Information processing apparatus and control method therefor

Abstract

To identify a user who made contributions to a learning model.SOLUTION: An information processing apparatus for managing a plurality of learning models comprises: model management means that manages, for each of the plurality of learning models, first information for identifying a dataset used to train the learning model, and second information for identifying an initial model used to train the learning model; data management means that manages, for each of a plurality of datasets identified by the first information of each of the plurality of learning models, third information for identifying an initial dataset used to create the datasets; and determination means that determines, based on the first information, the second information, and the third information, at least one learning model and at least one dataset used to train a learning model of interest included in the plurality of learning models.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a management technique for a learning model and learning data. [Background technology]

[0002] In recent years, the accuracy of image recognition technologies such as image classification, object detection, and object tracking has improved dramatically with the advent of deep neural networks (DNNs). Generally, training a DNN model requires a large amount of data, but it is difficult for an individual to collect such a large amount of data. Therefore, a technique called fine-tuning is sometimes used, in which a model trained with a large amount of data is used as an initial model, and only some of the layers of the DNN model are trained. It is also expected that in the future, users will use training models published by other users as initial models, perform additional training to suit their own purposes, and then publish the resulting training models.

[0003] Patent Document 1 discloses a technology that makes it possible to identify users who have participated in creating a learning model by associating identification information that identifies the users (contributors) with the learning model. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6858082 Summary of the Invention [Problem to be solved by the invention]

[0005] In the training of a learning model, not only the initial model but also the data used contributes significantly. Therefore, when releasing a learning model created by additional training for a fee, it is desirable to set a fee that takes into account the contribution of the user who created (edited or added) the data used in the additional training.

[0006] However, in the method disclosed in Patent Document 1, although it is possible to identify users who contributed to the model itself, it is not possible to identify users who contributed to the data used (data used to train the initial model, data used for additional training). Therefore, there is a problem in that it is difficult to appropriately set compensation for each user for publishing the trained model.

[0007] The present invention has been made in consideration of such problems, and aims to provide a technology that makes it possible to manage data used to create a learning model. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems, an information processing device according to the present invention has the following configuration. That is, the information processing device that manages a plurality of learning models includes: A model management means for managing, for each of the plurality of learning models, first information that identifies a dataset used in learning the learning model and second information that identifies an initial model used in learning the learning model; A data management means for managing third information that identifies an initial dataset used to create a dataset for each of a plurality of datasets identified by the first information for each of the plurality of learning models; A determination means for determining one or more learning models and one or more datasets used in learning of the attention learning model included in the plurality of learning models based on the first information, the second information, and the third information; Equipped with. [Effects of the Invention]

[0009] According to the present invention, it is possible to provide a technology that makes it possible to identify the data used to create a learning model. [Brief explanation of the drawings]

[0010] [Figure 1]FIG. 2 is a block diagram showing a functional configuration of the information processing device. [Figure 2] FIG. 2 is a diagram showing a processing flow in an information processing device. [Figure 3] 10 is a flowchart of a learning process in the information processing device. [Figure 4] FIG. 10 is a diagram illustrating an example of information assigned to a model. [Figure 5] FIG. 10 is a diagram illustrating an example of information added to data. [Figure 6] FIG. 10 is a diagram illustrating an example of traceability information. [Figure 7] FIG. 10 is a diagram illustrating an example of learning data. [Figure 8] FIG. 10 is a diagram illustrating an example of editing learning data. [Figure 9] FIG. 10 is a block diagram showing the functional configuration of an information processing device (second embodiment). [Figure 10] 10 is a flowchart of a learning process in an information processing device (second embodiment). [Figure 11] FIG. 10 is a diagram illustrating an example of information assigned to a model (second embodiment). [Figure 12] FIG. 10 is a diagram illustrating an example of traceability information (second embodiment). [Figure 13] FIG. 10 is a diagram illustrating an example of learning data (modification). [Figure 14] FIG. 10 is a diagram showing an example of information added to data (modification). DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention claimed. Although multiple features are described in the embodiments, not all of these multiple features are necessarily essential to the invention, and multiple features may be combined arbitrarily. Furthermore, in the accompanying drawings, the same reference numerals are used to designate the same or similar components, and redundant explanations will be omitted.

[0012] (First embodiment) As a first embodiment of an information processing device according to the present invention, an information processing device that manages the histories of a learning model and learning data as traceability information will be described below as an example.

