Data analysis system, data analysis device and data analysis method

The data analysis system addresses the challenge of distinguishing between multiple trained models by displaying them with associated training data and accuracy information, facilitating easy selection and improving usability.

JP2024159229APending Publication Date: 2024-11-08SHIMADZU SEISAKUSHO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023075080
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-04-28
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing data analysis systems struggle to easily distinguish and understand multiple trained models with different versions, making it difficult to select the appropriate model for analysis.

Method used

A data analysis system that associates and displays multiple trained models with the same analysis purpose but different versions on the same screen, along with their associated training data and accuracy information, allowing for easy comparison and selection.

Benefits of technology

Enables efficient selection of trained models by displaying version histories and accuracy evaluations, improving the understanding and usability of multiple trained models with different versions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024159229000001_ABST
    Figure 2024159229000001_ABST
Patent Text Reader

Abstract

To provide a data analysis system that allows multiple learned models with different versions to be easily grasped at a glance.SOLUTION: A data analysis system 100 comprises: a data acquisition unit 10; an analysis unit 12a which analyzes analysis object data 20 using a learned model 30; a learned model generation unit 12b which generates the learned model 30; a storage unit 11 which associates each of the generated learned models 30 with teacher data 21 used when generating them and stores the multiple learned models 30 with the same analysis purpose but the different versions in an associated manner; a display unit 2a which displays the learned models 30; and a control unit 12 which performs control so as to display, on the same screen, the selected learned model 30 among the displayed multiple learned models 30 and the learned model 30 with the different version associated with the selected learned model 30.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a data analysis system, a data analysis device, and a data analysis method, and more particularly to a data analysis system, a data analysis device, and a data analysis method that generate a trained model and analyze analysis target data using the generated trained model. [Background technology]

[0002] Conventionally, a data analysis system is known that generates a trained model and analyzes data to be analyzed using the generated trained model (see, for example, Patent Document 1).

[0003] The above-mentioned Patent Document 1 discloses a learning support system including a measuring device, a display unit, a storage unit, and a control unit. The above-mentioned Patent Document 1 discloses a configuration in which a plurality of trained models are displayed on a display unit, and a trained model is generated based on a trained model selected from the plurality of trained models. The learning support system disclosed in the above-mentioned Patent Document 1 is configured to analyze a cell image using the generated trained model. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2022-174740 A Summary of the Invention [Problem to be solved by the invention]

[0005] Here, although not disclosed in the above Patent Document 1, when analyzing a cell image (analysis target data), in order to improve the analysis accuracy of the analysis target data, multiple trained models with different versions may be generated by changing the teacher data and learning parameters. In this case, when analyzing the analysis target data, the operator selects a trained model to be used for analyzing the analysis target data from among multiple trained models with different versions. However, in the configuration disclosed in the above Patent Document 1, multiple trained models are displayed in a list on the display unit, so when the configuration of this Patent Document 1 is applied to a configuration that displays multiple trained models with different versions, there is a disadvantage that it is difficult to grasp the multiple trained models with different versions at a glance. Therefore, a technology that makes it possible to easily grasp multiple trained models with different versions at a glance is desired.

[0006] The present invention has been made to solve the above-mentioned problems, and one object of the present invention is to provide a data analysis system, a data analysis device, and a data analysis method that make it possible to easily understand multiple trained models of different versions at a glance. [Means for solving the problem]

[0007] In order to achieve the above object, a data analysis system according to a first aspect of the present invention is a data analysis system that analyzes analysis target data, which is data to be analyzed, using a trained model, and includes a data acquisition unit that acquires the analysis target data, an analysis unit that analyzes the analysis target data using the trained model, a trained model generation unit that generates a trained model using at least teacher data, a storage unit that associates each of the trained models generated by the trained model generation unit with the teacher data used in generating the trained models and associates and stores a plurality of trained models having the same analysis purpose and different versions, a display unit that displays the plurality of trained models stored in the storage unit, and a control unit that controls displaying a selected trained model and a trained model of a different version associated with the selected trained model on the same screen from the plurality of trained models displayed on the display unit. Here, "different versions" means that the source code described for generating a network structure is the same, and the teacher data and learning parameters, etc. are changed.

[0008] In addition, a data analysis device according to a second aspect of the present invention is a data analysis device that analyzes analysis target data, which is data to be analyzed, using a trained model, and includes a data acquisition unit that acquires the analysis target data, an analysis unit that analyzes the analysis target data using the trained model, a trained model generation unit that generates a trained model using at least teacher data, a memory unit that associates each of the multiple trained models generated by the trained model generation unit with the teacher data used when generating them, and associates and stores multiple trained models that have the same analysis purpose and different versions, a display unit that displays the multiple trained models stored in the memory unit, and a control unit that controls the display of a selected trained model from the multiple trained models displayed on the display unit and a trained model of a different version associated with the selected trained model on the same screen.

[0009] In addition, a data analysis method according to a third aspect of the present invention is a data analysis method for analyzing analysis target data, which is data to be analyzed, using a trained model, and includes the steps of generating a trained model using at least teacher data, associating each of the multiple trained models generated by the trained model generation unit with the teacher data used when generating the models, and associating and storing multiple trained models that have the same analysis purpose but different versions, and displaying a selected trained model from the multiple stored trained models and a trained model of a different version associated with the selected trained model on the same screen. Effect of the Invention

[0010] The data analysis system in the first aspect and the data analysis device in the second aspect include a display unit that displays a plurality of trained models stored in a storage unit, and a control unit that controls displaying a selected trained model and a trained model having a different version from the selected trained model on the same screen. The data analysis method in the third aspect includes a step of displaying a selected trained model and a trained model having a different version associated with the selected trained model on the same screen, among the plurality of trained models stored, as described above. As a result, the selected trained model and a trained model having a different version associated with the selected trained model are displayed on the same screen among the plurality of trained models stored, so that the plurality of trained models having different versions can be easily understood at a glance. As a result, it is possible to provide a data analysis system, a data analysis device, and a data analysis method that can easily understand a plurality of trained models having different versions at a glance. As a result, when analyzing data to be analyzed, an operator can efficiently select a trained model to be used for analyzing data to be analyzed from among a plurality of trained models having different versions. [Brief description of the drawings]

[0011] [Figure 1] 1 is a schematic diagram showing an overall configuration of a data analysis system according to an embodiment. [Diagram 2] FIG. 1 is a schematic diagram illustrating a configuration in which analysis target data according to one embodiment associates multiple trained models having the same analysis purpose but different versions. [Diagram 3] A schematic diagram showing an example screen for selecting a learning model, etc., when generating a trained model in a data analysis system according to one embodiment. [Figure 4] A schematic diagram showing an example of a screen displaying multiple trained models in a data analysis system according to one embodiment. [Diagram 5] A schematic diagram showing an example screen that displays the version history of trained models with different versions that have the same analysis purpose in a data analysis system according to one embodiment. [Figure 6] FIG. 13 is a schematic diagram showing an evaluation screen for the analysis accuracy of multiple trained models of different versions that have the same analysis purpose in a data analysis system according to one embodiment. [Figure 7] FIG. 13 is a schematic diagram showing an example of a screen for setting an analysis recipe when performing analysis using a trained model in a data analysis system according to one embodiment. [Figure 8] A schematic diagram showing an example screen displaying multiple trained models of different versions with the same analysis purpose when selecting an analysis recipe in a data analysis system according to one embodiment. [Figure 9] 13 is a flowchart for explaining a process for displaying multiple trained models of different versions having the same analysis purpose in a data analysis system according to one embodiment. [Figure 10] A flowchart for explaining a process of determining whether the version name of a trained model is the latest in a data analysis system according to one embodiment. [Figure 11] FIG. 13 is a schematic diagram showing an overall configuration of a data analysis system according to a first modified example. [Figure 12]FIG. 13 is a schematic diagram showing an overall configuration of a data analysis system according to a second modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

[0013] A configuration of a data analysis system 100 according to an embodiment will be described with reference to FIG. 1. The data analysis system 100 is a data analysis system that analyzes analysis target data 20, which is data to be analyzed, using a trained model 30. The data analysis system 100 is also a system that generates the trained model 30. For example, the data analysis system 100 analyzes an image showing a cell as the analysis target data 20. Note that the analysis target data 20 analyzed by the data analysis system 100 is not limited to an image showing a cell. For example, the analysis target data 20 may include an X-ray image.

[0014] (Data analysis system configuration) As shown in FIG. 1, the data analysis system 100 includes a server 1 and an analysis data display device 2.

