Information processing device, information processing method, and program

Abstract

This information processing device 100 comprises a data acquisition unit 121 that acquires use data that has been used by a prediction model, an analysis acquisition unit 122 that acquires the results of preset analysis of the prediction model or the use data, and a generation unit 123 that, in accordance with the analysis results, generates retraining use data that is based on the use data and is to be used for retraining of the prediction model. The present invention can thereby: generate use data that can be effectively used for machine learning–based retraining of a prediction model such as is used, for example, in the healthcare field; and assist decision making by a user with respect to prediction.

Description

Information processing device, information processing method, and program

[0001] The present disclosure relates to an information processing device, an information processing method, and a program.

[0002] A prediction model generated by machine learning is used to predict an output for new input data. As described in Patent Literature 1, the prediction model must be re-trained to maintain its prediction accuracy.

[0003] JP 2016-143354 A

[0004] However, when re-learning a predictive model, data to be used for re-learning, such as training data and various parameters, is required, but there is a problem in that it is difficult to generate data that is effective for re-learning.

[0005] Therefore, an object of the present disclosure is to solve the above-mentioned problem that it is difficult to generate data that is effective for relearning a prediction model.

[0006] An information processing device according to an embodiment of the present disclosure includes: a data acquisition unit that acquires usage data used in a prediction model; an analysis acquisition unit that acquires predetermined analysis results for the prediction model or the usage data; and a generation unit that generates re-learned usage data to be used for re-learning the prediction model based on the usage data in accordance with the analysis results. An information processing method according to an embodiment of the present disclosure includes: acquiring usage data used in the prediction model; acquiring predetermined analysis results for the prediction model or the usage data; and generating re-learned usage data to be used for re-learning the prediction model based on the usage data in accordance with the analysis results. A program according to an embodiment of the present disclosure includes: causing a computer to execute the following processes: acquiring usage data used in the prediction model; acquiring predetermined analysis results for the prediction model or the usage data; and generating re-learned usage data to be used for re-learning the prediction model based on the usage data in accordance with the analysis results.

[0007] With the above-described configuration, the present disclosure can generate usage data that is effective for relearning a prediction model.

[0008] FIG. 1 is a block diagram showing a configuration of an information processing device according to the present disclosure; FIG. 2 is a diagram showing an overview of processing by an information processing device according to the present disclosure; FIG. 3 is a flowchart showing an operation of an information processing device according to the present disclosure; FIG. 4 is a block diagram showing a hardware configuration of an information processing device according to the present disclosure; and FIG. 5 is a block diagram showing a configuration of an information processing device according to the present disclosure.

[0009] First Embodiment A first embodiment of the present disclosure will be described with reference to the drawings. Note that the drawings may be relevant to any embodiment.

[0010] The information processing device 10 in this embodiment generates relearning information (relearning data) used when relearning a predictive model generated by machine learning. Here, the predictive model is, for example, a machine learning model generated by supervised learning, and is generated by supervised learning of training data (x, y) consisting of a set of explanatory variable x and target variable y. The predictive model is configured to output a predicted value y when an unknown explanatory variable x is input. The relearning information generated in this embodiment includes relearning data, which is training data used to retrain the predictive model through machine learning, weights assigned to the relearning data, and hyperparameters set for the predictive model. The configuration and operation of the information processing device 10 are described in detail below.

[0011] The information processing device 10 is composed of one or more information processing devices each including a calculation device and a storage device. As shown in FIG. 1 , the information processing device 10 includes a reference information acquisition unit 11, an analysis information acquisition unit 12, and a relearning information generation unit 13. The functions of the reference information acquisition unit 11, the analysis information acquisition unit 12, and the relearning information generation unit 13 can be realized by the calculation device executing a program for realizing each function stored in the storage device. The information processing device 10 also includes an acquired information storage unit 15. The acquired information storage unit 15 is composed of a storage device. A reference information storage device 20 and an analysis information storage device 30 are connected to the information processing device 10.

[0012] The reference information acquisition unit 11 (data acquisition unit) acquires reference information, which is data used in the prediction model to be retrained (step S1 in FIG. 3). The reference information includes sample data D, which is training data used during machine learning of the prediction model and operational data used during operation of the prediction model. For example, the sample data D is expressed by the following formula 1 and consists of samples (x_i, y_i), which are sets of explanatory variable x and objective variable y. Note that the sample data D may consist of only training data, only operational data, or even a portion of the training data and operational data. However, the sample data D is not limited to the training data and operational data used in the current prediction model, but may also include even older data, such as data used in previous prediction models or data referenced as important samples.

