Data processing method, data processing device, and program

Abstract

To provide a technique for selecting data generated by a generation model according to the uniqueness of the data.SOLUTION: A data processing method includes: a generation step of generating target embedded data in which target data is projected to embedded space by inputting the target data generated by a generation model to a learning model for projecting input data to the embedded space; a calculation step of calculating, using the target embedded data and reference embedded data in which a plurality of pieces of reference data generated by the generation model are projected to the embedded space, density of the target data in the embedded space to which each piece of reference data is projected; a determination step of determining a level of uniqueness of the target data according to the density; and a presentation step of presenting a determination result of the level by the determination step to a user.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present disclosure relates to a data processing method, a data processing device, and a program. [Background technology]

[0002] Patent Document 1 discloses a technology for outputting an output image in which an input image of a natural object is multiplied with an artificial object, based on a learning model that has been trained to output an image in which an input image of a natural object is multiplied with an artificial object. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-168078 Summary of the Invention [Problem to be solved by the invention]

[0004] In a technology for outputting data generated by a generative model that generates data based on input data, as in Patent Document 1, the data generated by the generative model may be similar to each other, and the uniqueness of the output data may be lower than the desired diversity. Conversely, for example, when data with lower uniqueness is desired, the uniqueness of the output data may be higher than the desired uniqueness. Therefore, a technology for selecting data generated by a generative model based on the uniqueness of the data is desired. [Means for solving the problem]

[0005] According to a first aspect of the present disclosure, there is provided a data processing method, comprising: a generating step of generating target embedded data in which target data generated by a generative model is input into a learning model that projects input data into an embedding space, by which the target data is projected into the embedding space; a calculating step of calculating, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model are projected into the embedding space, a determining step of determining whether the uniqueness of the target data is high or low based on the density; and a presenting step of presenting the result of the determining step to a user.

[0006] According to a second aspect of the present disclosure, there is provided a data processing method, comprising: a generating step of generating target embedded data in which target data generated by a generative model is input into a learning model that projects input data into an embedding space, by which the target data is projected into the embedding space; a calculating step of calculating, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model are projected into the embedding space, a determining step of determining whether or not to present the target data to a user based on the density; and a presenting step of presenting the target data to the user when it is determined in the determining step that the target data should be presented.

[0007] According to a third aspect of the present disclosure, there is provided a data processing method, comprising: a generating step of generating target embedded data in which target data generated by a generative model is input to a learning model that projects input data into an embedding space, by which the target data is projected into the embedding space; a calculating step of calculating a density of the target data in the embedding space in which each of the reference data is projected, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model are projected into the embedding space; and a storing step of storing the target data in which the density is equal to or greater than a predetermined threshold.

[0008] According to a fourth aspect of the present disclosure, there is provided a data processing device. The data processing device includes: a learning model that projects features of input data into an embedding space; a memory unit that stores reference embedded data in which multiple reference data generated by a generative model are projected into the embedding space; and a processor. The processor executes the following operations: an acquisition process that acquires target data generated by the generative model; a generation process that inputs the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process that uses the target embedded data and the reference embedded data to calculate the density of the target data in the embedding space into which each reference data is projected; a determination process that determines whether the target data is highly unique based on the density; and a process that presents the determination result of the determination process to a user.

[0009] According to a fifth aspect of the present disclosure, there is provided a data processing device. The data processing device includes: a learning model that projects features of input data into an embedding space; a memory unit that stores reference embedded data in which multiple reference data generated by a generative model are projected into the embedding space; and a processor. The processor executes the following operations: an acquisition process that acquires target data generated by the generative model; a generation process that inputs the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process that uses the target embedded data and the reference embedded data to calculate the density of the target data in the embedding space into which each reference data is projected; a determination process that determines whether to present the target data to a user based on the density; and a process that presents the target data to the user if the determination process determines that the target data should be presented.

[0010] According to a sixth aspect of the present disclosure, there is provided a data processing device. The data processing device includes: a learning model that projects features of input data into an embedding space; a memory unit that stores reference embedded data in which multiple reference data generated by a generative model are projected into the embedding space; and a processor. The processor executes the following operations: an acquisition process that acquires target data generated by the generative model; a generation process that inputs the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process that uses the target embedded data and the reference embedded data to calculate a density of the target data in the embedding space into which each reference data is projected; and a process of saving the target data whose density is equal to or greater than a predetermined threshold.

[0011] According to a seventh aspect of the present disclosure, there is provided a program that causes a computer to implement the following functions: (1) inputting target data generated by a generative model into a learning model that projects features of the input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; (2) using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, (3) calculating the density of the target data in the embedding space into which each of the reference data is projected, (4) determining whether the target data is highly unique based on the density, and (5) presenting the result of the determination made by the determination function to a user.

[0012] According to an eighth aspect of the present disclosure, there is provided a program that causes a computer to implement the following functions: generating target embedded data in which target data generated by a generative model is projected into an embedding space by inputting the target data into a learning model that projects features of the input data into the embedding space; calculating, using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, a determination function determining whether to present the target data to a user based on the density; and, when the determination function determines that the target data should be presented, presenting the target data to the user.

[0013] According to a ninth aspect of the present disclosure, there is provided a program that causes a computer to perform the following functions: inputting target data generated by a generative model into a learning model that projects features of the input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, calculate a density of the target data in the embedding space into which each of the reference data is projected; and saving the target data whose density is equal to or greater than a predetermined threshold. [Brief explanation of the drawings]

[0014] [Figure 1] FIG. 1 is a block diagram showing a schematic configuration of a data processing system according to a first embodiment. [Figure 2] A conceptual diagram explaining the learning method of the learning model. [Figure 3] 10 is a flowchart of pre-processing. [Figure 4] 4 is a flowchart of a main process in the first embodiment. [Figure 5] FIG. 4 is an explanatory diagram showing an example of a probability density function calculated by a calculation unit. [Figure 6] 10A and 10B are diagrams showing examples of results of uniqueness determination of each piece of target data. [Figure 7] 10 is a flowchart of a main process in the second embodiment. [Figure 8] 10 is a flowchart of a main process according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0015] A. First embodiment: 1 is a block diagram showing a schematic configuration of a data processing system 50 according to the first embodiment. The data processing system 50 is used to select generated data generated by a generative model 170 according to the uniqueness of the data. The generated data is various data such as image data, spectral data, and character string data.

