JP2018045369A - Recognition device, recognition system, recognition method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To balance loads on apparatuses and shorten a response time until processing results being returned.SOLUTION: A recognition device includes: a reception unit for receiving a request for recognition processing to find a class to which input information belongs from a lower-level device; a recognition unit for executing recognition processing by using dictionary information and outputting a score indicating similarity to the class to which the input information belongs; a determination unit for determining whether to transfer the request to an upper-level device on the basis of the score; a transmission unit for transmitting the request to the upper-level device if it is determined that the request will be transferred; and an update unit for updating the dictionary information on the basis of the input information and the found class.SELECTED DRAWING: Figure 3

Description

  Embodiments described herein relate generally to a recognition apparatus, a recognition system, a recognition method, and a program.

  In cloud services such as image recognition, speech recognition, and speech synthesis, it is common to take a framework in which a request from a user terminal (user terminal) is transmitted to the cloud and the processing result in the cloud is returned to the user terminal. It is. The cloud is generally composed of a plurality of devices, but the number of locations where the cloud itself is installed is small.

Japanese Patent No. 4723170 Japanese Patent No. 4845321 Japanese Patent Laid-Open No. 01-219971

  In the cloud service as described above, it is required to distribute the load of devices in the cloud and to shorten the response time until the processing result is returned.

  The recognition apparatus according to the embodiment executes a recognition process using a dictionary unit and a receiving unit that receives a request for a recognition process for obtaining a class to which the input information belongs, and shows similarity between the class to which the input information belongs. A recognition unit that outputs a score, a determination unit that determines whether to transfer a request to a higher-level device based on the score, and a transmission unit that transmits a request to the higher-level device when it is determined to transfer, An updating unit that updates the dictionary information based on the input information and the obtained class.

FIG. 1 is a block diagram illustrating an example of a configuration of a recognition system according to the present embodiment. FIG. 2 is a diagram for explaining an overview of recognition processing according to the present embodiment. FIG. 3 is a block diagram illustrating an example of functional configurations of the edge server, the master server, and the user terminal. FIG. 4 is a flowchart showing an example of recognition processing in the present embodiment. FIG. 5 is an explanatory diagram illustrating a hardware configuration example of the apparatus according to the present embodiment.

  Exemplary embodiments of a recognition device, a recognition system, a recognition method, and a program according to the present invention will be described below in detail with reference to the accompanying drawings.

  As described above, conventionally, a request is transmitted and processed on a cloud installed in one or a small number of regions. For this reason, for example, even a request from a user terminal at a geographically close point where similar requests are often made may be transmitted and processed on a remote cloud. For this reason, response time may become long. In addition, since all requests are processed uniformly on the remote cloud, the geographical proximity between the requests is not reflected in the processing result.

  As a mechanism for improving the responsiveness to the same request from the user terminal, there is a configuration in which edge servers for relaying the request from the user terminal to the master server in the cloud are provided hierarchically in various places. The edge server has a function of caching a pair of past requests and processing results, for example. The edge server does not transfer the same request as the previous request to the upper server, and responds to the user with the cached result as it is. However, in the case of image recognition or the like where the request is hardly the same as the past, the above configuration is not effective.

  As a mechanism for reflecting the geographical proximity between requests, there is a method of constructing a system in each place in an on-premises form and maintaining a dictionary on the system separately. However, the cost of maintenance increases due to the distributed system.

In the present embodiment, by performing processing according to the following flow, without changing the amount of dictionary collation processing in the entire cloud system, by reducing the request sent to the upper server and the processing amount at the upper server, the cloud Reduce the load on each server. In addition, the edge server has a characteristic of automatically constructing dictionary information (local dictionary) reflecting regional characteristics at a position close to the request generation point. As a result, the responsiveness and response accuracy of the entire system can be efficiently increased.
(1) An edge server as a lower device arranged in various places receives a request for recognition processing from a user terminal.
(2) The server (edge server or highest-level master server) that has received the request executes recognition processing using dictionary information held by itself. If the server is not the master server and the score indicating the similarity of the recognition results is not high, the server transfers the request to a higher-level device (higher-level edge server or master server). In other cases, the server returns the recognition result to the lower server or user terminal.
(3) After the recognition result is returned to the user terminal through all the servers that have been routed, each routed server updates the dictionary held by itself using the recognition result.