[0015] FIG. 1 shows an example of a data analysis system 100 constructed in an on-premise client-server model. The analysis data display device 2 functions as a client terminal in the data analysis system 100. The server 1 functions as a server that performs data analysis in the data analysis system 100. The server 1 and the analysis data display device 2 are connected to each other via a network 90 so as to be able to communicate with each other. The server 1 performs various information processes in response to a request (processing request) from the analysis data display device 2 operated by a user. The server 1 analyzes the analysis target data 20 in response to the request. Note that on-premise means that the server 1 and the analysis data display device 2 are located in the same facility.

[0016] In addition, in response to a request, the server 1 creates a trained model display screen 41 that displays the trained model 30. In addition, the server 1 transmits the created trained model display screen 41 to the analysis data display device 2.

[0017] The network 90 connects the server 1 and the analysis data display device 2 so that they can communicate with each other. The network 90 can be, for example, a local area network (LAN) established within a facility.

[0018] The analysis data display device 2 is an analysis data display device that displays an analysis result obtained by analyzing analysis target data 20, which is data to be analyzed. The analysis data display device 2 is also configured to display a trained model 30 used to analyze the analysis target data 20. The analysis data display device 2 is a so-called personal computer, and includes a processor (not shown) and a storage unit (not shown). The analysis data display device 2 also includes a display unit 2a and an input receiving unit 2b.

[0019] The display unit 2a displays at least one of the analysis target data 20 and the trained model 30. In this embodiment, the display unit 2a displays a plurality of trained models 30 stored in a storage unit 11 described later. The display unit 2a is, for example, a display device such as a liquid crystal monitor or an organic EL (Electro Luminescence) monitor. The display unit 2a may be a projector or a head-mounted display.

[0020] The input receiving unit 2b is configured to receive input from an operator. The input receiving unit 2b is, for example, an input device including a mouse and a keyboard. The input receiving unit 2b may be a touch panel. One or more analysis data display devices 2 are provided in the data analysis system 100.

[0021] The server 1 includes a data acquisition unit 10, a storage unit 11, and a control unit 12.

[0022] The data acquiring unit 10 is configured to acquire the analysis target data 20. Specifically, the data acquiring unit 10 is configured to acquire the analysis target data 20 input from the analysis data display device 2 via a network 90. ​​In addition, in this embodiment, the data acquiring unit 10 is configured to acquire the teacher data 21 input from the analysis data display device 2 via the network 90. ​​The teacher data 21 is teacher data used to generate the trained model 30. The data acquiring unit 10 includes, for example, an input / output interface.

[0023] The storage unit 11 stores various programs executed by the control unit 12. The storage unit 11 is configured to store the analysis target data 20 and the trained model 30. In this embodiment, the storage unit 11 is configured to associate the plurality of trained models 30 generated by the trained model generating unit 12b described later with the teacher data 21 used when generating each of the trained models 30, and to store the trained models 30 having the same analysis purpose and different versions in association with each other. The storage unit 11 is configured to store information 22 regarding the accuracy of the analysis result described later and learning parameters 23 described later. The analysis purpose means what kind of analysis is performed on what kind of analysis target data 20. The analysis purpose includes information on what kind of processing is performed, such as segmentation processing of cell images and segmentation processing of X-ray images. The same analysis purpose means that the type of analysis on the analysis target data 20 is the same.

[0024] The storage unit 11 includes a non-volatile storage device such as a hard disk drive (HDD) or a solid state drive (SSD). Details of the configuration in which the storage unit 11 stores the trained model 30 and the teacher data 21 in association with each other, and the configuration in which the storage unit 11 stores a plurality of trained models 30 of different versions having the same analysis purpose in association with each other will be described later.

[0025] The trained model 30 is, for example, a convolutional neural network (CNN) or is generated by training a learning model that includes a convolutional neural network as a part of the learning model using the teacher data 21.

[0026] The control unit 12 is configured to analyze the analysis target data 20. The control unit 12 is also configured to perform control to display a plurality of trained models 30 on the display unit 2a. The control unit 12 includes a CPU (Central Processing Unit), a circuit, a ROM (Read Only Memory), a RAM (Random Access Memory), a GPU (Graphics Processing Unit), or an FPGA (Field-Programmable Gate Array) configured for image processing.

[0027] The control unit 12 includes an analysis unit 12a and a trained model generation unit 12b. The analysis unit 12a and the trained model generation unit 12b are configured in a software manner as functional blocks realized by the control unit 12 executing a program. The analysis unit 12a and the trained model generation unit 12b may be configured in a hardware manner by providing a dedicated processor (processing circuit).

[0028] The analysis unit 12a is configured to analyze the analysis target data 20 using the trained model 30. The analysis unit 12a analyzes the analysis target data 20 using the trained model 30 selected by the operator. For example, when the analysis target data 20 is a cell image, the analysis unit 12a uses the trained model 30 to analyze whether the cells shown in the cell image are differentiated or remain undifferentiated. Note that the analysis process performed by the analysis unit 12a may differ depending on the analysis target data 20 and the trained model 30 used by the analysis unit 12a.

[0029] The trained model generation unit 12b is configured to generate a trained model 30 using at least the teacher data 21. In this embodiment, the trained model generation unit 12b generates the trained model 30 based on the learning model algorithm 24 and any one of the trained models 30 selected by the operator, and the teacher data 21 selected by the operator. The learning model algorithm 24 includes, for example, a neural network.

[0030] The configuration in which the control unit 12 controls the display of the trained model 30 will be described in detail later.

[0031] (Version management of trained models) Next, a configuration in which the control unit 12 (see FIG. 1) manages the version of the trained model 30 (see FIG. 1) will be described with reference to FIG.

[0032] In this embodiment, the control unit 12 performs version management of the trained models 30 by associating multiple trained models 30 of different versions having the same analysis purpose and storing them in the memory unit 11 (see Figure 1).

[0033] In the example shown in Fig. 2, a first trained model 30a to a third trained model 30c, which have the same analysis purpose but different versions, are stored in association with each other. The first trained model 30a, the second trained model 30b, and the third trained model 30c are trained models 30 that have the same learning model algorithm 24 (see Fig. 1) and different version names 25.

[0034] In this embodiment, the control unit 12 is configured to store the analysis target data 20, teacher data 21, information 22 regarding the accuracy of the analysis results, learning parameters 23, and version name 25 in association with the trained model 30.

[0035] Moreover, the teacher data 21 includes learning data 21a, validity verification data 21b, and accuracy verification data 21c. The learning data 21a is data used for learning when generating the trained model 30. The validity verification data 21b is data for verifying the validity of the generated trained model 30. The accuracy verification data 21c is data for verifying the accuracy of the generated trained model 30.

[0036] In this embodiment, the storage unit 11 is configured to store trained models 30 that have the same algorithm 24 of the learning models and that have been generated in a state in which at least one of the teacher data 21 or the learning parameters 23 set when generating the trained model 30 has been changed, in association with each other as different versions of the trained models 30. The same algorithm 24 of the learning models means that the network structures of the algorithms 24 of the learning models are the same.

[0037] In this embodiment, the storage unit 11 (see FIG. 1 ) is configured to associate information 22 on the accuracy of the analysis result of the analysis target data 20 with each of the multiple trained models 30 together with the teacher data 21, and to store multiple trained models 30 of different versions having the same analysis purpose in association with each other. The information 22 on the accuracy of the analysis result includes first analysis accuracy information 22a, which is information 22 on the accuracy of the analysis result of multiple trained models 30 of different versions having the same analysis purpose, and second analysis accuracy information 22b, which is information 22 on the accuracy of the analysis result when different analysis target data 20 are analyzed. Therefore, in this embodiment, the storage unit 11 is configured to associate and store the first analysis accuracy information 22a and the second analysis accuracy information 22b with each trained model 30.

[0038] Here, the data analysis system 100 according to this embodiment is configured to generate a trained model 30 and analyze the analysis target data 20 using the trained model 30.

[0039] (Generating a trained model) First, a configuration in which the data analysis system 100 (see FIG. 1) generates the trained model 30 will be described with reference to FIGS. 3 to 6.

[0040] A trained model generation screen 40 shown in FIG. 3 is a screen that is displayed on the display unit 2a (see FIG. 1) when the operator generates the trained model 30 (see FIG. 1).

[0041] The trained model generation screen 40 displays a name input field 40a, a project selection field 40b, a model selection field 40c, a dataset selection field 40d, a trained model selection field 40e, a version name input field 40f, a summary input field 40g, a training parameter input field 40h, a generate button 40i, and a cancel button 40j.

[0042] The name input field 40a is an input field in which the name of the trained model 30 to be generated is input. Moreover, the project selection field 40b is a selection field in which the project of the trained model 30 to be generated is selected. Note that the project of the trained model 30 is a category set for the trained model 30. In other words, the trained model 30 is stored in the storage unit 11 for each selected project.