[0016] The uniqueness of data represents the degree to which the number of other data similar to the data in the data group to which the data belongs is small. The greater the number of other data similar to the data in the data group to which the data belongs, above a predetermined level, the higher the uniqueness of the data. Generally, whether or not two pieces of data are similar to each other can be determined based on the distance between the pieces of data or the similarity between the pieces of data. In this case, for example, Euclidean distance or Manhattan distance can be used as the distance between the pieces of data. Furthermore, for example, cosine similarity or similarity based on cross entropy can be used as the similarity. Furthermore, the uniqueness of each piece of data correlates with the diversity of each piece of data. Specifically, the greater the amount of unique data contained in a data group, the greater the diversity of each piece of data contained in the data group. Hereinafter, a data group with a high level of diversity of each piece of data or a data set with a high level of diversity of each piece of data will also be simply referred to as a "highly diverse data group" or a "highly diverse data set."

[0017] The data processing system 50 includes a data processing device 100. In this embodiment, the data processing device 100 is configured by a computer including one or more processors 101, a storage unit 102 including ROM and RAM, an input / output interface 103, and an internal bus 104. The processor 101, the storage unit 102, and the input / output interface 103 are connected via the internal bus 104 to enable bidirectional communication. An output device 105 and an input device 106 are connected to the input / output interface 103. In this embodiment, the storage unit 102 stores a program 155, a learning model 160, reference embedded data 165, and a generative model 170. By executing the program 155 stored in the storage unit 102, the processor 101 realizes various functions, including those of a data generation unit 108, an acquisition unit 110, an embedded data generation unit 115, a calculation unit 120, a determination unit 125, a presentation unit 130, a determination condition reception unit 135, and a learning execution unit 140.

[0018] The input device 106 is configured by, for example, a mouse and a keyboard. The data processing device 100 receives various inputs from the user Ur via the input device 106.

[0019] The output device 105 outputs various data processing results obtained by the data processing device 100. The data processing results include result information, which will be described later. In this embodiment, the output device 105 is configured as a display device that outputs the data processing results as visual information. The display device is configured, for example, by a liquid crystal panel or an organic EL panel. The output device 105 as a display device may be configured, for example, as a touch panel that can accept touch operations from the user Ur. In this case, the output device 105 may also function as the input device 106, for example. In other embodiments, the output device 105 may be configured, for example, as a speaker that outputs the data processing results as audio information, or as a device that outputs the data processing results by transmitting them to an external computer or recording medium.

[0020] The learning model 160 is trained to project input data into an embedding space. In the embedding space, each piece of data is projected so that the feature vectors of more similar data are closer in distance. The embedding space in this embodiment is a 64-dimensional vector space. The embedding space is also called a feature space. The feature vectors in the embedding space are also called embedding vectors or embedded representations.

[0021] In this embodiment, the learning model 160 has been trained by self-supervised learning (SSL). The learning model 160 is configured as a machine learning model using Resnet. More specifically, the learning model 160 is configured as a Resnet model that is a backbone included in an encoder in a network structure that has been trained using SimSiam. Details of the learning model 160 will be described later.

[0022] The generative model 170 is also called generative AI (generative artificial intelligence). In this embodiment, the generative model 170 is configured as a trained model that uses latent variables as input to generate generated data. The training of the generative model 170 may be executed in the data processing device 100 or may be executed outside the data processing device 100. The generative model 170 determines the latent variables used to generate the generated data based on, for example, random numbers. As the random numbers for determining the latent variables, for example, random numbers following a probability distribution such as a normal distribution are used. The latent variables determined in this manner are so-called noise data.

[0023] In particular, a generative adversarial network (GAN) that has been trained to generate image data using latent variables as input is used as the generative model 170 in this embodiment. In the GAN, a generator and a discriminator are trained using a training dataset for training the GAN. Specifically, the generator is trained to be able to generate fake data that approximates true data, and the discriminator is trained to be able to discriminate data generated by the generator. In this embodiment, a projected GAN is used as the generative model 170.

[0024] The data generation unit 108 generates a generated data group consisting of a plurality of pieces of generated data. Specifically, the data generation unit 108 generates the generated data group by generating a plurality of pieces of generated data using the generative model 170. The generated generated data group is stored in the storage unit 102. In this embodiment, the data generation unit 108 generates a first generated data group and a second data group as the generated data group. The first data group includes a reference data set and a training data set, which will be described later. The second generated data group includes a target data set, which will be described later.

[0025] The acquiring unit 110 acquires a target dataset. The target dataset is a dataset including target data. The target data is generated data generated by the generative model 170, and is data that is the target of data processing by the data processing device 100. In this embodiment, the target dataset is composed of multiple pieces of target data.

[0026] The embedding data generation unit 115 generates target embedding data in which target data is projected into the embedding space. Specifically, the embedding data generation unit 115 generates the target embedding data by inputting the target data into the trained learning model 160. The target embedding data may be generated for each target data. The embedding data generation unit 115 stores the generated target embedding data in the storage unit 102.

[0027] Furthermore, the embedding data generation unit 115 in this embodiment generates reference embedding data 165 in which multiple reference data included in the reference data set are projected into the embedding space. The reference embedding data 165 is generated in common for each reference data. The reference data is generated by the generative model 170 and serves as a reference for the target data. Specifically, the reference data is data that serves as a reference for calculating the density of the target data, which will be described later. In this way, the reference embedding data 165 is generated based on each reference data generated by the generative model 170, and therefore the reference embedding data 165 reflects the tendency of the characteristics of data that is likely to be generated by the generative model 170.

[0028] The calculation unit 120 calculates the density of the target data using the reference embedded data 165 and the target embedded data. That is, the calculation unit 120 calculates the density of the target data generated by the generative model 170 in the same way as the reference data, using the reference embedded data 165, which reflects the tendency of the features of the generated data generated by the generative model 170, as a reference. Note that the "density of the target data" here refers to the density of the target data of the reference embedded data 165 at the coordinates where the target data is located in the embedding space into which each reference data is projected.

[0029] In this embodiment, the calculation unit 120 calculates the density of the target data using kernel density estimation (KDE). Specifically, the calculation unit 120 first calculates a probability density function representing the probability distribution of each reference data in the embedded space using KDE, based on the embedded space into which each reference data is projected. The calculated probability density function represents the distribution density of data points in the embedded space. The calculation unit 120 then calculates the density of each target data based on the calculated probability density function and the vector coordinates of each target data projected into the embedded space. Note that in other embodiments, the embedded space into which the probability density function is calculated may include not only each projected reference data, but also, for example, one or more projected target data.

