Information processing device, information processing system, information processing method and program

Abstract

In order to solve the problem of assisting in streamlining the generation of a machine learning-based model, this information processing device (1) comprises: a first acquisition means (11) for acquiring data of interest, which is data relating to a machine learning operation of interest; a second acquisition means (12) for acquiring executed data, which is data relating to a plurality of machine learning operations that have already been executed; an identification means (13) for using the data of interest and the executed data to identify one or more executed machine learning operations similar to the machine learning operation of interest; a priority calculation means (14) for calculating the priority of each of a plurality of parameters relating to the one or more executed machine learning operations identified by the identification means, and assigning the calculated priority to the corresponding parameter; and a generation means (15) for generating output data which contains identification information for the plurality of parameters and a range of values for the plurality of parameters, the output data further including priorities in an identifiable manner.

Description

[Technical Field]

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

[0002] To generate a model with high analytical accuracy, it is necessary to manually select parameters from a large number of parameters and adjust them for each trial. Each trial requires a fairly large amount of calculation (several hours to several days), so parameter adjustment ultimately takes a significant amount of time.

[0003] Patent Document 1 discloses a parameter adjustment device that can efficiently perform grid search. This parameter adjustment device extracts parameter combination patterns that keep the model accuracy within an acceptable range, and uses the highest-ranked combination patterns the next time an analysis is performed. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. WO2020 / 189371 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the technique disclosed in Patent Document 1 has a problem in that it is difficult to identify which of multiple parameters contributes to improving the analytical accuracy of the model, because the combinations themselves are ranked.

[0006] One aspect of the present invention has been made in consideration of the above-mentioned problems, and one example of its purpose is to provide a technology that can support the efficiency of generating models through machine learning. [Means for solving the problem]

[0007] An information processing device according to one aspect of the present invention comprises a first acquisition means for acquiring target data, which is data related to a target machine learning; a second acquisition means for acquiring executed data, which is data related to multiple machine learnings that have been executed; an identification means for using the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning; a priority calculation means for calculating a priority for each of multiple parameters related to the one or more executed machine learnings identified by the identification means and assigning the calculated priority to the parameter; and a generation means for generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0008] An information processing system according to one aspect of the present invention comprises a first acquisition means for acquiring target data, which is data related to a target machine learning; a second acquisition means for acquiring executed data, which is data related to multiple machine learnings that have been executed; an identification means for using the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning; a priority calculation means for calculating a priority for each of multiple parameters related to the one or more executed machine learnings identified by the identification means and assigning the calculated priority to the parameter; and a generation means for generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0009] An information processing method according to one aspect of the present invention includes one or more processors acquiring target data, which is data related to a target machine learning, acquiring executed data, which is data related to multiple machine learnings that have been executed, using the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning, calculating a priority for each of multiple parameters related to the identified one or more executed machine learnings and assigning the calculated priority to the parameters, and generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0010] A program according to one aspect of the present invention causes a computer to execute a first acquisition process for acquiring target data, which is data related to a target machine learning; a second acquisition process for acquiring executed data, which is data related to multiple executed machine learning processes; an identification process for using the target data and the executed data to identify one or more executed machine learning processes that are similar to the target machine learning; a priority calculation process for calculating a priority for each of multiple parameters related to the identified one or more executed machine learning processes and assigning the calculated priority to the parameters; and a generation process for generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priorities in an identifiable manner. [Effects of the Invention]