[0013] <Summary> When additional learning is performed on a learning model (trained model) using learning data, the information processing device of this embodiment manages information about the used learning data (learning data ID) in association with the learning model obtained by the additional learning. In addition, if the learning model is a derived model, information about the used initial model (initial model ID) is managed in association with the learning model. Furthermore, if the learning data is derived data, information about the used initial dataset (initial data ID) is managed in association with the learning data. This enables the information processing device to identify past learning models and learning data that contributed to the creation of the learning model created by the additional learning.

[0014] <Device configuration> FIG. 1 is a block diagram showing the functional configuration of an information processing device 11. The information processing device 11 manages "models" and "data" used for model learning and evaluation. In addition, the information processing device 11 learns models, edits data, and registers model information in response to requests from users. Here, it is assumed that the models and data are available for use by third parties via a learning model publishing platform. In other words, the user is a user of the learning model publishing platform.

[0015] The information processing device 11 is communicably connected to the user terminal 12 and the user terminal 13 via a communication network such as a local area network (LAN) or the Internet. The method of connection is not particularly limited. For example, each may be connected separately via a wired connection or via wireless communication. Although two user terminals are shown in FIG. 1, this is not limiting. Each user terminal is assumed to be an information terminal such as a personal computer (PC), a mobile phone, or a tablet terminal device.

[0016] The information processing device 11 includes a control unit 10 and a storage unit 20. The control unit 10 can be realized, for example, by a central processing unit (CPU) executing various programs stored in the storage unit 20. The storage unit 20 is a large-capacity storage device such as random access memory (RAM), read-only memory (ROM), or a hard disk drive (HDD) or solid-state drive (SSD).

[0017] The control unit 10 includes a model management unit 101, a data management unit 102, a traceability information management unit 103, a usage reception unit 104, a learning and evaluation unit 105, a registration unit 106, and a display control unit 107. As described above, each of these functional units can be realized by, for example, a CPU executing various programs. However, some or all of these functional units may also be realized by hardware such as an application specific integrated circuit (ASIC).

[0018] The model management unit 101 manages models registered in the information processing device 11. In this embodiment, the model is assumed to be an "object detection model" intended to perform object detection, which is a form of image recognition, but is not limited to this. It may also be a voice recognition model, a natural language processing model, or a generative artificial intelligence (AI) model. The model management unit 101 assigns model information to the model and manages it.

[0019] 4 is a diagram showing examples of model information 401, 402 assigned to two models. The model information is attribute information of the model, and includes information on the model ID, user ID, initial model ID, learning data ID, tag, task, evaluation data ID, evaluation result, and disclosure range. In other words, it includes identification information (ID) that identifies the initial model and data used to generate the model.

[0020] The model ID is model-specific identification information (ID) for identifying each model. The user ID is a user-specific ID (user information) for identifying the user who performed training for each model. The initial model ID is a model ID that specifies the model that became the initial parameter when training each model.

[0021] 4, in model information 401, the model ID is "m0001" and the initial model ID is "m0000". In addition, in model information 402, the model ID is "m0002" and the initial model ID is "m0001". This indicates that the model with the model ID "m0002" is a model obtained as a result of learning using the model with the model ID "m0001" as the initial model.

[0022] The training data ID is a unique ID for the training data that identifies the dataset used to train each model. The tag contains information about the objects that each model targets and information necessary for users to search for models. The task contains the type of processing that each model targets (object detection, image generation, etc.). The evaluation data ID is a data ID that specifies the dataset used when evaluating each model. The evaluation results contain numerical values ​​obtained as a result of the evaluation. The disclosure range contains information indicating the disclosure range (the range of disclosure permitted) of each model. For example, it can be all (public (fully public)), private, or one or more user IDs. If a user ID is listed, only users with the listed ID will be able to access the model; other users will not be able to access the model.

[0023] 4, model information 401 and 402 are both model information for an object detection model. For example, if the model is another type of model (such as a voice recognition model), the items included in the model information may be different from those in the example of FIG.

[0024] The data management unit 102 manages data registered in the information processing device 11. The data in this embodiment is image data for object detection. The data management unit 102 accepts registration of a data set from a user, and manages the data set by assigning data information to the data set.