[0030] The determination unit 125 performs data determination according to the density of the target data calculated by the calculation unit 120. Specifically, the determination unit 125 performs at least one of a uniqueness determination and a presentation determination as data determination. The uniqueness determination refers to determining the level of uniqueness of the target data according to the density of the target data. The presentation determination refers to determining whether to present the target data to the user Ur according to the density of the target data. Herein, for example, when target data with a relatively high level of uniqueness is projected into the embedding space, the projected target data is located farther from each reference data in the embedding space. As a result, the density of the target data in the embedding space is relatively low. Conversely, when target data with a relatively high level of uniqueness is projected into the embedding space, the projected target data is located closer to each reference data in the embedding space. As a result, the density of the target data in the embedding space is relatively high. As such, since the density and uniqueness of the target data are suitably correlated, the determination unit 125 can appropriately perform uniqueness determination and presentation determination according to the density of the target data. Hereinafter, data determination will also be simply referred to as "determination." Furthermore, the step, processing, and function for executing data judgment are also referred to as a judgment step, judgment processing, and judgment function, respectively.

[0031] In this embodiment, the determination unit 125 performs data determination based on the result of normalizing the density of the target data. In this disclosure, "normalization" includes so-called min-max normalization and "standardization." Standardization is also called Z-score normalization. In this embodiment, the determination unit 125 performs uniqueness determination and presentation determination based on the result of normalizing the density of the target data by Z-score normalization.

[0032] More specifically, the determination unit 125 first uses the target data and each reference data as a population and normalizes each density in the population by Z-score normalization so that the average density in the population is 0 and the variance is 1. Then, if the density of the normalized target data is equal to or greater than a predetermined first threshold, the determination unit 125 determines that the uniqueness of the target data is low. Furthermore, if the density of the normalized target data is equal to or less than a second threshold that is less than the first threshold, the determination unit 125 determines that the uniqueness of the target data is high. Furthermore, if the density of the normalized target data is greater than the second threshold and less than the first threshold, the determination unit 125 determines that the uniqueness of the target data is about normal. Note that in other embodiments, the determination unit 125 may not determine the uniqueness of the target data in three stages as described above, but may instead determine it in two stages, four stages, or more stages, for example.

[0033] The presentation unit 130 presents result information to the user Ur according to the determination result of the determination step. When the determination unit 125 performs a uniqueness determination, the presentation unit 130 presents the result of the uniqueness determination, i.e., level information indicating the level of uniqueness of the target data, as result information. The level information may be, for example, a numerical value indicating the level of density of the target data, or a symbol or character string indicating the level of uniqueness of the target data. Furthermore, when the determination unit 125 performs a presentation determination, if the presentation determination determines that the target data should be presented, the presentation unit 130 presents the target data determined to be presented to the user Ur as result information. In this embodiment, the presentation unit 130 presents the result information to the user Ur via the output device 105. Hereinafter, the step and process of presenting the result information to the user Ur are also referred to as the presentation step and the presentation process, respectively. The processor 101 may store the target data determined to be presented, for example, on a recording medium. This allows for the creation of a dataset containing a large amount of highly unique data. Furthermore, as will be described later, the presentation unit 130 may present the target data to the user Ur, as well as information about the uniqueness of the target data, to the user Ur.

[0034] The judgment condition receiving unit 135 receives, from the user Ur, a designation of the judgment condition for the judgment step. In this embodiment, the judgment condition receiving unit 135 receives, as the judgment conditions, a designation of the first threshold value and a designation of the second threshold value. Specifically, the judgment condition receiving unit 135 receives the designation from the user Ur, for example, by displaying a dialog for designating the first threshold value and the second threshold value on the display unit of the output device 105. The user Ur designates the judgment condition by, for example, inputting the judgment condition via the input device 106.

[0035] The learning execution unit 140 executes machine learning of the learning model 160. In this embodiment, the learning execution unit 140 uses at least a portion of a reference dataset as a training dataset to train the learning model 160. The training dataset is a dataset for training the learning model 160. In this embodiment, the training dataset includes a plurality of pieces of generated data generated by the generative model 170. The training dataset also includes the entire reference dataset. In this embodiment, it can also be said that a portion of the training dataset and the entire reference dataset are the same data. Note that in other embodiments, the training dataset may not include the generated data generated by the generative model 170. For example, the training dataset may be a general-purpose dataset that does not include generated data.

[0036] FIG. 2 is a conceptual diagram illustrating a learning method of the learning model 160 in this embodiment. FIG. 2 shows a model structure SM for realizing learning by SimSiam. The model structure SM includes an encoder En and a predictor Pd. The encoder En is configured by a neural network. The encoder En includes a backbone Bb and a projector Pj as a head. The projector Pj is configured by a fully connected layer including an output layer among the layers of the neural network that configures the encoder En. The predictor Pd is a neural network configured by two fully connected layers.

[0037] When input data Dt, such as image data, is input to a model structure SM, first and second data augmentations DA, which are different data augmentations, are performed on the input data Dt. The first data Dt1, which is the input data Dt subjected to the first data augmentation, and the second data Dt2, which is the input data Dt subjected to the second data augmentation, are each input to an encoder En. When the input data Dt is image data, the data augmentation DA may be, for example, vertical flipping, horizontal flipping, rotation, enlargement, reduction, cropping, translation, color jitter, brightness jitter, contrast jitter, saturation jitter, hue jitter, grayscaling, Gaussian blur, partial masking, etc. "Different data augmentations" means that the types or numbers of data augmentations DA to be performed are different. The types and numbers of data augmentations to be performed as the first and second data augmentations are each determined probabilistically.

[0038] The encoder En outputs a feature vector z1 of the input first data Dt1 and a feature vector z2 of the input second data Dt2.

[0039] The predictor Pd converts and outputs the input feature vector so that the feature vector based on the first data Dt1 matches the feature vector based on the second data Dt2. The feature vector z1 and feature vector z2 output from the encoder En are each input to the predictor Pd alternatively. FIG. 2 shows a situation in which the feature vector z1 is input to the predictor Pd. When the feature vector z1 is input to the predictor Pd, the predictor Pd outputs a feature vector p1 based on the feature vector z1. On the other hand, when the feature vector z2 is input to the predictor Pd, the predictor Pd outputs a feature vector p2 based on the feature vector z2. Note that the feature vector p2 is not shown in FIG. 2.

[0040] In the similarity evaluation SA, the similarity between a feature vector based on the first data Dt1 and a feature vector based on the second data Dt2 is calculated. Specifically, in the similarity evaluation SA, the similarity D(p1, z2) between feature vector p1 and feature vector z2 and the similarity D(p2, z1) between feature vector z1 and feature vector p2 are calculated. The similarity D(p1, z2) is expressed by the following formula (1). Note that the similarity D(p2, z1) is expressed by a formula obtained by replacing p1 with p2 and z2 with z1 in formula (1).