[0011] According to one aspect of the present invention, it is possible to provide a technology that can support the improvement of the efficiency of model generation through machine learning. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a block diagram showing a configuration of an information processing device according to a first exemplary embodiment of the present invention. [Figure 2] 1 is a flowchart showing the flow of an information processing method according to the first exemplary embodiment. [Figure 3] FIG. 10 is a block diagram showing the configuration of an information processing device according to a second exemplary embodiment of the present invention. [Figure 4] 10 is a flowchart showing the flow of information processing involving an information processing device according to a second exemplary embodiment. [Figure 5] 1 is an example of data input to an analysis device and data output from the analysis device. [Figure 6] FIG. 10 is a diagram illustrating an example of a specification process performed by a specification unit. [Figure 7] 10A and 10B are diagrams illustrating results of priority calculation processing executed by an extraction unit and a priority calculation unit. [Figure 8] 10 is a flowchart showing the flow of an information processing method S2 according to the second exemplary embodiment. [Figure 9] FIG. 10 is a block diagram showing the configuration of an information processing system according to a third exemplary embodiment of the present invention. [Figure 10] 10 is a flowchart showing an example of a generation process S3 according to the fourth exemplary embodiment of the present invention. [Figure 11] 13 is an example of output data in which axes of hyperparameters are arranged in order of priority according to the fourth exemplary embodiment. [Figure 12] This is an example in which part of the identified machine learning input data is included in the output data. [Figure 13] This is an example of output data in which the thickness of the lines indicating combinations of multiple hyperparameters is changed depending on the value. [Figure 14] This is an example of output data when the user selects three hyperparameters. [Figure 15] 10 is an example of output data when the user selects non-adjacent hyperparameters. [Figure 16] 16 is a modified example of the output data shown in FIG. 15. [Figure 17] 10 is an example of output data that allows the user to select the sort order of hyperparameters. [Figure 18]This is an example of output data that allows the user to specify a predetermined relationship between combinations of values ​​of high-value hyperparameters. [Figure 19] FIG. 1 is a configuration diagram for realizing an information processing device by software. DETAILED DESCRIPTION OF THE INVENTION

[0013] Exemplary Embodiment 1 A first exemplary embodiment of the present invention will be described in detail with reference to the drawings. This exemplary embodiment is a basic form of the exemplary embodiments described below.

[0014] (Configuration of information processing device 1) The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the information processing device 1. As shown in the figure, the information processing device 1 includes a first acquisition unit 11, a second acquisition unit 12, an identification unit 13, a priority calculation unit 14, and a generation unit 15. The first acquisition unit 11, the second acquisition unit 12, the identification unit 13, the priority calculation unit 14, and the generation unit 15 are embodiments of the first acquisition means, the second acquisition means, the identification means, the priority calculation means, and the generation means, respectively, set forth in the claims. The information processing device 1 is an information processing device for efficiently setting parameters of a machine learning model, particularly parameters that are set in advance by a user, when performing machine learning on new data.

[0015] The first acquisition unit 11 acquires target data, which is data related to machine learning. "Machine learning" (hereinafter also referred to as "machine learning processing"), for example, refers to learning using a machine learning device that executes a machine learning algorithm. In this exemplary embodiment, it refers to a process in which a user sets certain parameters and trains a model for purposes such as regression, classification, and prediction. Furthermore, "machine learning" includes a machine learning algorithm. "Data related to machine learning" includes, for example, the type of machine learning device, the types and values ​​of parameters set by the user, data to be processed by the algorithm (data input to the machine learning device), and final or intermediate data calculated by the algorithm (data output from the machine learning device, etc.). Furthermore, "target data" includes the type of machine learning device that generates a new model, the data input to the machine learning device, the values ​​of parameters set by the user, and the data output from the machine learning device.

[0016] The machine learning device may be a device that uses various open source software (Scikit Learn, TensorFlow, etc.), or may be a machine learning device created independently by the user. The machine learning device is also called an analysis device, and may be a different type of machine learning device. Hereinafter, setting certain parameters and performing machine learning will also be referred to as "analysis." Generally, a high-performance model can be generated by changing parameters and repeating the analysis multiple times.

[0017] The second acquisition unit 12 acquires executed data, which is data related to multiple executed machine learning operations. "Executed data" refers to data analyzed in the past, such as the type of machine learning device, the numerical values ​​of set parameters, data input to the machine learning device, and data output from the machine learning device. The data output from the machine learning device also includes indicators, which will be described later.