[0043] The model selection field 40c is a selection field for selecting the learning model algorithm 24 (see FIG. 1) when generating the trained model 30. The dataset selection field 40d is a selection field for selecting the teacher data 21 (see FIG. 1) used to generate the trained model 30.

[0044] Further, the trained model selection field 40e is a selection field for the trained model 30 used to generate the trained model 30. By selecting the trained model 30 using the trained model selection field 40e, the trained model 30 can be generated based on the trained model 30 stored in the storage unit 11. That is, a trained model 30 that is an improvement of an already generated trained model 30 can be generated. In this embodiment, the control unit 12 is configured to perform control to cause the display unit 2a to selectably display multiple versions of the trained model 30 stored in the storage unit 11 when generating trained models 30 of different versions having the same analysis purpose.

[0045] Furthermore, the version name input field 40f is an input field for inputting the version name 25 of the trained model 30 to be generated. Furthermore, the summary input field 40g is an input field for inputting a summary of the trained model 30 to be generated, comments from the operator, etc. Furthermore, the learning parameter input field 40h is an input field for inputting the learning parameters 23 (see FIG. 1) used when generating the trained model 30. In the example shown in FIG. 3, the learning parameter input field 40h includes an epoch number input field 40k for inputting the number of epochs, and a batch size input field 40l for inputting the batch size.

[0046] The name input field 40a, the version name input field 40f, the summary input field 40g, the number of epochs input field 40k, and the batch size input field 40l are each, for example, a text box, and the project selection field 40b, the model selection field 40c, the dataset selection field 40d, and the trained model selection field 40e are each, for example, a combo box.

[0047] Moreover, the generate button 40i is a button of a GUI (Graphical User Interface) displayed on the trained model generation screen 40. When the generate button 40i is pressed, the trained model generation unit 12b (see FIG. 1) generates the trained model 30 based on the information entered in each input field and the information selected in each selection field. For example, when the generate button 40i is pressed with the algorithm 24 of the training model selected in the model selection field 40c, the trained model generation unit 12b generates the trained model 30 using the algorithm 24 of the selected training model.

[0048] Furthermore, for example, when the Generate button 40i is pressed while the trained model selection field 40e is selected, the trained model generation unit 12b is configured to be able to generate trained models 30 of different versions having the same analysis purpose based on a selected trained model 30 from among the multiple trained models 30 stored in the memory unit 11.

[0049] In addition, in this embodiment, when a trained model 30 with the same analysis objective is generated when a trained model 30 with the same analysis objective is selected on the trained model display screen 41, the trained model generation unit 12b is configured to generate a trained model 30 of a different version with the same analysis objective by training in a state in which at least one of the teacher data 21 (see Figure 1) or the learning parameters 23 (see Figure 1) that are set when generating the trained model 30 is changed.

[0050] In addition, in this embodiment, when different versions of trained models 30 with the same analysis purpose are generated, the control unit 12 is configured to perform control to sequentially add the latest version of the trained model 30 as an option.

[0051] Moreover, the cancel button 40j is a push button of a GUI displayed on the trained model generation screen 40. When the cancel button 40j is pressed, the control unit 12 is configured to perform control to close the trained model generation screen 40 without generating the trained model 30.

[0052] (Version name management) In this embodiment, the control unit 12 uses semantic versioning as a naming rule for the version name 25 (see FIG. 2) of the trained model 30. Semantic versioning is a rule for naming the version name 25 by separating three numbers called the major version, minor version, and patch version with commas. Note that the naming rule may include naming rules other than semantic versioning.

[0053] In addition, in this embodiment, the control unit 12 is configured to determine the version name 25 of the trained model 30 based on a naming rule set so that the version name 25 of the trained model 30 does not branch. The naming rule set so that the version name 25 of the trained model 30 newly generated is always the latest version name 25. For example, the naming rule set so that the version name 25 of the trained model 30 does not branch is a rule that prohibits the generation of a trained model 30 with the version name 25 of "1.1.2" when a trained model 30 with the version name 25 of "1.1.1" and a trained model 30 with the version name 25 of "1.1.3" are stored in the storage unit 11.

[0054] Therefore, in this embodiment, the control unit 12 is configured to determine whether or not the version name 25 input by the input receiving unit 2b is the latest version name 25 based on the naming rule, and if the version name 25 is not the latest version name 25 based on the naming rule, to perform control not to cause the trained model generation unit 12b to generate the trained model 30. Specifically, the control unit 12 compares the major version, minor version, and patch version in the version name 25 of each trained model 30, and determines whether the version name 25 is the latest.

[0055] (Displaying multiple trained models) Here, when analyzing the analysis target data 20 (see FIG. 1) using the trained model 30 (see FIG. 1), it is considered that at least one of the teacher data 21 (see FIG. 1) and the learning parameters 23 (see FIG. 1) is changed to generate multiple trained models 30, and the trained model 30 with the highest analysis accuracy is used to perform the analysis. In this case, since multiple trained models 30 with the same analysis purpose but different versions are generated, when the multiple trained models 30 are configured to be displayed in a list on the display unit 2a (see FIG. 1), it becomes difficult to grasp at a glance the multiple trained models 30 with the same analysis purpose but different versions. Therefore, in this embodiment, the control unit 12 (see FIG. 1) is configured to display multiple trained models 30 with the same analysis purpose but different versions in a comparable manner.

[0056] A configuration in which the control unit 12 displays a plurality of trained models 30 (see FIG. 1) having the same analysis purpose but different versions on the display unit 2a will be described with reference to FIGS. 4 and 5.

[0057] The trained model display screen 41 shown in Fig. 4 is a screen that displays a plurality of trained models 30 stored in the storage unit 11 (see Fig. 1). The trained model display screen 41 includes a trained model display area 41a. The trained models 30 displayed on the trained model display screen 41 are trained models 30 having different analysis purposes (see Fig. 1). In other words, the trained model display screen 41 is a screen that displays a list of a plurality of trained models 30 having different analysis purposes.

[0058] The trained model display area 41a includes a name display field 41b, a version name display field 41c, and a project name display field 41d.

[0059] The name display field 41b is a field that displays the name of the trained model 30. Moreover, the version name display field 41c is a field that displays the version name 25 (see FIG. 2) of the trained model 30. Moreover, the project name display field 41d is a field that displays the project name of the trained model 30. Moreover, in the project name display field 41d, a details button 41e and an update button 41f are displayed for each trained model 30.

[0060] The details button 41e is a push button of the GUI. When the details button 41e is pressed, a details screen of the corresponding trained model 30 is displayed. The details screen includes a version history screen 42 shown in FIG. 5 and an accuracy evaluation screen 43 shown in FIG.

[0061] (View version history of trained models) The version history screen 42 shown in Figure 5 is a screen that displays the trained model 30 (see Figure 1) selected on the trained model display screen 41 (see Figure 4) and trained models 30 of different versions that have the same analysis objective associated with the selected trained model 30.

[0062] In this embodiment, the control unit 12 (see FIG. 1) is configured to control the display of a selected trained model 30 and a trained model 30 of a different version associated with the selected trained model 30 on the same screen among the multiple trained models 30 displayed on the display unit 2a (see FIG. 1). In this embodiment, the control unit 12 is configured to control the display of a selected trained model 30 and a trained model 30 of a different version having the same analysis objective associated with the selected trained model 30 on the same screen. Specifically, the control unit 12 is configured to display the selected trained model 30 and a trained model 30 of a different version having the same analysis objective associated with the selected trained model 30 on the same screen by controlling the display unit 2a to display a version history screen 42. Note that the trained models 30 displayed by scrolling the screen are included in the trained models 30 displayed on the same screen.

[0063] The version history screen 42 includes a detailed information display field 42a that displays detailed information about a plurality of trained models 30. In the example shown in Fig. 5, the screen includes three detailed information display fields 42a for "trained model A" that have different version names 25 (see Fig. 2), and a detailed information display field 42a for the same version name 25 but different validation datasets. Specifically, in the example shown in Fig. 5, the screen includes four detailed information display fields 42a: one with the version name 25 of "1.0.0" and a validation dataset A, one with the version name 25 of "1.0.0." and a validation dataset B, one with the version name 25 of "1.0.1" and a validation dataset A, and one with the version name 25 of "2.0.0" and a validation dataset A.

[0064] Each of the detailed information display fields 42a includes a version name display field 42b, a name display field 42c, a model name display field 42d, a dataset display field 42e, a trained model display field 42f, an overview display field 42g, a learning parameter display field 42h, a verification dataset display field 42i, an inference parameter display field 42j, and an accuracy display field 42k.