  The recognition system of the present embodiment executes pattern recognition processing such as image recognition and voice recognition. In the recognition process, a class to which information (input information: image, sound, etc.) input as a target of the recognition process belongs is obtained. The class to be recognized may be any class. For example, when recognizing an object in an image, a plurality of classes (car, person, animal, etc.) for classifying the object may be set as classes to be recognized. When recognizing a person's face in an image, each person's face may be a recognition target class.

  Applicable recognition processing is not limited to image recognition and voice recognition, and any recognition processing conventionally used can be applied. Hereinafter, a case where the present invention is mainly applied to image recognition will be described as an example.

  FIG. 1 is a block diagram illustrating an example of a configuration of a recognition system according to the present embodiment. As shown in FIG. 1, the recognition system includes a cloud 10 and user terminals 300a to 300d.

  The cloud 10 receives a request for recognition processing from the user terminals 300a to 300d, executes the recognition processing in response to the request, and returns the recognition result to the user terminals 300a to 300d. The cloud 10 and the user terminals 300a to 300d are connected by any network form such as the Internet.

  Below, when it is not necessary to distinguish, user terminal 300a-300d is only called the user terminal 300. FIG. The number of user terminals 300 is not limited to four, and can be any number.

  The user terminal 300 receives a user operation and transmits a request including an image to the adjacent edge server 100 in the cloud 10. For example, the user captures an image including a face using an imaging device (such as a camera) provided in the user terminal 300. In response to a face recognition instruction from the user, the user terminal 300 transmits a request including an image and a request for image recognition processing (face authentication) to the cloud 10. The user terminal 300 receives the processing result of the recognition process from the cloud 10 and displays or accumulates the processing result.

  The user terminal 300 can be configured by mounting a client application having a request transmission function, a processing result reception function, and the like on a mobile phone, a smartphone, a tablet terminal, and a personal computer. The Web service provided by the cloud 10 may be provided with a function equivalent to these client applications, and the user terminal 300 may be configured to access this service.

  The cloud 10 includes a master server 200 and a plurality of edge servers 100a to 100f as recognition devices. The edge servers 100a to 100f are simply referred to as the edge server 100 when it is not necessary to distinguish them. The number of edge servers 100 is not limited to six and can be any number.

  The master server 200 includes a dictionary 250 used for recognition processing by the master server 200. The edge server 100 includes local dictionaries 150 a to 150 f used for recognition processing by the edge server 100. The local dictionaries 150a to 150f are simply referred to as the local dictionary 150 when it is not necessary to distinguish them.

  The cloud 10 includes a plurality of servers configured hierarchically. Servers in the cloud 10 are divided into a master server 200 and an edge server 100. In the master server 200, one or more edge servers 100 are connected to the lower level, and no server is connected to the higher level.

  One or more other edge servers 100 or user terminals 300 are connected to the edge server 100 at the lower level. The edge server 100 is connected to one server (another edge server 100 or master server 200) at the upper level. Although FIG. 1 shows an example of a configuration in which the edge server 100 has two layers, the number of layers of the edge server 100 may be one or three or more.

  For example, the lowermost edge server 100 is installed in any of a plurality of predetermined regions (such as prefectures). The lowermost edge server 100 may further be configured to accept a request from the user terminal 300 existing in the installed area. The middle-tier edge server 100 is installed, for example, in a region (such as a region) that includes a plurality of regions corresponding to each edge server 100 connected to the lower layer.

  By installing the edge server 100 according to the geographically determined area as described above, it is possible to reflect the geographical proximity between requests and improve the responsiveness and response accuracy of recognition processing.

  If the communication speed of the line connecting the edge server 100 and the user terminal 300 is high, an increase in response time can be suppressed. Therefore, in such a case, the area where the edge server 100 is installed may be different from the area where the user terminal 300 connected to the edge server 100 exists.

  Next, an overview of recognition processing according to the present embodiment will be described. FIG. 2 is a diagram for explaining an overview of recognition processing according to the present embodiment.

  First, the lower edge server 100 or the user terminal 300 transmits a request for recognition processing to the upper edge server 100 (step S1). Based on the received request, the upper edge server 100 executes recognition processing using the local dictionary 150 held by itself (step S2).