[0018] The identification unit 13 uses the target data and the executed data to identify one or more executed machine learning processes that are similar to the target machine learning process. "Similar" refers to having something in common. For example, machine learning processes that use the same type of machine learning device, machine learning processes that input similar types of data, and machine learning processes that have similar names of analysis projects can be considered similar machine learning processes. As an example, the identification unit 13 can use a machine model that evaluates the similarity of sentences to identify machine learning processes that have a similarity to the target data that is equal to or greater than a predetermined threshold from among the executed machine learning processes.

[0019] The priority calculation unit 14 calculates a priority for each of the multiple parameters related to one or more executed machine learning processes identified by the identification unit 13, and assigns the calculated priority to the parameter. In this exemplary embodiment, priority refers to the order of precedence in which parameters should be selected as parameters that are highly effective in improving machine learning. In other words, the priority calculation unit 14 calculates and assigns a higher priority to parameters with a higher learning effect.

[0020] The generating unit 15 generates output data including identification information of a plurality of parameters and value ranges of the plurality of parameters, and including the priority in an identifiable manner. The "output data" is, for example, data to be output to a display device for display. In this case, the generating unit 15 generates data for displaying the priority on the display device in an identifiable manner.

[0021] For example, a user can use the parameter value ranges and priority of each parameter displayed on the display device as a guide, and also take into consideration the user's personal experience and knowledge, to decide which parameters to set and at what values ​​to use in an analysis in a new machine learning run. This allows for the generation of a model with better performance more efficiently than by setting parameter values ​​randomly.

[0022] As described above, the information processing device 1 according to this exemplary embodiment is configured to include a first acquisition unit 11 that acquires target data, which is data related to the target machine learning; a second acquisition unit 12 that acquires executed data, which is data related to multiple machine learnings that have been executed; an identification unit 13 that uses the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning; a priority calculation unit 14 that calculates the priority of each of multiple parameters related to the one or more executed machine learnings identified by the identification unit 13 and assigns the calculated priority to the parameters; and a generation unit 15 that generates output data that includes identification information for the multiple parameters and value ranges for the multiple parameters, and that includes the priorities in an identifiable manner.

[0023] With this configuration, it is possible to derive the types of parameters with high learning effects and their value ranges, along with their priorities, in previously executed machine learning similar to the machine learning to be newly trained. The types of parameters, their value ranges, and their priorities can then be output in an identifiable manner. Therefore, the information processing device 1 according to this exemplary embodiment can provide the effect of supporting the efficient generation of models through machine learning.

[0024] (Flow of information processing method S1) The flow of the information processing method S1 according to this exemplary embodiment will be described with reference to the drawings. Figure 2 is a flow chart showing the flow of the information processing method S1.

[0025] As shown in the figure, the information processing method S1 includes the following steps: In step S11, one or more processors (for example, the first acquisition unit 11) acquire target data, which is data related to machine learning as a target. The target data is as described above.

[0026] In step S12, one or more processors (for example, the second acquisition unit 12) acquire executed data, which is data relating to multiple executed machine learning processes. The executed data is as described above.

[0027] Next, in step S13, one or more processors (for example, the identifying unit 13) use the target data and the executed data to identify one or more executed machine learning processes similar to the target machine learning. Examples of similar machine learning processes are as described above.

[0028] Next, in step S14, one or more processors (e.g., priority calculation unit 14) calculate the priority of each of the multiple parameters related to the identified one or more executed machine learning processes, and assign the calculated priority to the parameter. An example of how to assign priority is as described above.

[0029] Next, in step S15, one or more processors (e.g., generation unit 15) generate output data including identification information of the plurality of parameters and value ranges of the plurality of parameters, and including priority in an identifiable manner. Examples of the output data are as described above.