[0065] The version name display field 42b is a display field that displays the version name 25 of the trained model 30. The name display field 42c is a display field that displays the name of the trained model 30. The model name display field 42d is a display field that displays the learning model algorithm 24 (see FIG. 1) used when generating the trained model 30. The dataset display field 42e is a display field that displays the teacher data 21 used when generating the trained model 30. The trained model display field 42f is a display field that displays the trained model 30 used when generating the trained model 30. The summary display field 42g is a display field that displays the summary input by the operator when generating the trained model 30. The learning parameter display field 42h is a display field that displays the learning parameters 23 (see FIG. 1) used when generating the trained model 30. The verification dataset display field 42i is a display field that displays the analysis target data 20 (see FIG. 1) analyzed by the generated trained model 30. The inference parameter display field 42j is a display field that displays the inference parameters that were set when the analysis target data 20 was analyzed using the generated trained model 30. The accuracy display field 42k is a display field that displays information 22 (see FIG. 1) regarding the accuracy of the analysis result when the analysis target data 20 was analyzed using the generated trained model 30.

[0066] In this embodiment, the analysis unit 12a (see FIG. 1) is configured to analyze the analysis target data 20 using the generated trained model 30 based on the operation of the operator. When the analysis unit 12a analyzes the analysis target data 20, the control unit 12 stores the analysis target data 20, the inference parameters set during the analysis, and information 22 related to the accuracy of the analysis result in the storage unit 11 in association with the trained model 30. Therefore, the control unit 12 can display the analysis target data 20 analyzed by the generated trained model 30, the inference parameters, and the information 22 related to the accuracy of the analysis result in the verification dataset display field 42i, the inference parameter display field 42j, and the accuracy display field 42k of the version history screen 42.

[0067] (Displaying the accuracy evaluation of the trained model) Here, when improving the trained model 30, it is considered to generate a plurality of trained models 30 by changing at least one of the teacher data 21 (see FIG. 1) and the learning parameters 23 (see FIG. 1). However, even if a plurality of trained models 30 are generated by changing at least one of the teacher data 21 and the learning parameters 23, if the information 22 on the accuracy of the analysis results of the plurality of trained models 30 cannot be compared, it is difficult to grasp at a glance which trained model 30 is suitable for analyzing the analysis target data 20. Therefore, in this embodiment, the control unit 12 is configured to perform control to display the information 22 on the accuracy of the analysis results of the plurality of trained models 30 on the display unit 2a in a comparative manner.

[0068] 6 is a screen that displays information 22 (see FIG. 1) on the accuracy of analysis results of multiple trained models 30 (see FIG. 1) that have the same analysis purpose but different versions. The accuracy evaluation screen 43 includes a graph display area 43a that displays the information 22 on the accuracy of the analysis results in a graph format, and a matrix display area 43b that displays the information 22 on the accuracy of the analysis results in a matrix format.

[0069] As shown in Fig. 6, the control unit 12 (see Fig. 1) is configured to control the display unit 2a (see Fig. 1) to comparatively display information 22 about the accuracy of analysis results of multiple trained models 30 of different versions having the same analysis purpose. Specifically, the control unit 12 is configured to control the display unit 2a to comparatively display information 22 about the accuracy of analysis results based on trained models 30 having the same analysis purpose and analysis target data 20 selected based on a selection operation input received by the input receiving unit 2b (see Fig. 1).

[0070] In the graph 50 displayed in the graph display area 43a, the vertical axis indicates the accuracy index (Accuracy) and the horizontal axis indicates the dataset name. As shown in the legend 51, the graph 50 displays information 22 about the accuracy of the analysis results of a plurality of trained models 30 with different versions having the same analysis purpose. Specifically, as shown in the first legend 51a to the third legend 51c, the graph 50 displays information 22 about the accuracy of the analysis results of trained models 30 with the version names 25 (see FIG. 2) of "1.0.0", "1.0.1", and "2.0.0" among the trained models 30 with the name "Trained Model A". The first legend 51a indicates the trained model 30 with the version name 25 of "1.0.0". The second legend 51b indicates the trained model 30 with the version name 25 of "1.0.1". Moreover, the third legend 51c indicates a trained model 30 whose version name 25 is "2.0.0".

[0071] Moreover, curve 52a shown in graph 50 indicates information 22 relating to the accuracy of the analysis results of trained model 30 whose version name 25 is "1.0.0". Curve 52b indicates information 22 relating to the accuracy of the analysis results of trained model 30 whose version name 25 is "1.0.1". Curve 52c indicates information 22 relating to the accuracy of the analysis results of trained model 30 whose version name 25 is "2.0.0".

[0072] The graph 50 displays information 22 on the accuracy of the analysis result when each version of the trained model 30 is analyzed using each of the datasets A to D. That is, when focusing on one dataset, the graph 50 displays first analysis accuracy information 22a, which is information 22 on the accuracy of the analysis result of a plurality of trained models 30 of different versions having the same analysis purpose. Also, when focusing on one version, the graph 50 displays second analysis accuracy information 22b, which is information 22 on the accuracy of the analysis result when different analysis target data 20 are analyzed. In other words, the control unit 12 is configured to control the display unit 2a to display the first analysis accuracy information 22a and the second analysis accuracy information 22b in a comparable manner.

[0073] Furthermore, the matrix 53 displayed in the matrix display area 47b includes a version name display field 53a, a data set name display field 53b, a parameter display field 53c, an accuracy display field 53d, and an IoU display field 53e. Note that, although the matrix 53 displays information corresponding to the graph 50, in the example shown in FIG. 6, for convenience, only a part of the information is displayed in the matrix 53.

[0074] The version name display field 53a is a display field that displays the version name 25 of the trained model 30. The dataset name display field 53b is a display field that displays the name of the analysis target data 20 analyzed using the trained model 30. The parameter display field 53c is a display field that displays the inference parameters set when analyzing the analysis target data 20. The Accuracy display field 53d is a display field that displays the Accuracy value when the trained model 30 analyzes the analysis target data 20. The IoU display field 53e is a display field that displays the IoU when the trained model 30 analyzes the analysis target data 20. In other words, the Accuracy display field 53d and the IoU display field 53e are display fields that display information 22 regarding the accuracy of the analysis result. In the example shown in FIG. 6, the IoU display field 53e includes a display field that displays four IoUs, IoU_1, IoU_2, IoU_3, and IoU_4. The accuracy is a value indicating the percentage of correct inferences when the trained model 30 analyzes the analysis target data 20. The IoU is a value indicating the degree of match between the inference result and the true result when the trained model 30 analyzes the analysis target data 20.

[0075] In this embodiment, the control unit 12 is configured to be able to display a version history screen 42 and an accuracy evaluation screen 43 that display a version selected by the operator out of the trained model 30 selected by the operator. That is, the control unit 12 is configured to acquire various data associated with each version from the storage unit 11 (see FIG. 1) based on the trained model 30 selected by the operator and the selected version, and to display the version history screen 42 and the accuracy evaluation screen 43 on the display unit 2a.

[0076] (Analysis recipe settings) Next, a configuration for analyzing the analysis target data 20 (see FIG. 1) using the generated trained model 30 (see FIG. 1) will be described with reference to FIGS. 7 and 8.

[0077] The analysis recipe setting screen 44 shown in FIG. 7 includes an input data setting field 44a, a trained model setting field 44b, and an analysis result output destination setting field 44c.

[0078] The input data setting field 44a is a setting field for setting the analysis target data 20 to be analyzed by the trained model 30. The trained model setting field 44b is a setting field for setting the trained model 30 that analyzes the analysis target data 20 set by the input data setting field 44a. The analysis result output destination setting field 44c is a setting field for setting the output destination of the analysis result that is output when the trained model 30 analyzes the analysis target data 20.

[0079] By setting the analysis target data 20, the trained model 30, and the output destination of the analysis results in each setting field, the operator can create an analysis recipe that specifies what analysis target data 20 is to be analyzed using what trained model 30, and where the analysis results are to be output. The operator can also analyze the analysis target data 20 based on the created analysis recipe.

[0080] In addition, a setting button 44d is displayed in the trained model setting field 44b. The setting button 44d is a push button of the GUI. When the setting button 44d is pressed, the control unit 12 displays a trained model selection screen 45 shown in FIG.

[0081] 8 includes a trained model selection field 45a and an update button 45b. The trained model selection field 45a is a selection field that displays options 45c of the trained models 30. The trained model selection field 45a is, for example, a combo box. The update button 45b is a GUI push button displayed on the trained model selection screen 45.

[0082] The control unit 12 (see Figure 1) displays, as options 45c, the trained model 30 (see Figure 1) selected by the operator and a trained model 30 of a different version that has the same analysis purpose as the trained model 30 selected by the operator.