[0025] 5 is a diagram showing examples of data information 501, 502 assigned to data (image data set). The data information is attribute information of the data, and includes information on a data ID, a user ID, an initial data ID, a tag, a task, and a disclosure range.

[0026] The data ID is a data-specific ID used to identify each piece of data. The user ID is a user-specific ID (user information) used to identify the user who created each piece of data. The initial data ID is the data ID of the initial dataset that was used to create the data. The tag contains information about the object that each piece of data targets, as well as information necessary for users to search for data. The task contains the type of processing that each piece of data targets (object detection, image generation, etc.). The disclosure range contains information indicating the disclosure range of each model (the range of disclosure permitted). The information contained is the same as the disclosure range in the model information.

[0027] For example, data information 501 is data information for data created by user 1 (ID is "u0001"), and data information 502 is data information for data created by user 2 (ID is "u0002"). The initial data ID in data information 502 is "d0001", which indicates that the data corresponding to data information 502 was created using the data corresponding to data information 501 as initial data.

[0028] 5, data information 501 and 502 are both data information of data used in a model for an object detection model. For example, if the data is used in another type of model (such as a voice recognition model), the items included in the data information may be different from those in the example of FIG.

[0029] The traceability information management unit 103 manages traceability information of models and data. Here, the traceability information is information indicating the process by which each model and each data was created in the past (i.e., the past history).

[0030] Figure 6 is a diagram showing an example of traceability information. In Figure 6, solid arrows indicate derivation relationships between models and between data. That is, the relationships between multiple learning models and multiple data sets published on the learning model publishing platform are integrated and managed. Also, in Figure 6, dotted arrows indicate the data used to train each model.

[0031] For example, model 601 ("m0001") indicates that it is a model trained using data 603 ("d0001"). Also, model 602 ("m0002") indicates that it is a derived model created by additionally training model 601 ("m0001") as an initial model using data 604 ("d0002"). Also, data 604 indicates that it is derived data created (edited or added) based on data 603.

[0032] In this way, traceability information shows the records of how each model and data has been edited in the past. Therefore, by tracing the traceability information, it is possible to identify one or more models that were used to create each model. It is also possible to identify one or more data that were used to create each data.

[0033] The usage acceptance unit 104 accepts a request to use a model or data from a user and determines whether the model or data can be used by the user. The usability is determined based on the traceability information managed by the traceability information management unit 103. Models and data determined to be usable here become usable in the learning and evaluation unit 105, and are used as initial models and learning data.

[0034] The learning and evaluation unit 105 performs learning and evaluation of the model based on the model and data determined by the usage acceptance unit 104 to be usable.

[0035] The registration unit 106 registers models and data. For example, it registers a new model created by additional learning by the learning and evaluation unit 105. It also receives new data from a user and registers it as new data. As described with reference to FIGS. 4 and 5, when creating a newly registered model or data, if a model or data that has already been registered by the registration unit 106 is used, an initial model ID and initial data ID are also registered.

[0036] The display control unit 107 controls the display of the results of the information processing device 11 in response to requests from the user terminals 12 and 13 .

[0037] The storage unit 20 stores the various programs executed by the CPU described above, as well as the models, model information, data, data information, traceability information, and the like described above.

[0038] <Device Operation> Fig. 2 is a diagram showing the flow of processing in an information processing device in response to a request from a user terminal. Fig. 3 is a flowchart for learning and registering a model in information processing device 11. Fig. 3(a) is a flowchart for learning and registering a model based on a request from user terminal 12, and Fig. 3(b) is a flowchart for learning and registering a model based on a request from user terminal 13.

[0039] 2, in the following explanation, it is assumed that user 1 operates user terminal 12 to register data 603 and model 601. Then, it is assumed that user 2 subsequently operates user terminal 13 to edit data 603, create and register data 604, and register model 602 obtained by additional learning using model 601 as an initial model.

[0040] Also, assume that models 601 and 602 are object detection models for focusing a camera on, for example, a predetermined type of object. Here, assume that user 1 "wants to detect a person and a dog (as a predetermined type of object)." Also, assume that user 2 "wants to detect a dog's face (as a predetermined type of object)." However, the predetermined type of object is not limited to these.

[0041] First, the user 1 operates the user terminal 12 to request the information processing device 11 to add learning data that has been prepared in advance.