  If the score of the recognition process is not high, the edge server 100 that has executed the recognition process transfers the request to a higher-level server (another edge server 100 or master server 200) (step S3). The upper server executes recognition processing in response to the request and returns the recognition result to the requesting edge server 100. When the upper server is the edge server 100, the edge server 100 may further determine the score of the recognition process and execute the request transfer process to the upper level.

  The edge server 100 receives the recognition result from the host server (step S4). When the edge server 100 determines that the score is high, the edge server 100 returns the result of the recognition process executed in the own device to the lower edge server 100 or the user terminal 300 (step S5). When the request is transferred to the upper server, the edge server 100 returns the recognition result returned from the upper server to the lower edge server 100 or the user terminal 300.

  Thereafter, the edge server 100 executes an update process of the local dictionary 150 based on the recognition result at an arbitrary timing (step S6). The edge server 100 uses the recognition result of the recognition process when the recognition process is executed in its own device, and uses the recognition result of the recognition process when the recognition process is executed by the upper server. The local dictionary 150 is updated.

  In this way, the edge server 100 can update and maintain a dictionary unique to the area where the device is installed, for example.

  For example, consider an example in which recognition processing of an image of a captured signboard is requested from the user terminal 300 in each region. The user terminal 300 in the same area may be required to recognize the same signboard image, but it is considered that the images themselves are rarely the same. According to the present embodiment, even if the images themselves are different, the dictionary can be updated using the recognition result of the images obtained by capturing the same object as the learning data. This makes it possible to build a dictionary that reflects regional characteristics.

  As another example, consider an example of application to face authentication (face image recognition) of an employee of a company. In this case, the edge server 100 installed in a certain area may hold a dictionary for authenticating employees working at bases in this area. When authentication cannot be performed using this dictionary (when the score is small), authentication processing is performed on the other edge server 100 installed in a wider base or the master server 200 that holds a dictionary for authenticating all employees. Request is forwarded.

  As another example, consider an example in which speech recognition processing using a speech recognition dictionary that is different for each region is executed. The edge server 100 installed in a certain area can maintain and update a dictionary for recognizing speech according to a language (a native language, dialect, etc.) specific to this area.

  Next, a functional configuration example of each device according to the present embodiment will be described. FIG. 3 is a block diagram illustrating an example of functional configurations of the edge server 100, the master server 200, and the user terminal 300.

  The user terminal 300 includes a camera 321, a display unit 322, a control unit 301, a transmission unit 302, and a reception unit 303.

  The camera 321 is an imaging device that captures an image to be subjected to recognition processing. For example, the user of the user terminal 300 uses the camera 321 to capture an image used for face authentication. When performing voice recognition, the user terminal 300 may include a voice input device (such as a microphone) that acquires voice instead of the camera 321 or in addition to the camera 321.

  The display unit 322 displays various information (for example, the result of recognition processing) used in various processing. The display unit 322 can be realized by, for example, a liquid crystal display or a touch panel.

  The transmission unit 302 transmits various information to an external device such as the edge server 100. For example, the transmission unit 302 transmits a request for recognition processing to the edge server 100. The request for recognition processing includes an image (an example of input information) that is a target of recognition processing.

  The receiving unit 303 receives various information from the external device. For example, the receiving unit 303 receives the recognition processing result from the edge server 100 that has transmitted the recognition processing request.

  The control unit 301 controls the entire various processes performed by the user terminal 300. For example, the control unit 301 causes each unit to perform image capturing by the camera 321, transmission of a request for recognition processing including the captured image, reception of a recognition result, and the like.

  Next, a functional configuration example of the edge server 100 will be described. The edge server 100 includes a storage unit 121, a reception unit 101, a recognition unit 102, a determination unit 103, a transmission unit 104, and an update unit 105.

  The storage unit 121 stores various information used in various processes performed by the edge server 100. For example, the storage unit 121 stores dictionary information (such as the local dictionary 150 in FIG. 1) used for recognition processing.

  The receiving unit 101 receives various types of information from external devices such as the user terminal 300, the other edge server 100, and the master server 200 (when connected). For example, the receiving unit 101 receives a request for recognition processing from a lower-level device (the user terminal 300 or another edge server 100). When the edge server 100 is associated with the user terminal 300 in the same region as described above, the reception unit 101 may receive a request from one or more user terminals 300 in the corresponding predetermined region.