[0083] When the update button 45b is pressed while any of the options 45c is selected, the control unit 12 sets the selected version of the trained model 30 in the trained model setting field 44b (see Figure 7) of the analysis recipe setting screen 44 (see Figure 7) and closes the trained model selection screen 45.

[0084] (Display process of trained model) Next, with reference to FIG. 9, a process in which the data analysis system 100 (see FIG. 1) according to this embodiment displays a plurality of trained models 30 (see FIG. 1) on the display unit 2a (see FIG. 1) will be described.

[0085] In step 101, the trained model generation unit 12b (see FIG. 1) generates a trained model 30 using at least the teacher data 21 (see FIG. 1). In this embodiment, the trained model generation unit 12b generates the trained model 30 using the teacher data 21, the learning parameters 23, and either the trained model 30 or the algorithm 24 of the learning model.

[0086] In step 102, the analysis unit 12a (see FIG. 1) uses the generated trained model 30 to analyze the analysis target data 20 (see FIG. 1) and obtains information 22 (see FIG. 1) regarding the accuracy of the analysis result.

[0087] In step 103, the control unit 12 (see FIG. 1) associates each of the multiple trained models 30 generated by the trained model generation unit 12b with the teacher data 21 used when generating the models, and stores the multiple trained models 30 of different versions having the same analysis purpose in association with each other. Also, in this embodiment, in step 103, the control unit 12 associates information 22 (see FIG. 1) regarding the accuracy of the analysis result of the analysis target data 20 with each of the multiple trained models 30, together with the teacher data 21, and stores the multiple trained models 30 of different versions having the same analysis purpose in association with each other.

[0088] In step 104, the control unit 12 displays on the same screen the selected trained model 30 from among the multiple stored trained models 30, and trained models 30 of different versions associated with the selected trained model 30. In this embodiment, the control unit 12 displays on the same screen trained models 30 of different versions having the same analysis objective associated with the selected trained model 30. Then, the processing ends.

[0089] In addition, in step 103, if the control unit 12 does not store information 22 regarding the accuracy of the analysis results in association with the trained model 30 together with the teacher data 21 in the memory unit 11, the processing of step 102 can be skipped.

[0090] (Determining the version name of a trained model) Next, with reference to FIG. 11, a process in which the control unit 12 (see FIG. 1) determines the version name 25 (see FIG. 2) when generating the trained model 30 (see FIG. 1) will be described.

[0091] In step 110, the control unit 12 acquires the version name 25 input in the version name input field 40f (see FIG. 3) of the trained model generation screen 40 (see FIG. 3).

[0092] In step 111, the control unit 12 acquires the version name 25 of the trained model 30 selected on the trained model display screen 41 (see Figure 4) and the version names 25 of multiple trained models 30 having the same analysis purpose but different versions.

[0093] In step 112, the control unit 12 determines whether the version name 25 entered in the version name input field 40f is the latest. Specifically, the control unit 12 compares the major version, minor version, and patch version of the version name 25 entered in the version name input field 40f with the major version, minor version, and patch version of the version name 25 of each of the multiple trained models 30 acquired in step 111. If the control unit 12 determines that the version name 25 entered in the version name input field 40f is the latest, the process proceeds to step 113. If the control unit 12 determines that the version name 25 entered in the version name input field 40f is not the latest, the process proceeds to step 114.

[0094] In step 113, the trained model generation unit 12b (see FIG. 1) uses the version name 25 input in the version name input field 40f to generate the trained model 30. Then, the process ends.

[0095] Furthermore, when the process proceeds from step 112 to step 114, in step 114, the control unit 12 displays that the version name 25 entered in the version name input field 40f is not the latest. The control unit 12 displays, for example, in a pop-up format on the trained model generation screen 40, that the version name 25 entered in the version name input field 40f is not the latest. The control unit 12 also performs control not to cause the trained model generation unit 12b to generate the trained model 30. For example, the control unit 12 performs control to make it impossible to press the generate button 40i (see FIG. 3) displayed on the trained model generation screen 40. The process then ends.

[0096] (Effects of this embodiment) In this embodiment, the following effects can be obtained.

[0097] In this embodiment, as described above, the data analysis system 100 is a data analysis system that analyzes analysis target data 20, which is data to be analyzed, using a trained model 30, and includes a data acquisition unit 10 that acquires the analysis target data 20, an analysis unit 12a that analyzes the analysis target data 20 using the trained model 30, a trained model generation unit 12b that generates the trained model 30 using at least teacher data 21, a memory unit 11 that associates each of the multiple trained models 30 generated by the trained model generation unit 12b with the teacher data 21 used when generating them, and associates and stores multiple trained models 30 that have the same analysis purpose but different versions, a display unit 2a that displays the multiple trained models 30 stored in the memory unit 11, and a control unit 12 that controls the display of a selected trained model 30 from the multiple trained models 30 displayed on the display unit 2a and a trained model 30 of a different version associated with the selected trained model 30 on the same screen.

[0098] As a result, the selected trained model 30 and trained models 30 of different versions associated with the selected trained model 30 are displayed on the same screen, making it possible to easily understand the multiple trained models 30 of different versions at a glance. As a result, it is possible to easily understand the multiple trained models 30 of different versions at a glance. This allows the operator to efficiently select the trained model 30 to be used in analyzing the analysis target data 20 from the multiple trained models 30 of different versions when analyzing the analysis target data 20.

[0099] Furthermore, in this embodiment, as described above, the data analysis method is a data analysis method for analyzing analysis target data 20, which is data to be analyzed, using a trained model 30, and includes the steps of generating a trained model 30 using at least teacher data 21, associating each of the multiple trained models 30 generated by the trained model generation unit 12b with the teacher data 21 used when generating the models, and associating and storing multiple trained models 30 having the same analysis purpose but different versions, and displaying a selected trained model 30 from the multiple stored trained models 30 and a trained model 30 of a different version associated with the selected trained model 30 on the same screen.

[0100] This allows the operator to easily grasp at a glance a plurality of trained models 30 of different versions, similarly to the above-described data analysis system 100, and thus provides a data analysis method that allows the operator to easily grasp at a glance a plurality of trained models 30 of different versions. As a result, it is possible to provide a data analysis method that allows the operator to efficiently select a trained model 30 to be used in analyzing the analysis target data 20 from among a plurality of trained models 30 of different versions when analyzing the analysis target data 20.

[0101] In addition, in the above embodiment, the following additional effects can be obtained by configuring as follows.

[0102] That is, in this embodiment, as described above, the storage unit 11 is configured to associate information 22 on the accuracy of the analysis result of the analysis target data 20 with each of the multiple trained models 30 together with the teacher data 21, and to store multiple trained models 30 of different versions having the same analysis purpose in association with each other. This makes it possible to present to the operator the information 22 on the accuracy of the analysis result of the analysis target data 20 in the multiple trained models 30 having the same analysis purpose in a comparable manner. As a result, it becomes possible to present to the operator the information 22 on the accuracy of the analysis result of each trained model 30 in a comparable manner, so that the operator can more effectively select a trained model 30 suitable for analyzing the analysis target data 20 from among the multiple trained models 30 having the same analysis purpose.

[0103] Furthermore, in this embodiment, as described above, the control unit 12 is configured to control the display unit 2a to display information 22 on the accuracy of the analysis results of multiple trained models 30 of different versions having the same analysis purpose in a comparative manner. By comparing the information 22 on the accuracy of the analysis results, the operator can easily distinguish at a glance the trained model 30 suitable for analyzing the analysis target data 20 from among the multiple trained models 30 having the same analysis purpose. As a result, the operator can easily select the trained model 30 with high analysis accuracy when analyzing the analysis target data 20.

[0104] In addition, in this embodiment, as described above, the input receiving unit 2b that receives the selection operation input of the operator is further provided, and the control unit 12 is configured to control the display unit 2a to display the trained model 30 having the same analysis purpose selected based on the selection operation input received by the input receiving unit 2b and the information 22 regarding the accuracy of the analysis result based on the analysis target data 20 in a comparable manner. This allows the operator to easily distinguish the trained model 30 suitable for analyzing the analysis target data 20 from among multiple trained models 30 having the same analysis purpose as the selected trained model 30 but different versions. Therefore, the operator can easily select the trained model 30 suitable for analyzing the analysis target data 20. As a result, the work efficiency when selecting the trained model 30 suitable for analyzing the analysis target data 20 can be further improved.