[0042] FIG. 7 is a diagram showing an example of training data. The training data includes an image 701 and a ground truth value 704. The image 701 is an image showing subjects that the user wants to recognize (here, a person 702 and a dog 703). Furthermore, in order to train object detection, not only an image but also subject information as ground truth (GT) is required. The subject information is, for example, bounding box information (BB information) corresponding to information on the position and size of an object to be detected in an image. BB705 is a BB corresponding to the person 702, and BB706 is a BB corresponding to the dog 703. In other words, the training data includes one or more pairs of an image and a ground truth value corresponding to the image.

[0043] In S301, the registration unit 106 registers data in response to a data addition request from the user terminal 12 (user 1). The registered data is assigned data information such as that shown in data information 501. Here, the data ID of the data registered by user 1 is assumed to be "d0001." User 1 can assign information such as the disclosure range, tags, and tasks of the learning data along with the request to register the learning data. Note that in the data information 501, there is no initial data linked to the registered data, so the initial data ID is left blank.

[0044] In S302, the traceability information management unit 103 assigns new traceability information to the data registered in S301. Here, information such as that shown in data 603 in Fig. 6 is registered as traceability information. At this stage, there is no information linked to the data 603 registered by user 1, and only the data 603 is registered independently as traceability information.

[0045] In S303, the learning and evaluation unit 105 executes learning using the registered data in response to a learning execution request from the user terminal 12 (user 1). Here, the learning and evaluation unit 105 accepts specification of an initial model ("m0000") from the user, and executes learning for that model using data ("d0001"). It is assumed that the model with model ID "m0000" has been registered in advance in the storage unit 20.

[0046] As a learning method for object detection, for example, a method using a neural network (NN) exists. The learning method for object detection using a NN is described in detail in Literature A. The learning and evaluation unit 107 may evaluate the model trained using the data used for learning or any data.

[0047] (Reference A) Tian et al., "FCOS: Fully Convolutional One-Stage Object Detection", arXiv:1904.01355, 2019 In S304, the registration unit 106 registers the model learned in S303 in the storage unit 20. When registering the model, for example, model information such as that shown in model information 401 is assigned. The user ID is the user ID (“u0001”) of the user 1 operating the user terminal 12.

[0048] In S305, the traceability information management unit 103 updates the traceability information. Since the model ("m0001") was trained using the data ("d0001"), information such as that shown in data 601 in Fig. 6 is registered. In addition, the link (dotted arrow) between the model ("m0001") and the data ("d0001") is registered.

[0049] As a result of the above, the data prepared by the user 1 ("d0001") and the model trained using the data ("m0001") are registered in the information processing device 11.

[0050] Next, user 2 operates user terminal 13 to request information processing device 11 to use the data and model registered by user 1 to edit the data and learn the model.

[0051] In S306, the use acceptance unit 104 acquires the learning data in response to a data acquisition request from the user terminal 13 (user 2). At the same time, the use acceptance unit 104 inquires of the data management unit 102 whether user 2 can use the data ("d0001"). In response, the data management unit 102 checks the disclosure range of the data ("d0001"). In this case, the disclosure range is "all" (may be disclosed to everyone).

[0052] In S307, the traceability information management unit 103 checks the traceability information of the data ("d0001") and determines that no original initial data exists for the data ("d0001").

[0053] In S308, the traceability information management unit 103 determines whether the data can be used. In S306, the disclosure range of the data ("d0001") is "all," and in S307, there is no initial data that is the source of the data. Therefore, it is determined that the data can be used by all users who can access the information processing device 11, and user 2 can use the data. Note that, if it is determined in S307 that the traceability information includes data for which a disclosure range is set so that the data is not made public to user 2, user 2 cannot use the data.

[0054] In S309, the data management unit 102 edits the learning data in response to a request from the user terminal 13 (user 2).

[0055] 8 is a diagram showing an example of editing training data. As described above, here, user 2 is in a situation where he wants to "detect the face of a dog." Therefore, while using image 701 (FIG. 8(a)) of data ("d0001"), BB803 is assigned to the dog's face as the true value (GT) 801 corresponding to image 701.

[0056] In S310, the registration unit 106 registers the data edited by user 2. The data information assigned to the registered data ("d0002") is as shown in data information 502. Note that, here, as an example, the disclosure range is set to two users ("u0002" and "u0003").