[0041]

number

[0042] In the model structure SM, learning is performed so as to minimize the loss function L. The loss function L is expressed by the following equation (2).

[0043]

number

[0044] In learning in the model structure SM, the parameters of the encoder En and the predictor Pd are updated using gradient descent and backpropagation so that the loss function L is minimized. However, in learning in the model structure SM, a gradient stopping operation GS is executed. That is, the above formula (2) can be expressed by the following formula (3). In formula (3), stopgrad represents gradient stopping.

[0045]

number

[0046] That is, in updating the parameters of the encoder En, the feature vector z1 in the similarity D(p2, z1) and z2 in the similarity D(p1, z2) are treated as constants. Note that Fig. 2 shows how the gradient stopping operation GS is performed for the second data Dt2. Specifically, in Fig. 2, the encoder En receives the gradient Gd for the first data Dt1, but does not receive the gradient Gd for the second data Dt2.

[0047] Generally, collapse can be a problem in SSL. Collapse means that the learning model is trained so that a fixed feature vector is output from the encoder regardless of the data input to the encoder. In SimSiam, it is known that collapse is suppressed by arranging a predictor Pd corresponding to one of the first data Dt1 and the second data Dt2 and executing a gradient stopping operation GS corresponding to the other data. Therefore, by using SimSiam to train the learning model 160, the learning model 160 can be trained effectively without using negative samples or a momentum encoder.

[0048] As the learning model 160, a part or all of the encoder En can be used. In this embodiment, the backbone Bb of the encoder En is used as the learning model 160. That is, when data is input to the learning model 160, a feature vector based on the data is output by the learning model 160. The feature vector thus output corresponds to embedded data obtained by projecting the input data into the embedding space.

[0049] The type of data augmentation DA performed during training of the learning model 160 affects the factors that contribute to the uniqueness of the target data. For example, when grayscaling is used as the data augmentation DA, differences in hue contribute less to the uniqueness of the target data than when grayscaling is not used. Similarly, when left-right flipping is used as the data augmentation DA, differences in left-right orientation between image data contribute less to the uniqueness of the target data. Therefore, for example, by setting the type of data augmentation DA according to desired factors, it is possible to more effectively determine the level of uniqueness of target data in the uniqueness assessment, and more effectively present target data according to the uniqueness of the target data in the presentation assessment.

[0050] Fig. 3 is a flowchart of the pre-processing in this embodiment. For example, a predetermined input from the user Ur via the input device 106 is used as a trigger to cause the processor 101 to start the pre-processing shown in Fig. 3. The data processing method in this embodiment is realized by the pre-processing in Fig. 3 and the main processing described later.

[0051] In steps S105 to S115, the processor 101 executes a learning process for training the learning model 160 by machine learning. First, in step S105, the data generation unit 108 generates a first generated data group using the generative model 170 and stores it in the storage unit 102. In step S110, the acquisition unit 110 acquires a training dataset from the first generated data group stored in the storage unit 102. Note that the acquisition unit 110 may acquire, for example, the entire first generated data group as the training dataset. In step S115, the learning execution unit 140 executes machine learning of the learning model 160 using the training dataset acquired in step S110.

[0052] In step S120, the acquiring unit 110 acquires a reference dataset. As described above, in this embodiment, the reference dataset is included in the training dataset. Note that the acquiring unit 110 may acquire, for example, the entire training dataset as the reference dataset. In step S125, the embedded data generating unit 115 generates reference embedded data 165 using the training model 160 trained in step S115 and the reference dataset acquired in step S120. The embedded data generating unit 115 stores the reference embedded data 165 thus generated in the storage unit 102. The step of generating the reference embedded data 165, as in step S115, is also referred to as a pre-generation step. The pre-generation step is executed prior to the generation step, which will be described later.

[0053] 4 is a flowchart of the main processing in this embodiment. The main processing is started by the processor 101, for example, each time the pre-processing in FIG. 3 is completed. Note that in another embodiment, the processor 101 may start the main processing in response to a predetermined input by the user Ur via the input device 106, for example, after the pre-processing is completed.

[0054] In step S205, the data generation unit 108 generates a second generated data group using the generative model 170 and stores it in the storage unit 102. Note that the second generated data group may be generated together with the first generated data group, for example, in step S105 of FIG. 3. In step S210 of FIG. 4, the acquisition unit 110 acquires a target dataset from the second generated data group stored in the storage unit 102. It can also be said that the target data is acquired in step S205. The process of acquiring the target data, as in step S205, is also referred to as an acquisition process. Note that the acquisition unit 110 may acquire, for example, the entire second generated data group as the target dataset.

[0055] In step S215, the embedding data generation unit 115 selects one piece of data included in the target dataset acquired in step S210 as target data and inputs it to the learning model 160. As a result, target embedded data is generated. Specifically, in step S215 in this embodiment, target embedded data is generated by projecting the target data into the embedding space. The embedding data generation unit 115 stores the generated target embedded data in the storage unit 102. The step and process of generating target embedded data, such as step S215, are also referred to as a generation step and a generation process, respectively.

[0056] In step S220, the calculation unit 120 calculates the density of the target data in the embedding space using the target embedding data generated in step S215 and the reference embedding data 165 generated in step S125 of Fig. 3. That is, in step S220, the density is calculated for each target data. The step and process of calculating the density of the target data, as in step S220, are also referred to as a calculation step and a calculation process, respectively.

[0057] In step S225, the judgment condition receiving unit 135 receives a specification of a judgment condition from the user Ur. The process of receiving a specification of a judgment condition from the user Ur, as in step S225, is also referred to as a receiving process. Note that the reception of the judgment condition for the uniqueness judgment may be performed before the uniqueness judgment is started. Also, the reception of the judgment condition for the presentation judgment may be performed before the presentation judgment is started.

[0058] In steps S230 and S235, the determination unit 125 executes a determination process. First, in step S230, the determination unit 125 executes a uniqueness determination. That is, in step S230, the determination unit 125 determines whether the target data is highly unique, based on the density of the target data calculated in step S220. In step S230, the determination unit 125 determines whether the target data is highly unique, based on the result of normalizing the density of the target data, as described above. Also, in step S230, the determination unit 125 executes a uniqueness determination in accordance with the determination conditions specified in step S225.