[0013] The reference information also includes weight data w (setting data) assigned to the above-mentioned sample data D. The weight data w is a weight value assigned to the explanatory variable x and the objective variable y, which are samples included in the sample data D, when the sample data D is used to train a prediction model. The reference information also includes hyperparameters p (setting data) set in the prediction model. The hyperparameters p are data that are or can be set in the prediction model, such as the number of epochs representing the number of times a single training data set is trained when machine learning a prediction model, the depth of a hierarchy in a tree-structured prediction model, and the type of explanatory variable used.

[0014] The reference information including the above-mentioned sample data D, weight data w, and hyperparameter p is stored in advance in the reference information storage device 20, and is acquired by the reference information acquisition unit 11 of the information processing device 10 and stored in the acquired information storage unit 15 within the information processing device 10. However, the reference information may be acquired from any device, or may be stored in advance in the acquired information storage unit 15 within the information processing device 10.

[0015] The analytical information acquisition unit 12 (analysis acquisition unit) acquires analytical information S representing a preset analysis result for a prediction model or reference information (step S2 in FIG. 3 ). In particular, in this embodiment, the analytical information S is the analysis result for a prediction by a prediction model, for example, the analytical information S representing the analysis result of the factors that caused a prediction error. At this time, the analytical information is output from an analysis device configured to analyze the prediction model and generate analytical information S and stored in an analytical information storage device 30, and the analytical information acquisition unit 12 of the information processing device 10 acquires the analytical information S from the analytical information storage device 30. However, the analytical information acquisition unit 12 may acquire the analytical information S from the analysis device.

[0016] Here, a specific example of the analysis information S will be described. For example, the analysis information S ｉ is the sample (x_ ｉ , y_ ｉ ) characteristics. For example, ｉ is an abnormal value," "y_ ｉis an abnormal value (label noise)," "x_ ｉ "A change in distribution (concept drift) is occurring in the vicinity," "x_ ｉ Analysis information s that represent the analysis results such as "under-learning in the neighborhood" ｉ For example, analytical information s ｊ The characteristic of the subset Dj of sample data D used when a prediction error occurs in the prediction model is an example of the characteristic of the subset Dj of sample data D used when a prediction error occurs in the prediction model. ｊ "The explanatory variable of is an outlier," "D ｊ Analysis information that shows the analysis results, such as "The predictive model has low performance in ｊ For example, analytical information s ｋ The sample extraction condition c in the sample data D used when the prediction error occurred by the prediction model is ｋ For example, analytical information s that shows the analysis results such as "distribution change occurs in the area where the explanatory variable day of the week is Sunday and the weather is sunny" ｋ Examples include:

[0017] Furthermore, for example, the analytical information s may be characteristics of the prediction model according to the prediction result, as an analysis result of the prediction by the prediction model. Examples of analytical information s include "if the performance on the test data is lower than the performance on the training data, it is overfitting," "if the prediction model has a tree structure and is overfitted, the tree hierarchy is too deep," "if the prediction model is a neural network and is overfitted, the number of epochs is too large," "if the distribution variance is large (when k-neighbors are extracted, there are many samples with large variance of the objective variable y in the k-neighborhood), there are insufficient explanatory variables," and "if there are insufficient explanatory variables, and based on the attributes of the data being handled (e.g., weather), it is necessary to add an explanatory variable for that attribute (e.g., temperature)."

[0018] As described above, the analytical information S includes analytical information s consisting of multiple different analysis results, and the analytical information acquisition unit 12 acquires the analytical information s consisting of multiple different analysis results. Each analytical information s can be generated using various existing analytical techniques, such as the prediction error factor analysis technique disclosed in International Publication No. 2022 / 180749, and each analytical information s can be acquired from an analytical device that performs such analysis. However, the information processing device 10 itself may analyze the prediction model or usage data as described above and acquire each analytical information s. Furthermore, the analytical information s is not limited to information representing the above-described analysis results, and may also be information representing the analysis results of different prediction models or reference information.