[0057] In S311, the traceability information management unit 103 updates the traceability information. As described above, since the data ("d0002") was created based on the data ("d0001"), traceability information 606 indicating that the data 604 is derived data of the data 603 is created (updated).

[0058] In S312, the usage acceptance unit 104 accepts a request to acquire a model to be used as an initial model from the user terminal 13 (user 2). Here, it is assumed that user 2 uses the model (“m0001”) created by user 1 as the initial model.

[0059] In S313, the model management unit 101 checks the model information 401 assigned to the model for which the acquisition request has been made. It is found that the disclosure range of the model ("m0001") is "all." The traceability information management unit 103 also checks the traceability information of the model. As a result, it is determined that there is no original initial model for the model ("m0001").

[0060] In S314, the traceability information management unit 103 determines whether the model can be used. In S313, the disclosure range of the model ("m0001") is "all," and there is no initial model that is the basis of this model. Therefore, it is determined that the model can be used by all users who can access the information processing device 11, and user 2 can use this model. Note that, if it is determined in S313 that the traceability information includes a model for which the disclosure range is set so that it is not made public to user 2, user 2 cannot use this model.

[0061] In S315, the learning evaluation unit 105 executes learning using the data registered in S310 in response to a learning execution request from the user terminal 13 (user 2). Here, it accepts the specification of an initial model ("m0001") from the user, and executes learning for that model using data ("d0002"). The learning and evaluation procedures are the same as those in S303.

[0062] In S316, the registration unit 106 registers the model learned in S315 in the storage unit 20. When accepting registration of a model from user 2, for example, model information such as that shown in model information 402 is provided. The user ID is the user ID ("u0002") of user 2 operating the user terminal 13. Here, as an example, the disclosure range is set to two users ("u0002" and "u0003").

[0063] In S317, the traceability information management unit 103 updates the traceability information. As described above, the model ("m0002") was created by performing additional learning on the model ("m0001"), and therefore the traceability information 606 indicating that the model 602 is a derived model of the model 601 is created (updated).

[0064] As described above, according to the first embodiment, when additional learning is performed on an initial model using learning data to generate a new learning model (derived model), the history of each learning model and learning data is managed as traceability information. This makes it possible to identify one or more learning models and one or more datasets that contributed to the creation of any learning model of interest (included in multiple learning models published on the learning model publishing platform).

[0065] Furthermore, by referring to the traceability information, it becomes possible to appropriately determine whether other users can use the model and data obtained by additional learning, based on the scope of disclosure of past models and data used as the initial model and initial data. Note that instead of centrally managing the traceability information in the information processing device 11, it may be configured to manage it in a distributed manner using blockchain technology or the like.

[0066] In the above description, a set of an image and a true value is given as an example of the training data, but it does not necessarily have to be a set of an image and a true value. For example, it may be a set of a voice or text and a true value, or it may be only an image, a voice, or a text. It may also be a combination of an image and a text. Furthermore, it may be applied not only to image recognition such as the object detection described above, but also to natural language processing such as prediction and recognition of time series data and text classification.

[0067] (Second embodiment) In the second embodiment, a learning model created by additional learning is made public on a learning model publishing platform, and a method for calculating usage fees when the learning model is used by a third party will be described.

[0068] In particular, the following explanation will explain a method for determining the compensation that user 3 will pay to users (user 1 and user 2) who were involved in creating the model ("m0002") registered by user 2 in the first embodiment when user 3 uses the model.

[0069] <Device configuration> Fig. 9 is a block diagram showing the functional configuration of an information processing device 11 according to the second embodiment. Fig. 9 differs from the first embodiment (Fig. 1) in that a user terminal 91 is added and a value calculation unit 901 is added to the control unit 10.

[0070] It is assumed that user terminal 91 is a terminal operated by user 3, and that user 3's user ID is "u0003".

[0071] The fee calculation unit 901 is a functional unit for calculating the fee for using the published learning model using traceability information. As will be described in detail later, the fee is calculated based on the initial model and data used to train the learning model.

[0072] <Device Operation> 10 is a flowchart of the learning process in the information processing device according to the second embodiment. Specifically, the flowchart shows the operation when a user 3 performs learning and evaluation using a model ("m0002") and data ("d0002") to create a new model.