[0059] Next, in step S235, the determination unit 125 executes a presentation determination. That is, in step S235, the determination unit 125 determines whether to present the target data to the user Ur based on the density of the target data calculated in step S220. Specifically, in step S235 in this embodiment, the determination unit 125 determines whether to present the target data to the user Ur based on the level of uniqueness of the target data determined based on the density in step S230. As described above, in step S230 in this embodiment, the level of uniqueness of the target data is determined based on the result of normalizing the density of the target data. Therefore, in step S235, it can be said that a determination is made as to whether to present the target data to the user Ur based on the result of normalizing the density of the target data. Furthermore, in step S230 in this embodiment, the level of uniqueness of the target data is determined according to the determination condition. Therefore, in step S235, it can be said that a determination is made as to whether to present the target data to the user Ur according to the determination condition.

[0060] If it is determined in step S235 that the target data should be presented, the presentation unit 130 executes a presentation process in step S240. Specifically, the presentation unit 130 presents the target data determined to be presented in the presentation determination to the user Ur as result information of the presentation determination, and also presents level information indicating the level of the determined uniqueness to the user Ur as result information of the uniqueness determination.

[0061] If it is not determined in step S235 that the target data should be presented, the determination unit 125 proceeds to step S245. That is, in this embodiment, if it is not determined that the target data should be presented, the presentation process is not executed. Note that in other embodiments, if it is not determined that the target data should be presented, a presentation process may be executed to present high / low information about the target data.

[0062] In step S245, the processor 101 determines whether data determination has been completed for all data included in the target data set. If not, the processor 101 returns the process to step S215 and generates target embedding data for the next target data. In another embodiment, the presentation process may be executed after data determination has been completed for all data included in the target data set. In this case, the presentation process may, for example, present all of the target data determined to be presented together, or present the level information for each target data together. The processor 101 may also store the presented target data and level information, for example, on a recording medium.

[0063] Fig. 5 is an explanatory diagram showing an example of a probability density function calculated by the calculation unit 120. Fig. 6 is a diagram showing an example of the result of determining the uniqueness of each piece of target data.

[0064] The probability density function in FIG. 5 and the determination result in FIG. 6 were obtained using the learning model 160, the reference embedded data 165, the generative model 170, and the target dataset prepared under the following conditions. The generative model 170 was trained to generate image data as generated data using a dataset consisting of multiple image data. Each piece of image data included in the dataset for training the generative model 170 is classified into an "airplane class." That is, the generative model 170 was trained to generate image data that approximates images in the airplane class. The reference embedded data 165 and the target dataset were obtained by inputting each piece of image data generated by the generative model 170 as reference data into the trained learning model 160. A dataset including image data generated by the generative model 170 as target data was used.

[0065] In Fig. 5, the probability density function, which is actually calculated as a function representing the density of data points in a 64-dimensional embedding space, is schematically represented in a two-dimensional XY space having mutually orthogonal X and Y coordinates. In Fig. 5, areas of the XY space with higher density are indicated by darker hatching.

[0066] FIG. 6 shows a data group Gr1 including only target data determined to have low uniqueness in the target process, and a data group Gr2 including only target data determined to have high uniqueness. That is, the density of each target data included in data group Gr1 is higher than the density of each target data included in data group Gr2. FIG. 6 shows 10 image data as examples of target data included in each data group. In FIG. 6, colored portions of pixels included in each image data that differ from the white background are hatched. All of the image data included in data group Gr1 were image data in which the colored portions were located near the vertical center of each image, more specifically, image data in which the entire body of an airplane was depicted near the vertical center. On the other hand, the image data included in data group Gr2 varied, for example, in the location and area of ​​the colored portions compared to data group Gr1. That is, the uniqueness and diversity of each data in data group Gr2 were higher than those in data group Gr1.

[0067] According to the data processing method of the present embodiment described above, the density of the target data is calculated using the target embedded data in which the target data generated by the generative model 170 is projected into the embedding space and the reference embedded data 165 in which each reference data generated by the generative model 170 is projected into the embedding space. In the determination step, a presentation determination is made based on the calculated density of the target data. In the presentation step, the target data determined to be presented in the presentation determination is presented to the user Ur. In this way, the target data is presented to the user Ur based on a density that is suitably correlated with the uniqueness of the data. Therefore, by selecting the presented target data, the user Ur can select target data based on the uniqueness of the target data.

[0068] It is known that the generated data generated by the generative model 170 often become similar to each other due to various factors. In this embodiment, even when the generated data generated by the generative model 170 become similar to each other in this way, each piece of generated data can be selected, for example, so that the uniqueness of each piece of selected generated data is further improved. Note that in other embodiments, it is also possible to select each piece of generated data so that the uniqueness of each piece of selected generated data is further reduced, for example, by performing a presentation determination so that target data with high density and low uniqueness is presented to the user Ur.

[0069] In this embodiment, in the determination step, a uniqueness determination is performed according to the calculated density of the target data, and in the presentation step, the result of the uniqueness determination, that is, the level of uniqueness, is presented to the user Ur. Therefore, the user Ur can select target data according to the uniqueness of the target data by referring to the determination result presented in the presentation step.

[0070] Furthermore, in this embodiment, the learning model 160 has been trained by self-supervised learning. By training the learning model 160 in this manner by self-supervised learning, the learning model 160 can be appropriately trained to realize data projection into an embedding space in which feature vectors of more similar data are closer in distance, without using labeled training data. In this embodiment, by using the learning model 160 thus appropriately trained, the density of the target data can be more appropriately calculated in the calculation step. As a result, the determination step and presentation step using the calculation results from the calculation step can be more appropriately executed. Furthermore, the effort required to prepare a training dataset for training the learning model 160 can be reduced. Specifically, in this embodiment, for example, compared to when the learning model 160 is trained by supervised learning, there is no need to prepare a training dataset including labeled data, and therefore the effort required to prepare the training dataset can be reduced.

[0071] In this embodiment, prior to the judgment step, a receiving step is executed in which a specification of judgment conditions is received from the user Ur. Therefore, data judgment can be executed according to the desired judgment conditions specified by the user Ur.

[0072] Furthermore, in this embodiment, the learning model 160 has been trained using a training dataset including multiple pieces of generated data generated by the generative model 170. In this way, when reference data is input to the learning model 160, the learning model 160 can more appropriately output a feature vector of the reference data, compared to, for example, a case where no generation data is used in training the learning model 160. As a result, the learning model 160 can more appropriately project the reference data into the embedding space and generate more appropriate reference embedded data 165. Therefore, the calculation step can be more appropriately performed, and the determination step and presentation step using the calculation results from the calculation step can be more appropriately performed. Furthermore, in this embodiment, the learning model 160 has been trained using at least a portion of the reference dataset, so the learning model 160 can more appropriately output a feature vector of the reference data, compared to a case where no reference dataset is used in training the learning model 160. As a result, the learning model 160 can generate more appropriate reference embedded data 165. Therefore, the calculation step can be more appropriately performed, and the determination step and presentation step using the calculation results from the calculation step can be more appropriately performed.