[0030] As described above, the information processing method S1 according to this exemplary embodiment employs a configuration in which one or more processors acquire target data, which is data related to a target machine learning, acquire executed data, which is data related to multiple machine learnings that have been executed, use the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning, calculate a priority for each of multiple parameters related to the identified one or more executed machine learnings and assign the calculated priority to the parameters, and generate output data that includes identification information for the multiple parameters and value ranges for the multiple parameters, the output data including the priority in a manner that allows identification.

[0031] With this configuration, it is possible to derive the types of parameters with high learning effects and their value ranges, along with their priorities, in previously executed machine learning similar to the machine learning to be newly trained. The types of parameters, their value ranges, and their priorities can then be output in an identifiable manner. Therefore, the information processing method S1 according to this exemplary embodiment has the effect of supporting the efficiency of generating models through machine learning.

[0032] Exemplary Embodiment 2 A second exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first exemplary embodiment are given the same reference numerals, and their description will be omitted as appropriate.

[0033] (Configuration of information processing device 1A) FIG. 3 is a block diagram showing the configuration of an information processing device 1A according to exemplary embodiment 2. As shown in the figure, the information processing device 1A includes a control unit 10, a memory 17, a communication unit 18, and a database 19. The control unit 10 includes a first acquisition unit 11, a second acquisition unit 12, an identification unit 13, a priority calculation unit 14, a generation unit 15, and an extraction unit 16. The functions of the first acquisition unit 11, the second acquisition unit 12, the identification unit 13, the priority calculation unit 14, and the generation unit 15 are the same as those described in exemplary embodiment 1. Some or all of the first acquisition unit 11, the second acquisition unit 12, the identification unit 13, the priority calculation unit 14, the generation unit 15, the extraction unit 16, the memory 17, the communication unit 18, and the database 19 may be distributed. Furthermore, some or all of these units and the database 19 may be located on a cloud.

[0034] The information processing device 1A is an information processing device for efficiently setting parameters of a machine learning model when new data is to be subjected to machine learning.

[0035] In this exemplary embodiment, the "parameters of the machine learning model" include at least hyperparameters that are not subject to update through learning. Here, the hyperparameters are parameters that are set in advance by the user. Specifically, the hyperparameters include the learning rate, the number of epochs (the number of times learning of all data is repeated), the batch size (the number of divisions into which all data is divided), the window size (the size of a portion extracted from all data), the number of hidden units in the LSTM (long short-time memory) architecture, etc.

[0036] Furthermore, the "machine learning model parameters" may be configured to include parameters such as weighting coefficients that are subject to updating through learning, and loss parameters.

[0037] The extraction unit 16 extracts one or more machine learning processes associated with parameters that satisfy a predetermined condition from the one or more executed machine learning processes identified by the identification unit 13. The extraction unit 16 is one form of extraction means described in the claims. Furthermore, the priority calculation unit 14 calculates the priority of each of multiple parameters related to the one or more machine learning processes extracted by the extraction unit 16, and assigns the calculated priority to the parameter.

[0038] An example of the "predetermined condition" is that an index indicating the degree of learning (performance) of the machine learning device has reached a level set by the user. Specifically, an example of the "parameter satisfying the predetermined condition" is that a loss parameter (described below) indicating the performance of the machine learning device has reached a level set by the user.

[0039] As an example, the memory 17 includes a ROM (Read Only Memory) and a RAM (Random Access Memory), and stores one or more programs in the ROM, which are appropriately expanded into the RAM and executed to realize the functions of each part of the control unit 10.

[0040] In the database 19, the target data acquired by the first acquisition unit 11, the executed data acquired by the second acquisition unit 12, and the like are recorded.

[0041] The information processing device 1A is configured to be able to exchange information with the analysis result database 30 via the information communication network N. The analysis result database 30 stores executed data.

[0042] The information processing device 1A may include a display unit 20. The output data generated by the generation unit 15 is output to and displayed on the display unit 20. The display unit 20 may be, for example, a display.