[0073] In S1001, the use acceptance unit 104 acquires learning data in response to a data acquisition request from the user terminal 91 (user 3). At the same time, the use acceptance unit 104 inquires of the data management unit 102 whether user 3 can use the data ("d0002"). In response, the data management unit 102 checks the disclosure range of the data ("d0002"). As shown in FIG. 5, the disclosure range here is "u0002, u0003".

[0074] In S1002, the traceability information management unit 103 checks the traceability information of the data ("d0002"). As shown in Fig. 5, the initial data that is the source of the data (data ID is "d0002") is the data ("d0001"), and the disclosure range of the initial data is "all".

[0075] In S1003, the traceability information management unit 103 determines whether the data can be used. In S1001, the disclosure range of the data ("d0002") is "u0002, u0003", and in S1002, the disclosure range of the initial data that is the source of the data is "all". Based on the logical product (AND) of the disclosure ranges, it is determined that user 3 can use the data ("d0002").

[0076] In S1004, the usage acceptance unit 104 accepts a request to acquire a model to be used as an initial model from the user terminal 91 (user 3). Here, it is assumed that user 3 uses the model ("m0002") created by user 2 as the initial model.

[0077] In S1005, the model management unit 101 checks the model information 402 assigned to the model for which the acquisition request has been made. The disclosure range of the model ("m0002") is "u0002, u0003". In addition, the traceability information management unit 103 checks the traceability information of the model. As shown in FIG. 4, the initial model that is the basis of the model ("m0002") is the model ("m0001"), and the disclosure range of the initial model is "all".

[0078] In S1006, the traceability information management unit 103 determines whether the model can be used. In S1005, the disclosure range of the model ("m0002") is "u0002, u0003", and the disclosure range of the initial model that is the basis of this model is "all". Based on the logical product (AND) of the disclosure ranges, it is determined that user 3 can use the model.

[0079] In S1007, the learning evaluation unit 105 executes learning using registered data in response to a learning execution request from the user terminal 91 (user 3). Here, it accepts specification of an initial model ("m0002") from the user, and executes learning for that model using data ("d0002"). The learning and evaluation procedures are the same as those in S303.

[0080] In S1008, the registration unit 106 registers the model learned in S1007 in the storage unit 20. When registering a model, for example, model information such as that shown in model information 1101 is assigned. FIG. 11 is a diagram showing an example of model information 1101 assigned to a model. Note that the user ID is the user ID ("u0003") of user 3 operating the user terminal 91. Also, here, as an example, the disclosure range is set to user 3 ("u0003") (i.e., only oneself).

[0081] In S1009, the traceability information management unit 103 updates the traceability information. As described above, the model ("m0003") was created by performing additional learning on the model ("m0002"), and therefore, traceability information 1201 indicating that the model 1202 is a derived model of the model 602 is created (updated). FIG. 12 is a diagram showing an example of the traceability information 1201.

[0082] In S1010, the fee calculation unit 901 calculates the fee to be paid for user 3's use of the model ("m0002") and data ("d0002"). The fee calculation unit 901 first traces the traceability information 1201 for the model ("m0002") and data ("d0002") to acquire model information and data information. This makes it possible to determine the past models and data that contributed to the creation of these models and data. Here, it is clear that the data ("d0002") was created based on the initial data ("d0001"). It is also clear that the initial model of the model ("m0002") is ("m0001"). Therefore, user 2 (u0002) and user 1 (u0001) are identified based on the "user ID" entries in these models ("m0002" and "m0001") and data ("d0002" and "d0001").

[0083] Furthermore, each model and data is weighted in advance based on the degree of contribution, for example, the weighting for data is determined based on the "number of registered learning data" and the weighting for models is determined based on the "time required for learning", etc. The specific amount of compensation can be further determined based on these weightings.

[0084] As described above, according to the second embodiment, when additional learning is performed on an initial model using learning data to generate a new learning model (derived model), the history of each learning model and learning data is managed as traceability information. This makes it possible to determine the usage fee (compensation) when a user uses models and data registered by other users, taking into consideration all users who contributed to the creation of these models and data. In other words, it becomes possible to pay compensation according to the degree of contribution not only to the model but also to users who contributed to the creation of the data.

[0085] (Variation) In the above embodiment, the learning of an object detection model has been described as an example, but the learning of other types of models can be similarly applied. Below, as a modified example, a form applied to an image generation model used in an image generation task will be described. Image generation AI is an image generation technology as exemplified by Document B.