[0073] Furthermore, in this embodiment, in the determination step, data determination is performed based on the result of normalizing the density of the target data, which makes it possible to more easily perform data determination in the determination step.

[0074] B. Second embodiment: 7 is a flowchart of the main processing in the second embodiment. In this embodiment, unlike the first embodiment, the determination unit 125 determines in the presentation determination of the determination step that target data having a first density will be presented to the user Ur with a higher probability than target data having a second density. The second density is a density greater than the first density. The data processing device 100 in this embodiment is similar to the first embodiment in the points not specifically described.

[0075] In step S232, the determination unit 125 determines a presentation probability of presenting the target data to the user Ur depending on the level of uniqueness of the target data determined in step S230. Specifically, the determination unit 125 determines a presentation probability for target data determined to be highly unique as a first probability, and determines a presentation probability for target data determined to be low unique as a second probability. The first probability is higher than the second probability. As a result, it is determined that target data having a first density will be presented to the user Ur with a higher probability than target data having a second density. Note that in other embodiments, the determination unit 125 may determine the presentation probability for the target data depending on the density of the target data rather than the level of uniqueness of the target data.

[0076] In addition, in the reception process of step S230 in this embodiment, the judgment condition reception unit 135 may receive, as a judgment condition, a specification of a first probability or a second probability in addition to or instead of thresholds such as a first threshold or a second threshold.

[0077] In step S235b, the determination unit 125 executes the presentation determination in substantially the same manner as in step S235 in Fig. 4. However, unlike step S235, in step S235b, the determination unit 125 determines whether or not to present the target data to the user Ur in accordance with the presentation probability determined in step S232.

[0078] According to the data processing method of the present embodiment described above, in the determination step, it is determined that target data having a first density is more likely to be presented than target data having a second density. In this way, target data with higher uniqueness can be presented primarily to user Ur, while target data with lower uniqueness can also be presented to user Ur. Therefore, even when target data with higher uniqueness is presented primarily to user Ur, the possibility that target data with lower uniqueness will be wasted can be reduced.

[0079] C. Third embodiment: 8 is a flowchart of the main processing in the third embodiment. In this embodiment, unlike the first embodiment, the determination unit 125 executes only the presentation determination out of the presentation determination and the uniqueness determination in the presentation step. In the data processing device 100 in this embodiment, the points that are not particularly described are the same as those in the first embodiment.

[0080] As shown in Fig. 8, in this embodiment, after step S225, the uniqueness determination in step S230 is not performed, and instead a presentation determination in step S235c is performed. In step S235c, unlike step S235 in Fig. 4, the determination unit 125 determines whether or not to present the target data to the user Ur directly based on the density of the target data calculated in step S220, rather than on the level of uniqueness of the target data. Also, in step S235c, the determination unit 125 directly performs the presentation determination in accordance with the determination conditions specified in step S225.

[0081] According to the data processing method of the present embodiment described above, in the determination step, a presentation determination is made according to the density of the target data, and in the presentation step, the target data determined to be presented in the presentation determination is presented to the user Ur. Therefore, as in the first embodiment, target data can be selected according to the uniqueness of the target data.

[0082] In addition, even in a form in which only presentation determination is performed in the determination process as in the third embodiment, the presentation probability may be determined according to the density of the target data, and the presentation determination may be performed according to the determined presentation probability, as in the second embodiment.

[0083] D. Other Embodiments: (D-1) In each of the above embodiments, a presentation determination is performed in the determination step. However, the presentation determination need not be performed. For example, only a uniqueness determination may be performed. Even in this case, the user Ur can select target data according to the uniqueness of the target data by referring to the determination result presented in the presentation step. Furthermore, for example, a storage step may be performed in addition to or instead of the determination step and the presentation step. The storage step is a step of storing target data whose density is equal to or greater than a predetermined threshold. When the uniqueness determination and the storage step are performed, only target data determined to be highly unique may be stored in the storage step. In this case, the presentation step may or may not be performed. Furthermore, a normalized density may be used as the density in the storage step. By performing the storage step, target data is stored according to a density that is suitably correlated with the uniqueness of the data, allowing target data to be selected according to the uniqueness of the target data. Furthermore, a dataset containing a large amount of highly unique data can be created using each of the stored target data.

[0084] (D-2) In the above embodiment, a Projected GAN is used for the generative model 170, but this is not limited to this. For example, various GANs such as a Conditional GAN ​​(CGAN), a Deep Convolutional GAN ​​(DCGAN), a CycleGAN, or a StyleGAN are used for the generative model 170. Furthermore, the generative model 170 is not limited to a GAN, and for example, a Variational Autoencoder (VAE), a Flow-based Generative Model, or a Diffusion Model may also be used.

[0085] (D-3) In the above embodiment, the learning model 160 has been trained using SimSiam. In contrast, the learning model 160 may be trained using various methods other than SimSiam, as long as it has been trained to project input data into an embedding space. For example, the learning model 160 may be trained using SimCLR, BYOL, or SwAV. Furthermore, the learning model 160 may be trained using various learning methods, such as metric learning, other than SSL. In this case, the learning model 160 may be trained using, for example, unsupervised learning, supervised learning, semi-supervised learning, or reinforcement learning. Specifically, the learning model 160 may be a model trained by supervised learning or the like to perform class classification of input data.

[0086] (D-4) In each of the above embodiments, the training dataset includes the entire reference dataset. In contrast, the training dataset may include, for example, only a portion of the reference dataset, or may not include the reference dataset at all. Furthermore, all or a portion of the training dataset may be identical to all or a portion of the reference dataset. Furthermore, all or a portion of the training dataset may be identical to all or a portion of the target dataset. Furthermore, all or a portion of the reference dataset may be identical to all or a portion of the target dataset.

[0087] (D-5) In each of the above embodiments, the data processing method includes a pre-generation step, but it does not necessarily have to include a pre-generation step. In this case, for example, the reference embedding data 165 may be prepared in advance by a computer external to the data processing device 100. In this case, in step S220 of FIG. 4, the reference embedding data 165 prepared by the external computer may be used to calculate the density of the target data.

[0088] (D-6) In each of the above embodiments, the learning process is performed in the data processing device 100. However, the learning process does not have to be performed in the data processing device 100. In this case, machine learning of the learning model 160 may be performed, for example, on a computer external to the data processing device 100.

[0089] (D-7) In each of the above embodiments, data determination is performed based on the result of normalizing the density of the target data in the target process. In contrast, the uniqueness of the target data may be determined without normalizing the density of the target data.