[0019] The relearning information generator 13 (generator) generates relearning use data to be used for relearning a prediction model based on reference information, in accordance with the acquired analysis information S as described above. At this time, the relearning information generator 13 performs a correction process on the reference information to generate the relearning use data. The relearning information generator 13 determines each correction process for the reference information as a candidate process in accordance with each piece of analysis information s consisting of different analysis results acquired as described above (step S3 in FIG. 3 ), and then integrates the multiple candidate processes to determine a final correction process (step S4 in FIG. 3 ). The relearning use data generated includes relearning data to be used when relearning a prediction model, weight data w to be assigned to the relearning data when relearning a prediction model, and hyperparameters p to be set for the prediction model to be relearned.

[0020] Here, a specific example of a process performed by the relearning information generating unit 13 to determine candidate processes for correction of reference information according to each analysis information s will be described. For example, when the analysis information s is a sample (x_ ｉ , y_ ｉ ) characteristics,” correction processes such as “excluding (deleting) the sample from the re-learning data” and “setting the value of weight data w to be included and assigned in the re-learning data” are determined as processing candidates for the sample.ｉ is an abnormal value (label noise)" ｉ , y_ ｉ As an example, when the analysis information s is "x_ ｉ "There is a change in distribution (concept drift) in the vicinity" ｉ As an example, when the analytical information s is "x_ ｉ If "x_ is under-trained in the neighborhood," ｉ The modification process of "increasing the value of weight data w assigned to nearby samples" is determined as a candidate process.

[0021] As another specific example, if the analytical information s is "characteristics of the prediction model," a correction process such as "setting a hyperparameter p of the prediction model" is determined as a candidate process. As one example, if the analytical information s is "the hierarchical depth of the tree structure of the prediction model is too deep," a correction process such as "reducing the maximum hierarchical depth of the tree structure" is determined as a candidate process. As another example, if the analytical information s is "the number of epochs of the prediction model is too large," a correction process such as "reducing the number of epochs" is determined as a candidate process. As another example, if the analytical information s is "it is necessary to add an explanatory variable (e.g., temperature)," a correction process such as "adding an explanatory variable (e.g., temperature)" is determined as a candidate process.

[0022] The above-described process of determining candidate processes according to the analysis information s by the re-learning information generating unit 13 is an example, and different correction processes may be determined as candidate processes according to the content of the analysis information s.

[0023] Then, the relearning information generator 13 integrates the candidate processes determined as described above to determine a final correction process. At this time, when multiple candidate processes are determined, the relearning information generator 13 determines all or some of the multiple candidate processes as the final correction processes. For example, when multiple candidate correction processes are determined for the same sample (x_i, y_i) and the contents of these correction processes conflict, the relearning information generator 13 selects one of the correction processes to determine the final correction process.

[0024] As an example, "processing candidate 1" is "sample (x_ ｉ , y_ ｉ ) and "Processing candidate 2" are set to "x_ ｉ "Exclude nearby samples", "Processing candidate 3" and "x_ ｉ Three candidate processes, such as "double the value of weight data w assigned to nearby samples," are applied to the same sample (x_ ｉ , y_ ｉ ) is determined for the sample (x_ ｉ , y_ ｉ Regarding the above, candidate process 1 and candidate process 2 can be applied simultaneously because they have the same content, but candidate process 1, 2 and candidate process 3 cannot be applied simultaneously because the content of the process conflicts, with one being excluded and the other remaining and weighted. ｉ For the neighboring sample group, the candidate processes 2 and 3 cannot be simultaneously applied. ｉ , y_ ｉ ) and sample group (x_ ｉ For each sample (x_neighborhood), one of the candidate processes is determined as the final correction process. At this time, it is assumed that the candidate processes are prioritized according to their processing content. For example, it is assumed that the priority is set in the order of "candidate process 1 (exclusion of individual samples) > candidate process 2 (exclusion of sample groups) > candidate process 3 (weighting of samples)." Therefore, in the above example, the relearning information generator 13 determines the final correction process for the sample (x_ ｉ , y_ ｉ), "Processing Candidate 1 (Exclusion)" is determined as the final correction process, and the sample group (x_ ｉ For the candidate processes (neighborhood), "candidate process 2 (exclusion)" is determined as the final modification process. Note that the method for determining the final modification process from the candidate processes is not limited to the above-described method, and may be determined according to any standard (rule).