[0043] The first acquisition unit 11, the second acquisition unit 12, the identification unit 13, the priority calculation unit 14, the generation unit 15, the extraction unit 16, the memory 17, the communication unit 18, and the database 19 do not need to be integrated into one information processing device 1A. In other words, some or all of these may be distributed and located in different locations. Furthermore, some or all of these may be distributed and located on the cloud.

[0044] (Information processing flow) The flow of information processing related to the information processing device 1A having the above configuration will be described with reference to the drawings. Fig. 4 is a flowchart showing the flow of information processing related to the information processing device 1A.

[0045] As shown in the figure, executed data is stored in the analysis result database 30. Specifically, the analytical data, hyperparameters, etc. used in the analysis are input to the analysis device, and the output data from the analysis device is recorded in the analysis result database 30 together with the input analytical data, hyperparameters, etc. (step S20).

[0046] The first acquisition unit 11 acquires the target data input to the information processing device 1A and records it in the database 19 (step S21: first acquisition process). Alternatively, the target data may be stored in the analysis result database 30. In this case, the first acquisition unit 11 acquires the target data from the analysis result database 30 via the communication unit 18 and records it in the database 19.

[0047] Meanwhile, the second acquisition unit 12 acquires the executed data from the analysis result database 30 via the communication unit 18 and records it in the database 19 (step S22: second acquisition process). Alternatively, the second acquisition unit 12 may directly acquire the executed data held by the analysis device. The first acquisition process and the second acquisition process may be executed in parallel.

[0048] Next, the identifying unit 13 identifies one or more already-executed machine learning processes that are similar to the target machine learning process, using the target data and the already-executed data recorded in the database 19 (step S23: identifying process).

[0049] Next, the extraction unit 16 extracts one or more machine learning processes associated with parameters that satisfy a predetermined condition from the identified executed machine learning processes (step S24: extraction process).

[0050] As an example, the extraction unit 16 extracts one or more machine learning processes that output a loss parameter equal to or less than a predetermined threshold from among one or more executed machine learning processes identified by the identification unit 13. In this exemplary embodiment, the loss parameter refers to a loss value that is the output of a loss function included in a learning algorithm. The loss value is an index that indicates the performance of the model.

[0051] A loss function is a function that evaluates the difference between an output value and training data (correct answer data) in supervised learning. Examples of loss functions include, but are not limited to, mean squared error, mean absolute error, square root of mean squared error, Huber loss, and Poisson loss.

[0052] Next, the priority calculation unit 14 uses the loss parameters output by each of the multiple machine learning methods extracted by the extraction unit 16 and parameters other than the loss parameters to assign a priority to each of the parameters other than the loss parameters according to their contribution to improving the loss parameters (step S25: priority calculation process).

[0053] In this exemplary embodiment, the parameters other than loss parameters are hyperparameters. As an example, the contribution is the amount of reduction in the loss value when the hyperparameter is changed. The smaller the loss value indicating the performance of the model, the better. Therefore, when the loss value is used as the loss parameter, the priority calculation unit 14 assigns a higher priority to the hyperparameter that results in a larger reduction in the calculated loss value when the hyperparameter is changed.

[0054] Note that the change range of the hyperparameters in the executed data (the difference between the maximum and minimum values) varies greatly depending on the type of hyperparameter. Therefore, it may be difficult to compare the effect of changing each hyperparameter on the reduction of the loss value for each hyperparameter. In such cases, the priority calculation unit 14 may calculate the degree of contribution by comparing the amount of reduction in the loss value when the change range of each hyperparameter is changed by, for example, the entire change range, half, or one-fourth, etc., and assign a priority. This is because the change range of the hyperparameters in the executed data is determined to be within a range where the loss parameter can be improved as a result of trials.