[0090] (D-8) In each of the above embodiments, KDE is used in the calculation step. However, instead of KDE, other non-parametric density estimation methods such as histogram density estimation (HDE) or a parametric density estimation method may be used in the calculation step.

[0091] (D-9) In each of the above embodiments, the determination condition receiving unit 135 may be configured to be able to receive designation of various determination conditions, not limited to thresholds and presentation probabilities. For example, the determination condition receiving unit 135 may receive designation of whether or not to execute presentation determination according to the presentation probability, as a determination condition.

[0092] (D-10) In each of the above embodiments, the processor 101 may evaluate the diversity of the entire evaluation target dataset. For example, the processor 101 may evaluate the diversity of the evaluation target dataset based on the arithmetic mean value of the densities of multiple evaluation target data sets included in the evaluation target dataset. In this case, the processor 101 may evaluate the diversity as low if the arithmetic mean value is equal to or greater than a threshold, or may evaluate the diversity as low if the arithmetic mean value is less than the threshold. The processor 101 may also evaluate the diversity of the evaluation target dataset based on the uniqueness of multiple evaluation target data sets included in the evaluation target dataset. In this case, the processor 101 may evaluate the diversity as high if the number or proportion of evaluation target data sets determined to be highly unique is equal to or greater than a threshold, or may evaluate the diversity as low if the number or proportion is less than the threshold. The evaluation result of the diversity of the evaluation target dataset may be output using, for example, the output device 105.

[0093] E. Other forms: The present disclosure is not limited to the above-described embodiments and can be realized in various forms without departing from the spirit thereof. For example, the present disclosure can also be realized in the following forms. The technical features in the above embodiments corresponding to the technical features in each form described below can be appropriately replaced or combined to solve some or all of the problems of the present disclosure or to achieve some or all of the effects of the present disclosure. Furthermore, if a technical feature is not described as essential in this specification, it can be appropriately deleted.

[0094] (1) According to a first aspect of the present disclosure, there is provided a data processing method, comprising: a generation step of generating target embedded data in which target data generated by a generative model is input into a learning model that projects input data into an embedding space, by which the target data is projected into the embedding space; a calculation step of calculating, using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, a determination step of determining whether the uniqueness of the target data is high or low based on the density; and a presentation step of presenting the determination result of the determination step to a user. According to this aspect, the user can refer to the level of uniqueness of the target data presented in the presentation step, and can therefore select target data according to the uniqueness of the target data.

[0095] (2) According to a second aspect of the present disclosure, there is provided a data processing method, comprising: a generating step of generating target embedded data in which target data generated by a generative model is input into a learning model that projects input data into an embedding space, by which the target data is projected into the embedding space; a calculating step of calculating, using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, a determining step of determining whether to present the target data to a user based on the density; and a presenting step of presenting the target data to the user when it is determined in the determining step that the target data should be presented. According to this aspect, target data is presented to the user according to a density that is suitably correlated with the uniqueness of the data, and thus target data can be selected according to the uniqueness of the target data.

[0096] (3) According to a third aspect of the present disclosure, there is provided a data processing method, comprising: a generating step of generating target embedded data in which target data generated by a generative model is projected into an embedding space by inputting the target data into a learning model that projects input data into the embedding space, a calculating step of calculating, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model are projected into the embedding space, a density of the target data in the embedding space into which each of the reference data is projected, and a storing step of storing the target data whose density is equal to or greater than a predetermined threshold. According to this aspect, the target data is stored according to a density that is suitably correlated with the uniqueness of the data, and therefore the target data can be selected according to the uniqueness of the target data.

[0097] (4) In the above embodiment, the learning model may have been trained by self-supervised learning. According to this embodiment, by using a learning model that has been properly trained to realize data projection into an embedding space in which feature vectors of more similar data are closer in distance, the density of the target data can be calculated more appropriately in the calculation step. Therefore, subsequent steps that use the calculation results from the calculation step can be executed more appropriately.

[0098] (5) In the above embodiment, a receiving step may be provided for receiving, from the user, a designation of a determination condition for the determination step prior to the determination step. According to this embodiment, data determination can be performed according to the desired determination condition designated by the user.

[0099] (6) In the above embodiment, the determining step may determine that the target data having a first density is to be presented with a higher probability than the target data having a second density greater than the first density. According to this embodiment, while primarily presenting more unique target data to the user, it is also possible to present less unique target data to the user. Therefore, even when primarily presenting more unique target data to the user, it is possible to reduce the possibility that less unique target data will be wasted.

[0100] (7) In the above aspect, the learning model may have been trained using a training dataset including a plurality of pieces of data generated by the generative model. According to this aspect, the calculation step can be more appropriately executed, and subsequent steps using the calculation results from the calculation step can be more appropriately executed.

[0101] (8) In the above embodiment, the training data set may include at least a portion of each of the reference data. According to this embodiment, more appropriate reference embedding data can be generated by the training model. Therefore, the calculation step can be more appropriately performed, and subsequent steps using the calculation results from the calculation step can be more appropriately performed.

[0102] (9) In the above aspect, the determination step may be performed based on a result of normalizing the density. According to this aspect, the determination step can be performed more easily.

[0103] (10) In the above embodiment, the presenting step may present information indicating the level of uniqueness of the target data determined to be presented in the determining step. According to this embodiment, the user can obtain the target data and can confirm the level of uniqueness of the target data.

[0104] (11) According to a fourth aspect of the present disclosure, there is provided a data processing device. The data processing device includes: a learning model that projects features of input data into an embedding space; a memory unit that stores reference embedded data in which multiple reference data generated by a generative model are projected into the embedding space; and a processor. The processor executes the following operations: an acquisition process that acquires target data generated by the generative model; a generation process that inputs the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process that uses the target embedded data and the reference embedded data to calculate the density of the target data in the embedding space into which each reference data is projected; a determination process that determines whether the target data is highly unique based on the density; and a process that presents the determination result of the determination process to a user.

[0105] (12) According to a fifth aspect of the present disclosure, there is provided a data processing device. The data processing device includes: a learning model that projects features of input data into an embedding space; a memory unit that stores reference embedded data in which multiple reference data generated by a generative model are projected into the embedding space; and a processor. The processor executes the following operations: an acquisition process that acquires target data generated by the generative model; a generation process that inputs the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process that uses the target embedded data and the reference embedded data to calculate a density of the target data in the embedding space into which each reference data is projected; a determination process that determines whether to present the target data to a user based on the density; and a process that presents the target data to the user if the determination process determines to present the target data.