[0025] The relearning information generator 13 then performs the final correction process on the reference information to generate data to be used for relearning. For example, as described above, if the process to be excluded from sample data D is determined as the final correction process, as shown in FIG. 2 , sample d determined to be excluded as the final correction process is excluded from sample data D consisting of training data and operational data (e.g., the blank area in FIG. 2 ), and the remaining samples are generated as relearning data. Note that, if multiple candidate processes are not mutually exclusive, the relearning information generator 13 may determine the multiple candidate processes as the final correction processes and perform such correction processes on the reference information to generate data to be used for relearning. For example, weight data w having a set value may be assigned to the samples to generate relearning data, or a hyperparameter p may be set for the prediction model to be used for relearning.

[0026] The relearning information generator 13 outputs relearning use data, such as the relearning data generated as described above, to the learning device. As a result, the learning device re-learns the prediction model using the relearning use data. For example, the learning device generates a relearned prediction model by performing machine learning on a prediction model in which the hyperparameter p generated as the relearning use data is set, using the relearning use data generated as the relearning use data and the relearning use data to which the value of the weight data w generated as the relearning use data is assigned.

[0027] As described above, the information processing device 10 of this embodiment generates retraining use data by performing a correction process on the reference information in accordance with the results of a predetermined analysis of the prediction model or the usage data. For example, it is possible to generate retraining data by excluding certain samples from the sample data, which are the training data or the operational data, or by assigning weights to certain samples. This allows samples deemed unnecessary based on the analysis results to be excluded from the retraining data, even if the operational data, such as the training data shown in FIG. 2, was used in the prediction model at a time close to the training data. Furthermore, even if the training data, such as the training data shown in FIG. 2, was used in the prediction model at a time farther (older) than the operational data, samples deemed necessary based on the analysis results can be used as retraining data without being excluded, and can even be weighted to emphasize them. Furthermore, it is possible to set the hyperparameters of the prediction model during retraining. As a result, it is possible to generate usage data that can be effectively used for retraining the prediction model. Furthermore, outputting such information can effectively support user decision-making.

[0028] Here, as an example of an application of the information processing device 10 described above, a use example in the healthcare field will be described. First, assume that the "explanatory variable x" of sample (x, y) included in sample data D is "weather, temperature, day of the week, disease name, most recent bed occupancy rate, and holiday weekday label," the "target variable y" is "bed occupancy rate one week later," and the prediction model is "a model that predicts bed occupancy rates in a hospital one week later." Then, assume that the information processing device 10 acquires analysis information s as a result of analyzing the prediction model, such as "a prediction error caused by underlearning was discovered on sunny days with high temperatures" and "a distribution change occurred such that the bed occupancy rate of infectious disease patients on weekdays increased." Then, the information processing device 10 generates re-learning use data to be used for re-learning the prediction model based on the acquired analysis results. In this case, the information processing device 10 generates the re-learning data by performing correction processing on the samples, such as "increasing the sample weight of data on sunny days with high temperatures" or "excluding samples of infectious disease patients on weekdays." This enables the information processing device 10 to generate usage data that can be effectively used for relearning the prediction model through machine learning, thereby supporting the decision-making of hospital staff regarding bed management.

[0029] Second Embodiment Next, a second embodiment of the present disclosure will be described with reference to the drawings. Note that the drawings may be relevant to any of the embodiments.

[0030] The information processing device 10 of this embodiment further has the following functions in addition to the functions of the information processing device 10 of the first embodiment described above.

[0031] First, the reference information acquisition unit 11 (data acquisition unit) acquires sample data D consisting of learning data and operational data as shown in Fig. 2 as reference information. At this time, the learning data is data used when a prediction model is machine-learned, and the operational data is data input to the machine-learned prediction model and predicted. Therefore, it can be said that the group of learning data and the group of operational data belong to groups that were used for different times in the prediction model. For this reason, the reference information acquisition unit 11 converts the group of operational data into new group data D. １(first group data), a group of learning data is called old group data D ２ However, the reference information acquiring unit 11 divides the learning data and the operational data at an arbitrary time interval used in the prediction model, and acquires new group data D １ and old group data D ２ In addition, new group data D １ and old group data D ２ The samples (x_i, y_i) belonging to the vectors x_i and y_i are expressed by the following equation (2).