[0086] (Reference B) Rombach et al., "High-Resolution Image Synthesis with Latent Diffusion Models", CVPR 2022, arXiv:2112.10752, 2021 The functional configuration and operation of the information processing device are the same as those of the first embodiment (FIGS. 1 to 3), and therefore will not be described here. However, the difference is that character strings (text) are used instead of BBs as true value data used for model learning.

[0087] FIG. 13 is a diagram showing an example of training data in a modified example. The training data includes an image 1301 and a true value 1302. The image 1301 is an image showing a subject (here, a dog) that the user wants to recognize. Furthermore, in order to train a model for an image generation task, not only the image but also subject information as a true value (GT) is required. The subject information is, for example, text that represents the subject in the image. Here, the true value 1302 specifies "dog" as the text corresponding to the subject shown in the image 1301.

[0088] Figure 14 is a diagram showing an example of information assigned to data in a modified example. Here, it is assumed that user 1 (u0001) creates a model ("m0001") using data ("d0001") as shown in Figure 13 for the initial model ("m0000") and publishes it on the learning model publishing platform. In this case, user 2 (u0002) who uses the learning model publishing platform can create a model ("m0002") using data ("d0002") that he created for the published model ("m0001").

[0089] The disclosure of this specification includes the following information processing device, control method, and program. (Item 1) An information processing device that manages a plurality of learning models, A model management means for managing, for each of the plurality of learning models, first information that identifies a dataset used in learning the learning model and second information that identifies an initial model used in learning the learning model; A data management means for managing third information that identifies an initial dataset used to create a dataset for each of a plurality of datasets identified by the first information for each of the plurality of learning models; A determination means for determining one or more learning models and one or more datasets used in learning of the attention learning model included in the plurality of learning models based on the first information, the second information, and the third information; An information processing device comprising: (Item 2) Further provided is a management means for managing traceability information that integrates the relationship between learning models generated based on the second information of each of the plurality of learning models, the relationship between data sets generated based on the third information of each of the plurality of data sets, and the first information of each of the plurality of learning models; The determining means determines one or more learning models and one or more data sets that contributed to the attention learning model based on the traceability information. 2. The information processing device according to item 1, (Item 3) The traceability information is managed in a distributed manner by a blockchain. 3. The information processing device according to item 2, (Item 4) The model management means further manages, for each of the plurality of learning models, first user information that identifies a user who participated in learning of the learning model; The data management means further manages, for each of the plurality of data sets, second user information that identifies a user who participated in creating the data set; The determining means further determines one or more users who contributed to the attention learning model based on the first information, the second information, the third information, the first user information, and the second user information. 4. The information processing device according to any one of items 1 to 3. (Item 5) The system further includes a compensation calculation means for calculating compensation for one or more users who contributed to the attention learning model based on one or more learning models and one or more data sets that contributed to the attention learning model. 5. The information processing device according to item 4. (Item 6) a preset weight is assigned to each of the plurality of learning models and the plurality of data sets; The compensation calculation means calculates compensation for the one or more users based on the weightings set for the one or more learning models and the one or more data sets. 6. The information processing device according to item 5, (Item 7) Each of the plurality of data sets includes one or more pairs of an image and a corresponding true value. 7. The information processing device according to any one of items 1 to 6, (Item 8) When used for training an object detection model, the true value is bounding box information of a detection target object included in the image. 8. The information processing device according to item 7, (Item 9) When used to train an image generation model, the true value is a character string that indicates a subject included in the image. 8. The information processing device according to item 7, (Item 10) The plurality of learning models and the plurality of datasets are made public via a model publishing platform; The system further includes a registration acceptance means for accepting registration of a second learning model created by performing additional learning on the first learning model published on the model publishing platform, The registration receiving means registers identification information of the first learning model as the second information of the second learning model in the model management means. 10. The information processing device according to any one of items 1 to 9, (Item 11) The registration accepting means further accepts registration of a second dataset created using the first dataset published on the model publishing platform; The registration receiving means registers the identification information of the first data set as the second information of the second data set in the data management means. 11. The information processing device according to item 10. (Item 12) A control method for an information processing device that manages a plurality of learning models, comprising: a model information acquisition step of acquiring, for each of the plurality of learning models, first information that identifies a dataset used in learning the learning model and second information that identifies an initial model used in learning the learning model; a data information acquisition step of acquiring third information that identifies an initial dataset used to create a dataset for each of a plurality of datasets identified by the first information for each of the plurality of learning models; A determination step of determining one or more learning models and one or more datasets used in learning of the attention learning model included in the plurality of learning models based on the first information, the second information, and the third information; A control method comprising: (Item 13) Item 13. A program for causing a computer to execute the control method according to Item 12.