[0105] In addition, in this embodiment, as described above, the storage unit 11 is configured to store the first analysis accuracy information 22a, which is information 22 on the accuracy of the analysis results of multiple trained models 30 of different versions having the same analysis purpose, in association with the second analysis accuracy information 22b, which is information 22 on the accuracy of the analysis results when the trained models 30 of the same version are made to analyze different analysis target data 20, and the control unit 12 is configured to control the display unit 2a to display the first analysis accuracy information 22a and the second analysis accuracy information 22b in a comparable manner. As a result, by comparing the first analysis accuracy information 22a output from multiple trained models 30 of different versions having the same analysis purpose, the operator can easily determine the trained model 30 suitable for analyzing the analysis target data 20. In addition, by comparing the second analysis accuracy information 22b, the operator can easily determine whether the analysis accuracy of the trained model 30 has decreased. Therefore, the operator can easily determine whether the trained model 30 needs to be improved in order to improve the accuracy of the trained model 30. As a result, it becomes possible to easily determine a trained model 30 suitable for analyzing the data 20 to be analyzed, and whether or not the trained model 30 needs to be opened, thereby improving convenience (usability) for the operator.

[0106] Furthermore, in this embodiment, as described above, the trained model generation unit 12b is configured to be able to generate trained models 30 of different versions having the same analysis purpose based on a selected trained model 30 from among the multiple trained models 30 stored in the storage unit 11. This allows trained models 30 of different versions having the same analysis purpose to be generated based on the selected trained model 30, so that the trained models 30 having the same analysis purpose and already generated can be easily improved. As a result, it becomes possible to easily improve the trained models 30 already generated, so that it is possible to easily generate trained models 30 with improved analysis accuracy of the analysis target data 20.

[0107] In addition, in this embodiment, as described above, when the trained model 30 having the same analysis purpose is generated in a state where the trained model 30 is selected, the trained model generation unit 12b is configured to generate trained models 30 having different versions with the same analysis purpose by performing training in a state where at least one of the teacher data 21 or the learning parameters 23 set when generating the trained model 30 is changed. As a result, trained models 30 having the same analysis purpose but different versions can be easily generated by changing at least one of the teacher data 21 and the learning parameters 23. Therefore, when improving the trained model 30 to improve the analysis accuracy of the analysis target data 20 by the trained model 30, multiple trained models 30 having the same analysis purpose but different versions can be easily generated. As a result, the trained model 30 can be efficiently improved to improve the analysis accuracy of the analysis target data 20 by the trained model 30.

[0108] In addition, in this embodiment, as described above, the control unit 12 is configured to control the display unit 2a to display multiple versions of the trained model 30 stored in the storage unit 11 in a selectable manner when generating trained models 30 of different versions with the same analysis purpose. This allows the operator to easily grasp trained models 30 of different versions with the same analysis purpose when selecting trained models 30 of different versions that have already been generated and are the same analysis purpose, and generating a new version of the trained model 30 by improving the trained model 30. As a result, the convenience (usability) for the operator can be improved.

[0109] In addition, in this embodiment, as described above, the control unit 12 is configured to perform control to sequentially add the latest version of the trained model 30 as an option when trained models 30 of different versions with the same analysis purpose are generated. Here, when trained models 30 of different versions with the same analysis purpose are generated, the accuracy of the analysis of the analysis target data 20 is not necessarily improved. Therefore, for example, when trained models 30 of different versions with the same analysis purpose are generated, in a configuration in which the latest version of the trained model 30 is automatically set to analyze the analysis target data 20, the analysis accuracy of the analysis target data 20 may decrease. Therefore, as described above, by configuring to perform control to sequentially add the latest version of the trained model 30 as an option, it is possible to suppress the latest version of the trained model 30 from being automatically set as the trained model 30 that analyzes the analysis target data 20. As a result, it is possible to suppress the analysis accuracy of the analysis target data 20 from decreasing due to the latest version of the trained model 30 being automatically set as the trained model 30 that analyzes the analysis target data 20.

[0110] In addition, as described above, this embodiment further includes an input receiving unit 2b that receives input from an operator, and the control unit 12 is configured to determine whether the version name 25 input by the input receiving unit 2b is the latest version name 25 based on the naming rule, and to control the trained model generation unit 12b not to generate the trained model 30 if the version name 25 is not the latest version name 25 based on the naming rule. This makes it possible to prevent the generation of a trained model 30 of a different system of version derived from a past version name 25, where the version name 25 of the trained model 30 has the same analysis purpose but is a different version. As a result, it is possible to prevent the management of the version name 25 of the trained model 30 having the same analysis purpose from becoming complicated.

[0111] In addition, in this embodiment, as described above, the storage unit 11 is configured to store the trained models 30 generated in a state in which the algorithm 24 of the learning model is the same and at least one of the teacher data 21 or the learning parameters 23 set when generating the trained model 30 is changed, as trained models 30 of different versions in association with each other. Here, when the version of the algorithm 24 of the learning model is updated, the analysis operation in the data analysis system 100 may not be guaranteed. If the trained models 30 that cannot guarantee the analysis operation are stored in association with each other, the maintenance work of the data analysis system 100 becomes complicated. Therefore, by configuring as described above, the trained models 30 having different algorithms 24 of the learning model can be version-managed in different systems. Therefore, it is possible to prevent the trained models 30 that may not guarantee the analysis operation from being included in the version management. As a result, it is possible to suppress the maintenance work of the data analysis system 100 from becoming complicated.

[0112] In addition, in the present embodiment, in the step of associating and storing a plurality of trained models 30 of different versions as described above, information 22 on the accuracy of the analysis result of the analysis target data 20 is associated with each of the plurality of trained models 30 together with the teacher data 21, and a plurality of trained models 30 of different versions having the same analysis purpose are associated and stored. This makes it possible to present to the operator the information 22 on the accuracy of the analysis result of the analysis target data 20 in the plurality of trained models 30 having the same analysis purpose in a comparable manner. As a result, it becomes possible for the operator to present the information 22 on the accuracy of the analysis result of each trained model 30 in a comparable manner, and therefore it is possible to provide a data analysis method that allows the operator to easily select a trained model 30 suitable for analyzing the analysis target data 20 from among a plurality of trained models 30 having the same analysis purpose.

[0113] [Variations] The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is indicated by the claims rather than the description of the above embodiments, and further includes all modifications (variations) within the meaning and scope of the claims.

[0114] For example, in the above embodiment, the data analysis system 100 is configured as a so-called on-premise system, but the present invention is not limited to this. For example, as in a first modified example shown in Fig. 11, the data analysis system 200 may be configured as a so-called cloud service type system.

[0115] In the first modified example shown in Fig. 11, the server 1 and the analysis data display device 2 are connected via a network 91. The network 91 may be, for example, the Internet. In the first modified example, since the network 91 is the Internet, the data analysis system 200 is configured as a system constructed in the form of so-called cloud computing. Even with this configuration, it is possible to provide a data analysis system 200 that allows the user to easily grasp multiple trained models 30 of different versions at a glance, similar to the data analysis system 100 according to the above embodiment.

[0116] In the above embodiment, the data analysis system 100 is configured such that the server 1 and the analysis data display device 2 are different devices, but the present invention is not limited to this. For example, they may be configured in one information processing device, as in the data analysis device 300 shown in FIG.

[0117] As shown in FIG. 12, the data analysis device 300 includes a data acquisition unit 310, a storage unit 311, a control unit 312, a display unit 313, and an input reception unit 314.

[0118] The data acquisition unit 310, memory unit 311, and control unit 312 of the data analysis device 300 have the same configuration as the data acquisition unit 10, memory unit 11, and control unit 12 of the data analysis system 100 of the above embodiment, so detailed explanations are omitted.

[0119] The control unit 312 also includes an analysis unit 312a and a trained model generation unit 312b. The analysis unit 312a and the trained model generation unit 312b have the same configurations as the analysis unit 12a and the trained model generation unit 12b according to the above embodiment, and therefore detailed description thereof will be omitted.

[0120] Furthermore, the display unit 313 and the input receiving unit 314 provided in the data analysis device 300 according to the second modified example have the same configuration as the display unit 2a and the input receiving unit 2b provided in the data analysis system 100 according to the above embodiment, so detailed explanations are omitted.

[0121] As described above, the data analysis device 300 according to the second modified example is a data analysis device that analyzes the analysis target data 20, which is data to be analyzed, using the trained model 30, and includes a data acquisition unit 310 that acquires the analysis target data 20, an analysis unit 312a that analyzes the analysis target data 20 using the trained model 30, a trained model generation unit 312b that generates the trained model 30 using at least the teacher data 21, a memory unit 311 that associates each of the multiple trained models 30 generated by the trained model generation unit 312b with the teacher data 21 used when generating them, and associates and stores multiple trained models 30 that have the same analysis purpose but different versions, a display unit 313 that displays the multiple trained models 30 stored in the memory unit 311, and a control unit 312 that controls the display of a selected trained model 30 from the multiple trained models 30 displayed on the display unit 313 and a trained model 30 of a different version associated with the selected trained model 30 on the same screen.

