by sensors such as a camera and a microphone, for example. The discrimination unit A3 has a function of discriminating a predetermined inspection object such as non-defective products or defective products, based on information (e.g., images, etc.) of the inspection object input from the input unit A2. Also, the output unit A4 has a function of outputting the determination result of the discrimination unit A3. Note that outputting the determination result here includes not only presenting the determination result to the user by displaying the determination result on a display apparatus or the like, but also operations of the inspection device, such as sorting out a predetermined inspection object from other inspection objects. The inspection device a is further provided with an acceptance means, an acquisition means, a determination means, and a notification means, other than the configuration serving as the above inspection device. The acceptance means, the acquisition means, the determination means and the notification means have equivalent functions to the acceptance unit 1511, the acquisition unit 1512, the determination unit 1513 and the notification unit 1514 in the above embodiment.

Next, processing for determining and notifying that a new capability has been procured in the inspection device a will be described. First, when the manufacturer Z requests the inspection device a to perform notification in the case where a learning result that satisfies a predetermined condition is acquired, the acceptance means of the inspection device a accepts notification request information, and store the notification request information in a storage unit. For example, the manufacturer Z, when determining non-defective products or defective products based on a plurality of images of the electronic circuit boards captured by a camera provided in the inspection device a, issues a notification request whose condition is that the accuracy rate improves by 10% or more over the learning model that is implemented in the inspection device a.

The acquisition means of the inspection device a acquires a learning result via a network regularly or irregularly. For example, the acquisition means accesses the learning service provision system 1 and acquires a learning result stored since the previous access and identification information of the learning result from the learning database 11.

The determination means of the inspection device a determines whether the acquired learning result satisfies conditions included in the notification request information, and, in the case where the conditions are satisfied, notifies the manufacturer Z using the notification means. The inspection device a is able to replace the learning result currently being applied in the learning model that is applied in the inspection device a with the learning result newly acquired this time, in response to an instruction from the manufacturer Z or the like. In this way, by applying the notification apparatus according to the present embodiment to the inspection device, the inspection device a becomes able to autonomously update the learning model that is implemented, and the reliability of the inspection device further improves.

One embodiment of the present invention has been described above. Note that the present embodiment is for facilitating understanding of the present invention, and is not for interpreting the present invention in a limited manner. Also, changes or improvements can be made to the present invention without departing from the gist of the invention. Also, the steps in the abovementioned processing flow can be partially omitted in a range that does not produce inconsistencies in the processing contents, and the processing steps can be suitably rearranged or performed in parallel.

In the abovementioned embodiment, the learning request and the notification request were described as separate requests, but the learning request may be configured to include the notification request. That is, based on request contents designated by the requester, the learning request apparatus 22 creates learning request information and notification request information, and transmits this information to the learning service provision system 1. Subsequent processing is performed as described above, and after receiving a learning result that is based on the learning request, the requester is further able to receive notification when a new capability is procured with regard to the capability that is required by the requester.

Note that, in this specification, a "unit" or "means" does not merely signify a physical configuration, and also includes the case where the functions of the "unit" are realized by software. Also, the functions of one "unit", "means" or apparatus may be realized by two or more physical configurations or apparatuses, or the functions of two or more "units", "means" or apparatuses may be realized by one physical means or apparatuses.

The above embodiment can also be partially or wholly described as in the following additional remarks, but is not limited to the following.
Additional Remark 1
A notification apparatus including at least one hardware processor,
the hardware processor:
accepting a condition designated by a requester,
acquiring a learning result obtained due to predetermined learning being performed by machine learning,
determining whether a capability procured by the acquired learning result satisfies the accepted condition, and
notifying the requester if necessary, based on a result of the determination.
Additional Remark 2
A notification method for executing, with at least one or more hardware processors, the steps of:
accepting a condition designated by a requester;
acquiring a learning result obtained due to predetermined learning being performed by machine learning;
determining whether a capability procured by the acquired learning result satisfies the accepted condition; and
notifying the requester if necessary, based on a result of the determination.

Claims (9)

1. A notification apparatus comprising:
   an acceptance unit that accepts a condition designated by a requester;
   an acquisition unit that acquires a learning result obtained due to predetermined learning being performed by machine learning;
   a determination unit that determines whether a procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit; and
   a notification unit that notifies the requester if necessary, based on a result of the determination.
2. The notification apparatus according to claim 1,
   wherein the determination unit determines whether an application range of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit.
3. 3The notification apparatus according to claim 1 or 2,
   wherein the determination unit determines whether a learning object of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit.
4. The notification apparatus according to any of claims 1 to 3,
   wherein the determination unit performs the determination based on whether the procured capability of the learning result acquired by the acquisition unit is equivalent from a viewpoint of the condition, through comparison with a capability procured by a predetermined learning result.
5. The notification apparatus according to any of claims 1 to 4,
   wherein the determination unit determines whether the procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit.
6. The notification apparatus according to any of claims 1 to 5,
   wherein the notification apparatus is connected to a database that stores a plurality of learning results, and
   the determination unit further determines whether the procured capability of the learning result acquired by the acquisition unit satisfies the condition, through comparison with a capability procured by each of the plurality of learning results stored in the database.
7. An inspection device in which a learning result that is a result of predetermined learning performed in order to procure a predetermined capability is applicable, comprising:
   an acceptance unit that accepts a condition designated by a requester;
   an acquisition unit that acquires a learning result obtained due to predetermined learning being performed by machine learning;
   a determination unit that determines whether a procured capability of the learning result acquired by the acquisition unit satisfies the condition accepted by the acceptance unit; and
   a notification unit that notifies the requester if necessary, based on a result of the determination.
8. A notification method according to which a computer executes the steps of:
   accepting a condition designated by a requester;
   acquiring a learning result obtained due to predetermined learning being performed by machine learning;
   determining whether a capability procured by the acquired learning result satisfies the accepted condition; and
   notifying the requester if necessary, based on a result of the determination.
9. A computer program that causes a computer to function as:
   a unit that accepts a condition designated by a requester;
   a unit that acquires a learning result obtained due to predetermined learning being performed by machine learning;
   a unit that determines whether a capability procured by the acquired learning result satisfies the accepted condition; and
   a unit that notifies the requester if necessary, based on a result of the determination.

Images