[0032] The analysis information acquisition unit 12 (analysis acquisition unit) acquires the new group data D １ and old group data D ２ The analytical information s representing the analytical results based on the comparison of each sample between the new group data D １ The samples belonging to the old group data D ２ As an example, the analysis information s is an analysis result that indicates the ratio of samples belonging to "D １ and D ２ The density ratio function r of (x, y) ｘｙ (x, y)" and "a set of samples (x, y) is used as a new group data D １ Probability of appearing in old group data D ２ The analytical information s is the probability of x appearing in the ｘ (x)" and "a sample (x) is added to the new group data D １ Probability of appearing in old group data D ２ The probability that a certain event will occur in a certain period (or its approximation or estimate).

[0033] The analytical information acquisition unit 12 acquires the above-mentioned analytical information s, which is the "density ratio function r ｘｙ (x, y)" and "density ratio r ｘ (x)" may be acquired from another storage device, and the acquired new group data D １ and old group data D２ It may be calculated and obtained from

[0034] The re-learning information generating unit 13 (generating unit) generates re-learning use data to be used for re-learning the prediction model based on the reference information, according to the analysis information s acquired as described above. Specifically, in this embodiment, the re-learning information generating unit 13 generates new group data D １ and old group data D ２ For the sample data D belonging to the above, the re-learning information generator 13 determines, as a candidate process, a modification process such as "setting the value of weight data w to be included in and assigned to the re-learning data." As an example, the re-learning information generator 13 determines the following three candidate processes:

[0035] First, as a candidate process 1, new group data D １ For samples belonging to the group, the weight data w to be assigned is set to "(1-α / n １ In other words, in the process candidate 1, the new group data D １ The old group data D is included in the re-learning data. ２ For samples (x, y) belonging to ｘｙ (x, y) / n ２ In other words, in the process candidate 2, the new group data D １ Old group data similar to D ２ In addition, as a processing candidate 3, an even older group data D ２ For samples (x, y) belonging to ｘ If (x)≦τ, the value of the weight data w to be assigned is set to the value shown in the following formula 3. That is, in the process candidate 3, the old group data D ２ is the new group data D １ Although it does not appear in the １ Old group data D with explanatory variables that do not appear in ２ is included in the retraining data. Note that α, τ, and ε are set as hyperparameters according to the user's wishes and usage conditions.

[0036] Then, the relearning information generator 13 integrates the candidate processes generated as described above to determine a final modification process. In this embodiment, when multiple candidate modification processes are determined for the same sample (x, y), that is, when the above-described candidate processes 2 and 3 are determined for the same sample, the relearning information generator 13 adds up the respective weight values ​​and assigns them.

[0037] Thereafter, the re-learning information generating unit 13 performs the final correction process on the reference information and generates the re-learning use data. １ and old group data D ２ The weight data w is assigned to the samples belonging to the new group data D １ and old group data D ２ In this example, weights are assigned to samples according to the analysis results obtained by comparing them with the data to generate re-learning data. However, other data to be used for re-learning may be generated by deleting some samples according to the analysis results described above, or by setting hyperparameters to be set in a predictive model.

[0038] As described above, the information processing device 10 of this embodiment generates re-learned usage data by performing a correction process based on a preset analysis result on the usage data used in the prediction model. For example, a weight for inclusion in the re-learned data is set based on a comparison between the learning data grouped according to the time used in the prediction model and the operational data. As a result, for example, the above-mentioned candidate process 1 can emphasize new group data as re-learned data, the above-mentioned candidate process 2 can use data similar to the new group data for old group data as re-learned data, and the above-mentioned candidate process 3 can use useful data from old group data that does not appear in the new group data but may appear in the future as re-learned data.

[0039] Third Embodiment Next, a third embodiment of the present disclosure will be described with reference to the drawings. This embodiment shows an outline of the configuration of the information processing device described in the above-mentioned embodiment. Note that Figures 4 and 5 are diagrams for explaining the configuration, and these diagrams may be relevant to any of the embodiments.

[0040] First, the hardware configuration of the information processing device 100 will be described with reference to Fig. 4. The information processing device 100 is configured as a general information processing device, and is equipped with the following hardware configuration, for example: CPU (Central Processing Unit) 101 (arithmetic unit); ROM (Read Only Memory) 102 (storage device); RAM (Random Access Memory) 103 (storage device); programs 104 loaded into RAM 103; storage device 105 storing programs 104; drive device 106 for reading and writing data from and to a storage medium 110 external to the information processing device; communication interface 107 for connecting to a communication network 111 external to the information processing device; input / output interface 108 for inputting and outputting data; and bus 109 for connecting the various components.