  The recognition unit 102 executes recognition processing in response to the request, and outputs a class and a score for this class. The score is a degree indicating the certainty that the input information belongs to the class. Although the algorithm of the recognition process by the recognition unit 102 is arbitrary, for example, the nearest neighbor determination method can be applied.

  The local dictionary 150 is learned by learning data in advance according to an algorithm to be applied, for example, and is stored in the storage unit 121. When the edge server 100 is associated with the user terminal 300 in the same region, the local dictionary 150 may be learned and stored for each region. The recognition unit 102 collates the local dictionary 150 with the input information, and calculates a score for each class included in the local dictionary 150.

For example, when using the nearest neighbor determination method, the recognizing unit 102 recognizes a feature vector and class label pair (x _n, y _n ) (n = 1 to N, N is a learning value ) extracted from a learning image (learning data). The number of data ) is used as the local dictionary 150. The recognition unit 102 uses the feature vector x extracted from the image (input information) included in the request to determine a score s _c indicating the similarity between the class c and the feature vector x using the following equation (1). .

The determination unit 103 determines whether to transfer a request for recognition processing to a higher-level device (another edge server 100 or master server 200) using the score. For example, the determination unit 103 determines to transfer the request to the host device when the score is smaller than a predetermined threshold (first threshold). Score is compared with a threshold value, for example, the maximum value of the scores s _c determined for all classes.

  When it is not determined to transfer, that is, when the score of the recognition result by the recognition unit 102 is equal to or greater than the threshold, the transmission unit 104 displays the recognition result obtained by the recognition unit 102 as a request source (another edge server 100 or user). Terminal 300). The recognition result is, for example, a class that gives the maximum score value.

  The transmission unit 104 transmits a request to the higher-level device when the determination unit 103 determines to transfer. When the request is transferred to the higher-level device, the receiving unit 101 receives the recognition result from the higher-level device that is the transfer destination. The recognition result received from the host device is used as the recognition result of the own device.

  The update unit 105 updates the local dictionary 150 using the recognition result. For example, the update unit 105 adds input information and a class obtained as a recognition result as learning data, and updates the local dictionary 150 by re-learning using the added learning data.

The timing of the update process is arbitrary, but may be executed at the following timing, for example.
・ After each recognition process is executed ・ When the recognition process is executed a predetermined number of times ・ When the load such as CPU usage rate and network usage rate falls below a certain threshold ・ When a predetermined time has passed

  The update unit 105 may be configured to execute the update process when the recognition result is reliable. For example, the updating unit 105 may calculate the reliability of the recognition result and update the dictionary information when the reliability is greater than a predetermined threshold (second threshold). For example, the update unit 105 adds, as a class label, a recognition result having a sufficiently high reliability among the recognition results obtained after the previous update process to the learning data. The update unit 105 re-learns the local dictionary 150 using the added learning data.

The reliability of the recognition result can be calculated by, for example, a determination function represented by the following equation (2) or (3). s _{1 to} s _C are obtained by arranging the scores s _c for the C classes to be recognized in order from the top. α _c and β _c are constants.

The reliability is not limited to the above example. For example, the determination function may be represented by at least one linear combination among the values obtained as follows.
-Multiple scores calculated for multiple classes-A value obtained by multiplying the value of the score by a constant-A value obtained by multiplying the value of the score by a constant and giving it to the exponential function-Any linear form of the above values Logarithm of sum For example, when the score for a certain class matches the likelihood or logarithmic likelihood for that class except for a constant multiple or a constant difference, the posterior probability can be used as the reliability.

  In addition, when the number of determination functions that exceed a predetermined threshold (third threshold) among a plurality of determination functions is defined as reliability, and the reliability (number of determination functions) is greater than the threshold (second threshold), The update unit 105 may perform update processing.

  Next, a functional configuration example of the master server 200 will be described. The master server 200 includes a storage unit 221, a reception unit 201, a recognition unit 202, a transmission unit 204, and an update unit 205.

  The master server 200 differs from the edge server 100 in that it does not include at least the determination unit 103 and does not transfer a request according to the determination result of the determination unit 103.