[0122] This allows the operator to easily grasp at a glance a plurality of trained models 30 of different versions, similarly to the above-described data analysis system 100, and thus makes it possible to provide a data analysis device 300 that allows the operator to easily grasp at a glance a plurality of trained models 30 of different versions. As a result, it is possible to provide a data analysis device 300 that allows the operator to efficiently select a trained model 30 to be used in analyzing the analysis target data 20 from among a plurality of trained models 30 of different versions when analyzing the analysis target data 20.

[0123] In the above embodiment, the control unit 12 controls the display of a plurality of trained models 30 of different versions having the same analysis purpose associated with the selected trained model 30 on the same screen, but the present invention is not limited to this. For example, the control unit may be configured to display a plurality of trained models 30 having different versions from the selected trained model 30 and not having the same analysis purpose on the same screen.

[0124] In the above embodiment, the storage unit 11 stores the information 22 on the accuracy of the analysis result of the analysis target data 20 in association with the teacher data 21 for each of the trained models 30, but the present invention is not limited to this. For example, if the storage unit 11 stores a plurality of trained models 30 having different versions but the same analysis purpose, the information 22 on the accuracy of the analysis result of the analysis target data 20 may not be stored in association with each of the trained models 30. However, if the storage unit 11 does not store the information 22 on the accuracy of the analysis result of the analysis target data 20 in association with each of the trained models 30, it becomes difficult to present the information 22 on the accuracy of the analysis result of the trained models 30 having different versions but the same analysis purpose to the operator in a comparable manner. Therefore, it is preferable that the storage unit 11 is configured to store the information 22 on the accuracy of the analysis result of the analysis target data 20 in association with each of the trained models 30, together with the teacher data 21.

[0125] In the above embodiment, the control unit 12 has been configured to display information 22 about the accuracy of the analysis results of multiple trained models 30 with different versions having the same analysis purpose on the display unit 2a in a comparative manner, but the present invention is not limited to this. The control unit 12 does not need to display information 22 about the accuracy of the analysis results of multiple trained models 30 with different versions having the same analysis purpose on the display unit 2a in a comparative manner, as long as the selected trained model 30 and the trained models 30 with different versions having the same analysis purpose are displayed on the same screen. However, if the control unit 12 is configured not to display information 22 about the accuracy of the analysis results of multiple trained models 30 with different versions having the same analysis purpose on the display unit 2a in a comparative manner, it becomes difficult for the operator to easily compare information 22 about the accuracy of the analysis results of multiple trained models 30 with different versions having the same analysis purpose. Therefore, it is preferable that the control unit 12 is configured to display information 22 about the accuracy of the analysis results of multiple trained models 30 with different versions having the same analysis purpose on the display unit 2a in a comparative manner.

[0126] In the above embodiment, an example of a configuration in which the storage unit 11 stores the first analysis accuracy information 22a and the second analysis accuracy information 22b as information 22 on the accuracy of analysis results of multiple trained models 30 of different versions having the same analysis purpose in association with the trained model 30 has been shown, but the present invention is not limited to this. For example, the storage unit 11 may be configured to store at least one of the first analysis accuracy information 22a and the second analysis accuracy information 22b in association with the trained model 30.

[0127] In the above embodiment, the control unit 12 controls the display unit 2a to display the first analysis accuracy information 22a and the second analysis accuracy information 22b in a comparative manner, but the present invention is not limited to this. For example, the control unit 12 may be configured to control the display unit 2a to display either the first analysis accuracy information 22a or the second analysis accuracy information 22b in a comparative manner.

[0128] In the above embodiment, the trained model generating unit 12b generates trained models 30 of different versions having the same analysis purpose based on a selected trained model 30 from among the multiple trained models 30 stored in the storage unit 11, but the present invention is not limited to this. For example, the trained model generating unit 12b may be configured to generate a trained model 30 based on the algorithm 24 of a training model having the same analysis purpose, in which at least one of the teacher data 21 and the learning parameters 23 is different, instead of the selected trained model 30.

[0129] In the above embodiment, the control unit 12 controls the display unit 2a to selectably display the multiple versions of the trained model 30 stored in the storage unit 11 when generating trained models 30 with different versions having the same analysis purpose, but the present invention is not limited to this. For example, the control unit 12 does not need to selectably display the multiple versions of the trained model 30 stored in the storage unit 11 on the display unit 2a when generating trained models 30 with different versions having the same analysis purpose. However, if the control unit 12 is configured not to selectably display the multiple versions of the trained model 30 stored in the storage unit 11 on the display unit 2a when generating trained models 30 with different versions having the same analysis purpose, it becomes difficult for the operator to grasp the trained models 30 with different versions having the same analysis purpose. Therefore, it is preferable that the control unit 12 is configured to control the display unit 2a to selectably display the multiple versions of the trained model 30 stored in the storage unit 11 when generating trained models 30 with different versions having the same analysis purpose.

[0130] In the above embodiment, the control unit 12 is configured to perform control to sequentially add the latest version of the trained model 30 as an option when different versions with the same analysis purpose are generated, but the present invention is not limited to this. For example, the control unit 12 does not need to add the latest version of the trained model 30 as an option.

[0131] In the above embodiment, the control unit 12 determines whether the version name 25 input by the input receiving unit 2b is the latest version name 25 based on the naming rule, and if the version name 25 is not the latest version name 25 based on the naming rule, controls the trained model generation unit 12b not to generate the trained model 30. However, the present invention is not limited to this. For example, the control unit 12 may be configured to generate the trained model 30 by the trained model generation unit 12b even if the version name 25 input by the input receiving unit 2b is not the latest version name 25 based on the naming rule. However, if the control unit 12 controls the trained model generation unit 12b to generate the trained model 30 even if the version name 25 input by the input receiving unit 2b is not the latest version name 25 based on the naming rule, management of the version name 25 of the generated trained model 30 becomes complicated. Therefore, it is preferable that the control unit 12 is configured to control the trained model generation unit 12b not to generate the trained model 30 if the version name 25 input by the input receiving unit 2b is not the latest version name 25 based on the naming rules.

[0132] In the above embodiment, the control unit 12 controls the learned model 30 generated in a state where the algorithm 24 of the learning model is the same and at least one of the learning parameters 23 set when generating the trained model 30 is changed, to be stored in association with each other as a trained model 30 of a different version, but the present invention is not limited to this. For example, the control unit 12 may be configured to control the learned model 30 generated in a state where the learning parameters 23 set when generating the trained model 30 are the same and the version of the algorithm 24 of the learning model is changed, to be stored in association with each other as a trained model 30 of a different version. However, when the version of the algorithm 24 of the learning model is updated, the generated trained model 30 may not operate normally. Therefore, it is preferable that the control unit 12 is configured to control the learned model 30 generated in a state where the algorithm 24 of the learning model is the same and at least one of the learning parameters 23 set when generating the trained model 30 is changed, to be stored in association with each other as a trained model 30 of a different version.

[0133] [Aspects] It will be appreciated by those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0134] (Item 1) A data analysis system that analyzes analysis target data, which is data to be analyzed, using a trained model, A data acquisition unit that acquires the analysis target data; an analysis unit that analyzes the analysis target data using the trained model; A trained model generation unit that generates the trained model using at least teacher data; A storage unit that associates the plurality of trained models generated by the trained model generation unit with the teacher data used in generating each of the trained models, and associates and stores the plurality of trained models having the same analysis purpose and different versions; A display unit that displays the plurality of trained models stored in the storage unit; and A data analysis system comprising: a control unit that controls displaying, on the same screen, a selected trained model from the plurality of trained models displayed on the display unit and a trained model having a different version associated with the selected trained model.

[0135] (Item 2) 2. The data analysis system according to item 1, wherein the storage unit is configured to associate, with each of the plurality of trained models, information regarding accuracy of the analysis result of the analysis target data together with the teacher data, and to store the plurality of trained models of different versions having the same analysis purpose in association with each other.

[0136] (Item 3) 3. The data analysis system according to item 2, wherein the control unit is configured to control the display unit to display information regarding accuracy of the analysis results of the plurality of trained models of different versions having the same analysis purpose in a comparative manner.

[0137] (Item 4) An input receiving unit that receives an operation input of a selection by an operator, 4. The data analysis system according to item 2 or 3, wherein the control unit is configured to control the display unit to display information on the accuracy of the analysis result based on the trained model and the analysis target data having the same analysis purpose, selected based on the selection operation input received by the input receiving unit, in a comparative manner.