[0041] 4 shows an example of the hardware configuration of the information processing device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with only a part of the above-described configuration, such as excluding the drive device 106. Furthermore, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof, instead of the above-described CPU.

[0042] The information processing device 100 can be equipped with the data acquisition unit 121, analysis acquisition unit 122, and generation unit 123 shown in FIG. 5 by having the CPU 101 acquire and execute the program group 104. The program group 104 is stored in advance in the storage device 105 or the ROM 102, for example, and is loaded into the RAM 103 and executed by the CPU 101 as needed. The program group 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, and the drive device 106 may read and supply the programs to the CPU 101. However, the data acquisition unit 121, analysis acquisition unit 122, and generation unit 123 described above may be constructed using dedicated electronic circuits for realizing such means.

[0043] The data acquisition unit 121 acquires usage data used in the prediction model. The analysis acquisition unit 122 acquires a preset analysis result for the prediction model or the usage data. The generation unit 123 generates re-learning usage data to be used for re-learning the prediction model based on the usage data according to the analysis result.

[0044] With the above-described configuration, the present disclosure can generate retrained usage data to be used for retraining a prediction model based on usage data in accordance with a predetermined analysis result of the prediction model or usage data, thereby generating usage data that can be effectively used for retraining a prediction model.

[0045] In addition, at least one or more of the functions of the above-mentioned data acquisition unit 121, analysis acquisition unit 122, and generation unit 123 may be executed by an information processing device installed and connected anywhere on the network, that is, they may be executed by so-called cloud computing.

[0046] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-RWs, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can be supplied to a computer via wired communication paths such as electric wires and optical fibers, or via wireless communication paths.