[0090] (Other Examples) The present invention can also be realized by supplying a program that realizes one or more functions of the above-described embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program.The present invention can also be realized by a circuit (e.g., ASIC) that realizes one or more functions.

[0091] The invention is not limited to the above-described embodiments, and various changes and modifications can be made without departing from the spirit and scope of the invention. Accordingly, the following claims are appended to apprise the public of the scope of the invention. [Explanation of symbols]

[0092] 10 control unit; 11 information processing device; 20 storage unit; 12, 13 user terminal

Claims

1. An information processing device that manages a plurality of learning models, a model management means for managing, for each of the plurality of learning models, first information that identifies a dataset used in learning the learning model and second information that identifies an initial model used in learning the learning model; a data management means for managing third information that identifies an initial dataset used to create a dataset for each of a plurality of datasets identified by the first information for each of the plurality of learning models; A determination means for determining one or more learning models and one or more datasets used in learning of an attention learning model included in the plurality of learning models based on the first information, the second information, and the third information; An information processing device comprising:

2. Further provided is a management means for managing traceability information that integrates the relationship between learning models generated based on the second information of each of the plurality of learning models, the relationship between data sets generated based on the third information of each of the plurality of data sets, and the first information of each of the plurality of learning models; The determining means determines one or more learning models and one or more data sets that contributed to the attention learning model based on the traceability information.

2. The information processing apparatus according to claim 1, wherein:

3. The traceability information is managed in a distributed manner by a blockchain.

3. The information processing apparatus according to claim 2, wherein:

4. The model management means further manages, for each of the plurality of learning models, first user information that identifies a user who participated in learning of the learning model; The data management means further manages, for each of the plurality of data sets, second user information that identifies a user who participated in creating the data set; The determining means further determines one or more users who contributed to the attention learning model based on the first information, the second information, the third information, the first user information, and the second user information.

2. The information processing apparatus according to claim 1, wherein:

5. The system further includes a compensation calculation means for calculating compensation for one or more users who contributed to the attention learning model based on one or more learning models and one or more data sets that contributed to the attention learning model.

5. The information processing apparatus according to claim 4,

6. a preset weight is assigned to each of the plurality of learning models and the plurality of data sets; The compensation calculation means calculates compensation for the one or more users based on the one or more learning models and the weights set for each of the one or more data sets.

6. The information processing apparatus according to claim 5,

7. Each of the plurality of data sets includes one or more pairs of an image and a corresponding true value.

2. The information processing apparatus according to claim 1, wherein:

8. When used for training an object detection model, the true value is bounding box information of a detection target object included in the image.

8. The information processing apparatus according to claim 7,

9. When used to train an image generation model, the true value is a character string that indicates a subject included in the image.

8. The information processing apparatus according to claim 7,

10. The plurality of learning models and the plurality of datasets are made public via a model publishing platform; The system further includes a registration acceptance means for accepting registration of a second learning model created by performing additional learning on the first learning model published on the model publishing platform, The registration receiving means registers identification information of the first learning model as the second information of the second learning model in the model management means.

2. The information processing apparatus according to claim 1, wherein:

11. The registration accepting means further accepts registration of a second dataset created using the first dataset published on the model publishing platform; The registration accepting means registers the identification information of the first data set as the second information of the second data set in the data management means.

11. The information processing apparatus according to claim 10,

12. A control method for an information processing device that manages a plurality of learning models, comprising: a model information acquisition step of acquiring, for each of the plurality of learning models, first information that identifies a dataset used in learning the learning model and second information that identifies an initial model used in learning the learning model; a data information acquisition step of acquiring third information that identifies an initial dataset used to create each of a plurality of datasets identified by the first information for each of the plurality of learning models; A determination step of determining one or more learning models and one or more datasets used in learning of an attention learning model included in the plurality of learning models based on the first information, the second information, and the third information; A control method comprising:

13. A program for causing a computer to execute the control method according to claim 12.

Images