[0138] (Item 5) The storage unit is configured to store, in association with each other, first analysis accuracy information, which is information about accuracy of the analysis results of the plurality of trained models having different versions and having the same analysis purpose, and second analysis accuracy information, which is information about accuracy of the analysis results when the trained models having the same version are made to analyze the different analysis target data, 5. The data analysis system according to item 3 or 4, wherein the control unit is configured to control the display unit to display the first analysis accuracy information and the second analysis accuracy information in a comparative manner.

[0139] (Item 6) The data analysis system according to any one of items 2 to 4, wherein the trained model generation unit is configured to be capable of generating trained models of different versions having the same analysis purpose based on a selected trained model from the plurality of trained models stored in the storage unit.

[0140] (Item 7) Item 7. The data analysis system of item 6, wherein the trained model generation unit is configured to generate a different version of the trained model having the same analysis purpose by training in a state in which at least one of the teacher data or the learning parameters set when generating the trained model is changed when the trained model having the same analysis purpose is generated in a state in which the trained model is selected and the trained model having the same analysis purpose is generated.

[0141] (Item 8) 8. The data analysis system according to item 7, wherein the control unit is configured to control the display unit to selectably display multiple versions of the trained models stored in the memory unit when generating different versions of the trained models having the same analysis purpose.

[0142] (Item 9) The data analysis system according to any one of items 1 to 8, wherein the control unit is configured to perform control such that, when different versions of the trained models having the same analysis purpose are generated, the latest version of the trained model is sequentially added as an option.

[0143] (Item 10) further comprising an input receiving unit for receiving an input from an operator; The data analysis system according to any one of items 1 to 3, wherein the control unit is configured to determine whether the version name input by the input receiving unit is the latest version name based on a naming rule, and if the version name is not the latest version name based on a naming rule, to control the trained model generation unit not to generate the trained model.

[0144] (Item 11) The data analysis system according to any one of items 1 to 10, wherein the memory unit is configured to store the trained models generated in a state in which the algorithm of the learning model is the same and at least one of the teacher data or the learning parameters set when generating the trained model is changed, as trained models of different versions in association with each other.

[0145] (Item 12) A data analysis device that analyzes analysis target data, which is data to be analyzed, using a trained model, A data acquisition unit that acquires the analysis target data; an analysis unit that analyzes the analysis target data using the trained model; A trained model generation unit that generates the trained model using at least teacher data; A storage unit that associates the plurality of trained models generated by the trained model generation unit with the teacher data used in generating each of the trained models, and associates and stores the plurality of trained models having the same analysis purpose and different versions; A display unit that displays the plurality of trained models stored in the storage unit; and a control unit that performs control to display, on the same screen, a selected trained model from the plurality of trained models displayed on the display unit and a trained model of a different version associated with the selected trained model.

[0146] (Item 13) A data analysis method for analyzing analysis target data, which is data to be analyzed, using a trained model, comprising: generating the trained model using at least training data; Associating the plurality of trained models generated by the trained model generation unit with the teacher data used in generating each of the trained models, and associating and storing the plurality of trained models having the same analysis purpose and different versions; A data analysis method comprising a step of displaying, on the same screen, a selected trained model from the plurality of stored trained models and a trained model having a different version associated with the selected trained model.

[0147] (Item 14) Item 14. The data analysis method according to item 13, wherein in the step of associating and storing the plurality of trained models of different versions, information on accuracy of an analysis result of the analysis target data is associated with each of the plurality of trained models together with the teacher data, and the plurality of trained models of different versions having the same analysis purpose are associated and stored. [Explanation of symbols]

[0148] 2a, 313 display section 2b, 314 Input reception section 10, 310 Data Acquisition Section 11, 311 Storage section 12, 312 Control section 12a, 312a analysis section 12b, 312b Trained model generation unit 20 Data to be analyzed 21 Teacher Data 22 Information regarding the accuracy of the analysis results 22a First analysis accuracy information 22b Second analysis accuracy information 23 Learning parameters 24 Learning model algorithm 25 Version Name 30, 30a, 30b, 30c Pre-trained models 100, 200 Data Analysis System 300 Data Analysis Equipment

Claims

1. A data analysis system that analyzes analysis target data, which is data to be analyzed, using a trained model, A data acquisition unit that acquires the analysis target data; an analysis unit that analyzes the analysis target data using the trained model; A trained model generation unit that generates the trained model using at least teacher data; A storage unit that associates the plurality of trained models generated by the trained model generation unit with the teacher data used in generating each of the trained models, and associates and stores the plurality of trained models having the same analysis purpose and different versions; A display unit that displays the plurality of trained models stored in the storage unit; and A data analysis system comprising: a control unit that controls displaying, on the same screen, a selected trained model from the plurality of trained models displayed on the display unit and a trained model having a different version associated with the selected trained model.

2. 2. The data analysis system according to claim 1, wherein the memory unit is configured to associate information regarding the accuracy of the analysis results of the analysis target data together with the teacher data for each of the plurality of trained models, and to associate and store the plurality of trained models of different versions having the same analysis purpose.

3. 3. The data analysis system according to claim 2, wherein the control unit is configured to control the display unit to display information regarding accuracy of the analysis results of the plurality of trained models of different versions having the same analysis purpose in a comparative manner.

4. An input receiving unit that receives an operation input of a selection by an operator, 4. The data analysis system according to claim 3, wherein the control unit is configured to control the display unit to display information regarding accuracy of the analysis result based on the trained model and the analysis target data having the same analysis purpose, selected based on the selection operation input received by the input receiving unit, in a comparative manner.

5. The storage unit is configured to store, in association with each other, first analysis accuracy information, which is information about accuracy of the analysis results of the plurality of trained models having different versions and having the same analysis purpose, and second analysis accuracy information, which is information about accuracy of the analysis results when the trained models having the same version are made to analyze the different analysis target data, The data analysis system according to claim 3 , wherein the control unit is configured to perform control so as to cause the display unit to display the first analysis accuracy information and the second analysis accuracy information in a comparable manner.

6. 3. The data analysis system according to claim 2, wherein the trained model generation unit is configured to be capable of generating trained models of different versions having the same analysis purpose based on a selected trained model from the plurality of trained models stored in the memory unit.

7. The data analysis system of claim 6, wherein the trained model generation unit is configured to generate a different version of the trained model having the same analysis purpose by training in a state in which at least one of the teacher data or the learning parameters set when generating the trained model is changed when the trained model is generated in a state in which the trained model having the same analysis purpose is selected.

8. 8. The data analysis system according to claim 7, wherein the control unit is configured to control the display unit to selectably display multiple versions of the trained models stored in the memory unit when generating different versions of the trained models having the same analysis purpose.

9. 2. The data analysis system of claim 1, wherein the control unit is configured to perform control such that, when different versions of the trained models having the same analysis purpose are generated, the latest version of the trained model is sequentially added as an option.

10. further comprising an input receiving unit for receiving an input from an operator; 2. The data analysis system of claim 1, wherein the control unit is configured to determine whether the version name input by the input receiving unit is the latest version name based on a naming rule, and if the version name is not the latest version name based on a naming rule, to control the trained model generation unit not to generate the trained model.

11. 2. The data analysis system of claim 1, wherein the memory unit is configured to store, in association with each other, a trained model generated when the learning model has the same algorithm and at least one of the teacher data or the learning parameters set when generating the trained model has been changed, as a trained model of a different version.

12. A data analysis device that analyzes analysis target data, which is data to be analyzed, using a trained model, A data acquisition unit that acquires the analysis target data; an analysis unit that analyzes the analysis target data using the trained model; A trained model generation unit that generates the trained model using at least teacher data; A storage unit that associates the plurality of trained models generated by the trained model generation unit with the teacher data used in generating each of the trained models, and associates and stores the plurality of trained models having the same analysis purpose and different versions; A display unit that displays the plurality of trained models stored in the storage unit; and a control unit that controls displaying, on the same screen, a selected trained model from the plurality of trained models displayed on the display unit and a trained model of a different version associated with the selected trained model.

13. A data analysis method for analyzing analysis target data, which is data to be analyzed, using a trained model, comprising: generating the trained model using at least training data; Associating the plurality of trained models generated by the trained model generation unit with the teacher data used in generating each of the trained models, and associating and storing the plurality of trained models having the same analysis purpose and different versions; A data analysis method comprising a step of displaying, on the same screen, a selected trained model from the plurality of stored trained models and a trained model having a different version associated with the selected trained model.

14. 14. The data analysis method according to claim 13, wherein in the step of associating and storing the plurality of trained models of different versions, information regarding accuracy of an analysis result of the data to be analyzed is associated with each of the plurality of trained models, together with the teacher data, and the plurality of trained models of different versions having the same analysis purpose are associated and stored.

Citation Information

Patent Citations

  • Learning support system and learning support method

    JP2022174740A