[0047] Although the present disclosure has been described above with reference to the above-described embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0048] <Supplementary Notes> Some or all of the above embodiments may be described as in the following supplementary notes. The following provides an overview of the configurations of an information processing device, an information processing method, and a program according to the present disclosure. However, the present disclosure is not limited to the following configurations. (Supplementary Note 1) An information processing device comprising: a data acquisition unit that acquires usage data used in a prediction model; an analysis acquisition unit that acquires predetermined analysis results for the prediction model or the usage data; and a generation unit that generates re-learned usage data to be used for re-learning the prediction model based on the usage data according to the analysis results. (Supplementary Note 2) The information processing device according to Supplementary Note 1, wherein the generation unit performs a correction process on the usage data according to the analysis results to generate the re-learned usage data. (Supplementary Note 3) The information processing device according to Supplementary Note 2, wherein the analysis acquisition unit acquires a plurality of different analysis results, and the generation unit sets candidate correction processes for the usage data according to each of the analysis results, and performs a final correction process determined based on the set candidate correction processes on the usage data to generate the re-learned usage data. (Supplementary Note 4) The information processing device according to Supplementary Note 1, wherein the analysis and acquisition unit acquires the analysis result for a prediction made by the prediction model using the usage data. (Supplementary Note 5) The information processing device according to Supplementary Note 4, wherein the analysis and acquisition unit acquires the analysis result obtained by analyzing factors that caused a prediction error in a prediction made by the prediction model using the usage data. (Supplementary Note 6) The information processing device according to Supplementary Note 5, wherein the data acquisition unit acquires, as the usage data, learning data used when training the prediction model and / or operation data used during operation, and the analysis and acquisition unit acquires, as the analysis result, characteristics of the usage data as factors that caused a prediction error made by the prediction model using the usage data. (Supplementary Note 7) The information processing device according to Supplementary Note 6, wherein the generation unit performs a correction process on the usage data in accordance with the characteristics of the usage data, and generates re-learned data to be used when re-learning the prediction model as the re-learned usage data.(Supplementary Note 8) The information processing device according to Supplementary Note 7, wherein the generation unit performs a correction process of deleting a portion of the usage data or assigning a weight to the portion of the usage data according to characteristics of the usage data, thereby generating the re-learned data. (Supplementary Note 9) The information processing device according to Supplementary Note 1, wherein the data acquisition unit acquires, as the usage data, learning data used when learning the prediction model and / or operation data used during operation, or setting data of the prediction model, and the generation unit performs at least one of correction processes of deleting a portion of the learning data and / or operation data, assigning a weight to the learning data and / or operation data, or setting new setting data according to a result of the analysis, thereby generating the re-learned usage data. (Supplementary Note 10) The information processing device according to Supplementary Note 1, wherein the data acquisition unit acquires as the usage data, from among learning data used when training the prediction model and / or operation data used during operation, data belonging to two groups having different times used in the prediction model, the analysis and acquisition unit acquires the analysis result based on a comparison of the usage data between the groups, and the generation unit performs a correction process on the usage data according to the analysis result to generate the re-learned usage data. (Supplementary Note 11) The information processing device according to Supplementary Note 10, wherein the analysis and acquisition unit acquires as the analysis result a ratio of usage data belonging to a second group having a time used in the prediction model farther away than the first group to appear in usage data belonging to a first group having a close time used in the prediction model, and the generation unit performs a correction process to assign a weight to the usage data according to the analysis result to generate the re-learned usage data. (Supplementary Note 12) An information processing method comprising: acquiring usage data used in a prediction model; acquiring a predetermined analysis result for the prediction model or the usage data; and generating re-learning usage data to be used for re-learning the prediction model based on the usage data according to the analysis result.(Supplementary Note 13) The information processing method according to Supplementary Note 12, comprising performing a correction process on the usage data in accordance with the analysis results to generate the re-learned usage data. (Supplementary Note 14) The information processing method according to Supplementary Note 13, comprising obtaining a plurality of different analysis results, setting candidate correction processes for the usage data in accordance with each of the analysis results, and performing a final correction process determined based on the set candidate correction processes on the usage data to generate the re-learned usage data. (Supplementary Note 15) The information processing method according to Supplementary Note 12, comprising obtaining the analysis results for predictions made by the prediction model using the usage data. (Supplementary Note 16) The information processing method according to Supplementary Note 15, comprising obtaining the analysis results of analyzing factors that caused a prediction error in predictions made by the prediction model using the usage data. (Supplementary Note 17) The information processing method according to Supplementary Note 16, wherein learning data used when training the prediction model and / or operation data used during operation are acquired as the usage data, and characteristics of the usage data are acquired as the analysis result as a cause of a prediction error by the prediction model using the usage data. (Supplementary Note 18) The information processing method according to Supplementary Note 17, wherein a correction process is performed on the usage data according to the characteristics of the usage data, and re-learned data to be used when re-learning the prediction model is generated as the re-learning usage data. (Supplementary Note 19) The information processing method according to Supplementary Note 18, wherein a correction process is performed to delete some data from the usage data or to assign a weight to some data of the usage data according to the characteristics of the usage data, and the re-learned data is generated.(Supplementary Note 20) The information processing method according to Supplementary Note 12, comprising: acquiring, as the usage data, learning data used when training the prediction model and / or operation data used during operation, or setting data of the prediction model; and performing at least one of correction processes such as deleting a portion of the learning data and / or operation data, assigning a weight to a portion of the learning data and / or operation data, and setting new setting data according to the analysis results, to generate the re-learning usage data. (Supplementary Note 21) The information processing method according to Supplementary Note 12, comprising: acquiring, as the usage data, data belonging to two groups that have been used for the prediction model for different periods of time from the learning data used when training the prediction model and / or the operation data used during operation; acquiring the analysis result based on a comparison of the usage data between the groups; and performing correction processes on the usage data according to the analysis result, to generate the re-learning usage data. (Supplementary Note 22) The information processing method according to Supplementary Note 21, comprising: acquiring, as the analysis result, a ratio at which the usage data belonging to a second group, the time at which the usage data was used in the prediction model being further away than the first group, appears in the usage data belonging to a first group, the time at which the usage data was used in the prediction model being closer than the first group; and performing a correction process to assign a weight to the usage data according to the analysis result, thereby generating the re-learned usage data. (Supplementary Note 23) A computer-readable storage medium having stored thereon a program that causes a computer to execute the steps of: acquiring usage data used in a prediction model; acquiring predetermined analysis results for the prediction model or the usage data; and generating re-learned usage data to be used for re-learning the prediction model based on the usage data according to the analysis result.