[0106] (13) According to a sixth aspect of the present disclosure, there is provided a data processing device. The data processing device includes: a learning model that projects features of input data into an embedding space; a memory unit that stores reference embedded data in which multiple reference data generated by a generative model are projected into the embedding space; and a processor. The processor executes the following operations: an acquisition process that acquires target data generated by the generative model; a generation process that inputs the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process that uses the target embedded data and the reference embedded data to calculate a density of the target data in the embedding space into which each reference data is projected; and a process of saving the target data whose density is equal to or greater than a predetermined threshold.

[0107] (14) According to a seventh aspect of the present disclosure, there is provided a program that causes a computer to implement the following functions: (1) inputting target data generated by a generative model into a learning model that projects features of the input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; (2) using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, (3) calculating the density of the target data in the embedding space into which each of the reference data is projected, (4) determining whether the target data is highly unique based on the density, and (5) presenting the result of the determination made by the determination function to a user.

[0108] (15) According to an eighth aspect of the present disclosure, there is provided a program that causes a computer to implement the following functions: a function of generating target embedded data in which target data generated by a generative model is projected into an embedding space by inputting the target data into a learning model that projects features of the input data into the embedding space; a function of calculating, using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, a function of determining whether to present the target data to a user based on the density; and a function of presenting the target data to the user when the determination function determines that the target data should be presented.

[0109] (16) According to a ninth aspect of the present disclosure, there is provided a program that causes a computer to perform the following functions: inputting target data generated by a generative model into a learning model that projects features of the input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; using the target embedded data and reference embedded data in which multiple reference data generated by the generative model are projected into the embedding space, calculate a density of the target data in the embedding space into which each of the reference data is projected; and saving the target data whose density is equal to or greater than a predetermined threshold.

[0110] In addition to the above-described embodiments, the present disclosure can be realized in the form of, for example, a data processing system, a non-transitory recording medium on which a program is recorded, a program product, etc. Note that the program product may be provided as a recording medium on which a program is recorded, or may be provided as a program product that can be distributed via a network, for example. [Explanation of symbols]

[0111] 50...data processing system, 100...data processing device, 101...processor, 102...memory unit, 103...input / output interface, 104...internal bus, 105...output device, 106...input device, 108...data generation unit, 110...acquisition unit, 115...embedded data generation unit, 120...calculation unit, 125...judgment unit, 130...presentation unit, 135...judgment condition reception unit, 140...learning execution unit, 155...program, 160...learning model, 165...reference embedded data, 170...generative model

Claims

1. a generation step of inputting target data generated by a generative model into a learning model that projects input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; a calculation step of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model is projected into the embedding space; a determination step of determining whether the uniqueness of the target data is high or low according to the density; a presentation step of presenting to a user the high / low determination result obtained by the determination step.

2. a generation step of inputting target data generated by a generative model into a learning model that projects input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; a calculation step of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model is projected into the embedding space; a determination step of determining whether or not to present the target data to a user according to the density; a presentation step of presenting the target data to the user when it is determined in the determination step that the target data should be presented.

3. a generation step of inputting target data generated by a generative model into a learning model that projects input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; a calculation step of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedded data and reference embedded data in which a plurality of reference data generated by the generative model is projected into the embedding space; a storing step of storing the target data whose density is equal to or greater than a predetermined threshold.

4. 4. A data processing method according to claim 1, further comprising: A data processing method, wherein the learning model has been trained by self-supervised learning.

5. 3. A data processing method according to claim 1 or 2, The data processing method further comprises, prior to the determining step, a receiving step of receiving, from the user, a designation of a determination condition for the determining step.

6. 3. A data processing method according to claim 2, A data processing method in which, in the determination step, it is determined that the target data having a first density will be presented with a higher probability than the target data having a second density that is greater than the first density.

7. 4. A data processing method according to claim 1, further comprising: A data processing method in which the learning model has been trained using a training dataset including multiple pieces of data generated by the generative model.

8. 8. A data processing method according to claim 7, A data processing method, wherein the training dataset includes at least a portion of each of the reference data.

9. 3. A data processing method according to claim 1 or 2, A data processing method, wherein in the determining step, the determination is made depending on a result of normalizing the density.

10. 3. A data processing method according to claim 2, A data processing method, wherein the presentation step presents information indicating the level of uniqueness of the target data determined to be presented in the determination step.

11. a memory unit that stores a learning model that projects the features of input data into an embedding space, and reference embedding data in which a plurality of reference data generated by a generative model are projected into the embedding space; a processor; The processor: an acquisition process for acquiring target data generated by the generative model; a generation process of inputting the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedding data and the reference embedding data; a determination process for determining whether the target data is highly unique or not according to the density; and executing a process of presenting the high / low determination result obtained by the determination process to a user. Data processing device.

12. a memory unit that stores a learning model that projects the features of input data into an embedding space, and reference embedding data in which a plurality of reference data generated by a generative model are projected into the embedding space; a processor; The processor: an acquisition process for acquiring target data generated by the generative model; a generation process of inputting the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedding data and the reference embedding data; a determination process for determining whether or not to present the target data to a user according to the density; and if it is determined in the determination process that the target data should be presented, executing a process of presenting the target data to the user. Data processing device.

13. a memory unit that stores a learning model that projects the features of input data into an embedding space, and reference embedding data in which a plurality of reference data generated by a generative model are projected into the embedding space; a processor; The processor: an acquisition process for acquiring target data generated by the generative model; a generation process of inputting the target data into the learning model to generate target embedded data in which the target data is projected into the embedding space by the learning model; a calculation process of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedding data and the reference embedding data; and storing the target data whose density is equal to or greater than a predetermined threshold. Data processing device.

14. A function of inputting target data generated by a generative model into a learning model that projects the features of input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; a function of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedding data and reference embedding data in which a plurality of reference data generated by the generative model is projected into the embedding space; a determination function for determining the level of uniqueness of the target data according to the density; and a function of presenting the result of the high / low judgment made by the judgment function to a user.

15. A function of inputting target data generated by a generative model into a learning model that projects the features of input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; a function of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedding data and reference embedding data in which a plurality of reference data generated by the generative model is projected into the embedding space; a determination function that determines whether or not to present the target data to a user according to the density; and a function of presenting the target data to the user when the determination function determines that the target data should be presented.

16. A function of inputting target data generated by a generative model into a learning model that projects the features of input data into an embedding space, thereby generating target embedded data in which the target data is projected into the embedding space; a function of calculating a density of the target data in the embedding space into which each of the reference data is projected, using the target embedding data and reference embedding data in which a plurality of reference data generated by the generative model is projected into the embedding space; and a function of storing the target data whose density is equal to or greater than a predetermined threshold.

Images