  The storage unit 221 stores various information used in various processes by the master server 200. For example, the storage unit 221 stores dictionary information (such as the dictionary 250 in FIG. 1) used for recognition processing. The dictionary 250 is learned by learning data in advance according to an algorithm to be applied, for example, and is stored in the storage unit 221.

  The receiving unit 201 receives various types of information from an external device such as the edge server 100. For example, the receiving unit 201 receives a request for recognition processing from the lower edge server 100.

  The recognition unit 202 executes recognition processing in response to a request, and outputs a class and a score for this class. The recognition process by the recognition unit 202 is the same as the recognition process by the edge server 100 except that the dictionary 250 stored in the storage unit 221 is used.

  The transmission unit 204 transmits the recognition result obtained by the recognition unit 202 to the edge server 100 that is a request source.

  The update unit 205 updates the dictionary 250 using the recognition result. The update process by the update unit 205 is the same as the update process by the update unit 105 of the edge server 100.

  Each of the above units may be realized by causing a processing device such as a CPU (Central Processing Unit) to execute a program, that is, realized by software or hardware such as an IC (Integrated Circuit). It may be realized by using software and hardware together.

  Each storage unit (storage unit 121, storage unit 221) can be configured by any commonly used storage medium such as a hard disk drive (HDD), an optical disk, a memory card, and a random access memory (RAM). .

  Next, recognition processing by the edge server 100 according to the present embodiment configured as described above will be described with reference to FIG. FIG. 4 is a flowchart showing an example of recognition processing in the present embodiment.

  The receiving unit 101 waits for reception of a request for recognition processing from a lower-level device (the user terminal 300 or another edge server 100) (step S101). The receiving unit 101 determines whether or not a request for recognition processing has been received (step S102). When the request is received (step S102: Yes), the recognition unit 102 executes recognition processing according to the request (step S103).

  The determination unit 103 determines whether or not the score obtained by the recognition process is smaller than the threshold (step S104). When the score is smaller than the threshold value (step S104: Yes), the transmission unit 104 transfers the recognition process request to the upper server (other edge server 100 or master server 200) (step S105).

  After transferring the request or when the request has not been received in step S102 (step S102: No), the receiving unit 101 determines whether or not a recognition result has been received from a higher-level server (step S106).

  When the recognition result has not been received (step S106: No), the process returns to step S101 and is repeated. When the recognition result is received (step S106: Yes), the receiving unit 101 records the received recognition result in, for example, the storage unit 121 (step S107). When it is determined in step S104 that the score is equal to or greater than the threshold (step S104: No), the recognition unit 102 records the result of recognition processing by itself in the storage unit 121, for example. The transmission unit 104 transmits the recorded recognition result to the request source (step S108).

  The updating unit 105 determines whether it is the dictionary update timing (step S109). If it is not the update timing (step S109: No), the process returns to step S101 and is repeated. When it is an update timing (step S109: Yes), the update part 105 adds a recognition result to learning data, and performs the update process of a dictionary (step S110).

  In the case of the master server 200, steps S104 to S106 are not executed, and the processing after step S107 is executed after the recognition processing (step S103). In this case, after step S102: No, the process may transition to step S101.

(Modification)
In the above embodiment, an example in which the nearest neighbor determination method is used for the recognition process has been described. The recognition process is not limited to this. As the dictionary update process, an arbitrary update process according to the algorithm of the recognition process to be applied can be applied.

  For example, in the recognition process, a subspace method, a modified secondary discriminant function method, or a spherical pseudo Bayes method may be used. In this case, in the update process, a learning subspace method or an average learning subspace method may be used.

  In the recognition process, a recognition process using a neural network may be used. In this case, the error back propagation method may be used in the update process.

  As described above, according to the present embodiment, it is possible to distribute the load on the device and shorten the response time.

  Next, the hardware configuration of each device (user terminal, edge server, master server) according to the present embodiment will be described with reference to FIG. FIG. 5 is an explanatory diagram illustrating a hardware configuration example of the apparatus according to the present embodiment.

  The apparatus according to the present embodiment communicates with a control device such as a CPU (Central Processing Unit) 51 and a storage device such as a ROM (Read Only Memory) 52 and a RAM (Random Access Memory) 53 by connecting to a network. A communication I / F 54 and a bus 61 for connecting each unit are provided.