[0049] REFERENCE SIGNS LIST 10 Information processing device 11 Reference information acquisition unit 12 Analysis information acquisition unit 13 Re-learning information generation unit 15 Acquired information storage unit 20 Reference information storage device 30 Analysis information storage device 100 Information processing device 101 CPU 102 ROM 103 RAM 104 Program group 105 Storage device 106 Drive device 107 Communication interface 108 Input / output interface 109 Bus 110 Storage medium 111 Communication network 121 Data acquisition unit 122 Analysis acquisition unit 123 Generation unit

Claims

1. An information processing apparatus comprising: a data acquisition unit that acquires usage data used in a prediction model; an analysis acquisition unit that acquires a preset analysis result for the prediction model or the usage data; and a generation unit that generates re-learning usage data used for re-learning the prediction model based on the usage data according to the analysis result.

2. The information processing apparatus according to claim 1, wherein the generation unit generates the re-learning usage data by performing a correction process on the usage data according to the analysis result.

3. The information processing apparatus according to claim 2, wherein the analysis acquisition unit acquires a plurality of different analysis results, and the generation unit sets candidate correction processes for the usage data according to each of the analysis results, and performs a final correction process determined based on the set candidate correction processes on the usage data to generate the re-learning usage data.

4. The information processing apparatus according to claim 1, wherein the analysis acquisition unit acquires the analysis result for the prediction by the prediction model using the usage data.

5. The information processing apparatus according to claim 4, wherein the analysis acquisition unit acquires the analysis result obtained by analyzing the cause of a prediction error in the prediction by the prediction model using the usage data.

6. The information processing apparatus according to claim 5, wherein the data acquisition unit acquires the learning data used during learning of the prediction model and / or the operation data used during operation as the usage data, and the analysis acquisition unit acquires the characteristics of the usage data as the analysis result as the cause of the prediction error in the prediction by the prediction model using the usage data.

7. The information processing apparatus according to claim 6, wherein the generation unit performs a correction process on the usage data according to the characteristics of the usage data to generate re-learning data used during re-learning of the prediction model as the re-learning usage data.

8. The information processing apparatus according to claim 7, wherein the generation unit performs a correction process of deleting a part of the usage data or assigning weights to a part of the usage data according to the characteristics of the usage data to generate the re-learning data.

9. An information processing apparatus according to claim 1, wherein the data acquisition unit acquires, as the usage data, learning data used during learning of the prediction model and / or operation data used during operation, or setting data of the prediction model, and the generation unit performs at least one of correction processes such as deleting some data of the learning data and / or operation data, assigning weights to some data of the learning data and / or operation data, and newly setting the setting data, according to the analysis result, to generate the re-learning usage data. Information processing apparatus.

10. An information processing apparatus according to claim 1, wherein the data acquisition unit acquires, as the usage data, data belonging to two groups with different times used in the prediction model, among the learning data used during learning of the prediction model and / or operation data used during operation, the analysis acquisition unit acquires the analysis result based on comparison of the usage data between the groups, and the generation unit performs a correction process on the usage data according to the analysis result to generate the re-learning usage data. Information processing apparatus.

11. An information processing apparatus according to claim 10, wherein the analysis acquisition unit acquires, as the analysis result, the ratio at which the usage data belonging to the second group, in which the time used in the prediction model is farther than that of the first group, appears in the usage data belonging to the first group, in which the time used in the prediction model is close, and the generation unit performs a correction process of assigning weights to the usage data according to the analysis result to generate the re-learning usage data. Information processing apparatus.

12. An information processing method for acquiring usage data used in a prediction model, acquiring a preset analysis result for the prediction model or the usage data, and generating re-learning usage data used for re-learning the prediction model based on the usage data according to the analysis result.

13. An information processing method according to claim 12, wherein a correction process is performed on the usage data according to the analysis result to generate the re-learning usage data. Information processing method.

14. An information processing method according to claim 13, comprising: obtaining a plurality of different analysis results; setting candidate correction processes for the usage data according to each of the analysis results; and performing a final correction process determined based on the set candidate correction processes on the usage data to generate the re-learning usage data.

15. A computer-readable storage medium storing a program that causes a computer to execute a process of obtaining usage data used in a prediction model, obtaining a preset analysis result for the prediction model or the usage data, and generating re-learning usage data for re-learning the prediction model based on the usage data according to the analysis result.

Images