  A program executed by the apparatus according to the present embodiment is provided by being incorporated in advance in the ROM 52 or the like.

  A program executed by the apparatus according to the present embodiment is a file in an installable format or an executable format, and is a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), or a CD-R (Compact Disk Recordable). Alternatively, the program may be recorded on a computer-readable recording medium such as a DVD (Digital Versatile Disk) and provided as a computer program product.

  Furthermore, the program executed by the apparatus according to the present embodiment may be provided by being stored on a computer connected to a network such as the Internet and downloaded via the network. The program executed by the apparatus according to the present embodiment may be provided or distributed via a network such as the Internet.

  The program executed by the apparatus according to the present embodiment can cause a computer to function as each unit of the above-described apparatus. In this computer, the CPU 51 can read a program from a computer-readable storage medium onto a main storage device and execute the program.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 100 Edge server 101 Reception part 102 Recognition part 103 Determination part 104 Transmission part 105 Update part 121 Storage part 200 Master server 201 Reception part 202 Recognition part 204 Transmission part 205 Update part 221 Storage part 300 User terminal 301 Control part 302 Transmission part 303 Reception Section 321 Camera 322 Display section

Claims (11)

A receiving unit for receiving a request for recognition processing for obtaining a class to which the input information belongs, from a lower-level device;
A recognition unit that performs the recognition process using dictionary information and outputs a score indicating similarity to the class to which the input information belongs;
A determination unit that determines whether or not to transfer the request to a higher-level device based on the score;
A transmission unit that transmits the request to the higher-level device when it is determined to transfer,
An update unit that updates the dictionary information based on the input information and the determined class;
A recognition device comprising: The determination unit determines to transfer the request to the higher-level device when the score is smaller than a first threshold.
The recognition device according to claim 1. The receiving unit receives the request from one or more subordinate devices in a predetermined area,
The recognizing unit executes the recognition process using the dictionary information determined according to the region.
The recognition device according to claim 1. The update unit calculates the reliability of the obtained class, and updates the dictionary information when the reliability is greater than a second threshold;
The recognition device according to claim 1. The update unit calculates the reliability based on a plurality of scores calculated for a plurality of classes.
The recognition apparatus according to claim 4. The update unit updates the dictionary information using learning data to which information that associates the input information with the obtained class is added.
The recognition apparatus according to claim 4. The recognition unit performs the recognition process by a subspace method, a modified secondary discriminant function method, or a spherical pseudo Bayes method,
The update unit updates the dictionary information by a learning subspace method or an average learning subspace method.
The recognition apparatus according to claim 4. The recognition unit performs a recognition process using a neural network,
The updating unit updates the dictionary information by an error back propagation method.
The recognition apparatus according to claim 4. A recognition system comprising a lower device, a recognition device, and a higher device,
The subordinate device is:
A first transmission unit that transmits a recognition processing request for obtaining a class to which the input information belongs to the recognition device;
The host device includes a first recognition unit that executes the recognition process in response to the request,
The recognition device is
A receiving unit that receives the request from the lower-level device;
Executing a recognition process using dictionary information, and outputting a class and score to which the input information belongs;
A determination unit that determines whether or not to transfer the request to the host device based on the score;
A second transmission unit that transmits the request to the higher-level device when it is determined to transfer,
An update unit that updates the dictionary information based on the input information and the determined class,
Recognition system. A receiving step of receiving a request for recognition processing for obtaining a class to which the input information belongs, from a lower-level device;
A recognition step of performing the recognition process using dictionary information and outputting a class and score to which the input information belongs;
A determination step of determining whether to transfer the request to a higher-level device based on the score;
A transmission step of transmitting the request to the higher-level device when it is determined to transfer,
An updating step for updating the dictionary information based on the input information and the determined class;
A recognition method including: Computer
A receiving unit for receiving a request for recognition processing for obtaining a class to which the input information belongs, from a lower-level device;
A recognition unit that executes the recognition process using dictionary information and outputs a class and a score to which the input information belongs;
A determination unit that determines whether or not to transfer the request to a higher-level device based on the score;
A transmission unit that transmits the request to the higher-level device when it is determined to transfer,
An update unit that updates the dictionary information based on the input information and the determined class;
Program to function as.

Images