[0055] More specifically, the priority calculation unit 14 may determine the contribution level based on a comparison result between loss parameters output by each of the multiple machine learning methods and a comparison result between parameters (hyperparameters) other than the loss parameters used in each of the multiple machine learning methods. For example, the loss parameters are compared to extract two machine learning methods with different loss parameter values. Then, hyperparameters whose values ​​have been changed between the two machine learning methods are compared. Such a comparison may also be performed between different machine learning methods. The priority calculation unit 14 calculates and determines the contribution level of a hyperparameter estimated to have the effect of reducing the loss parameter by changing its value, based on the amount of reduction in the loss parameter. Then, the priority calculation unit 14 may assign a priority based on the contribution level. The number of hyperparameters estimated by the priority calculation unit 14 to have the effect of reducing the loss parameter is not limited to one, and may be multiple. The priority calculation unit 14 may determine the contribution level for each of the multiple hyperparameters. Then, the priority calculation unit 14 may assign a higher priority to a hyperparameter with a higher contribution level.

[0056] Next, the generating unit 15 generates output data including identification information of a plurality of parameters and value ranges of the plurality of parameters, and including priority in an identifiable manner (step S26: generating process).

[0057] The information processing device 1A may output the generated output data to the display unit 20 (step S27: output process).

[0058] (Example of information processing) An example of the above information processing will be specifically described with reference to the drawings. FIG. 5 shows an example of data input to the analysis device ANL1 and data output from the analysis device ANL1. The input shown in FIG. 5 is a portion of data that the user intends to have a new model learn (analyze), and as shown in the Model_id column of the "hyperparameter identification information" among the data, the name "TDA" is given as the name of the machine learning analysis project. Below, an example will be described in which this analysis project TDA is used as the target data.

[0059] Figure 5 shows the input data, which are the data to be analyzed, as input data (X.csv) and input data (Y.csv), and hyperparameters. In the example shown, four datasets from sensors 1 to 4 are shown as input data (X.csv) and input data (Y.csv), each with three data sets, but in reality, a large amount of input data is included.

[0060] Additionally, Learning_rate, Training_epochs, Batch_size, Input File, and Model_id are shown as identification information indicating hyperparameters. Learning_rate is the learning rate. Training_epochs is the number of epochs. Batch_size is the batch size. Input File is the name of the input data mentioned above. Model_id is the name of an individual analysis within one analysis project.

[0061] The output data includes the execution time, the name (type) of the analysis device, loss (loss parameter), Learning_rate, Training_epochs, Batch_size, Input File, and Model_id. Of these, the data output corresponding to the input data are the execution time, the name (type) of the analysis device, and loss, and the rest is the input data output as is. This output data is recorded in the analysis result database 30. The first acquisition unit 11 acquires this output data recorded in the analysis result database 30 and records it in the database 19 as target data. In addition, the second acquisition unit 12 acquires the executed data recorded in the analysis result database 30 and records it in the database 19. The executed data acquired by the second acquisition unit 12 may be only the output data of the analysis.

[0062] In this way, the first acquisition unit 11 acquires, for example, target data, which is data related to the target machine learning (analysis project), from the analysis result database 30, and records it in the database 19. In addition, the second acquisition unit 12 acquires, from the analysis result database 30, executed data, which is data related to multiple executed machine learning projects, and records it in the database 19.

[0063] (Specific processing) 6 is a diagram showing an example of the identification process executed by the identification unit 13. Table 601 shows the target data acquired by the first acquisition unit 11 and a portion of the executed data acquired by the second acquisition unit 12, with one row recording one analysis data. The types of data shown in table 601 are, from the left, the execution time, which is the time when the analysis started, the name (type) of the analysis device, the loss value obtained by the analysis, the learning rate (Learning_rate1), the number of epochs (Training_epochs), the batch size (Batch_size), the name of the input data (Input File), and the name of the analysis project (Model_id). Note that the types of hyperparameters are not limited to these.

[0064] The data in the first row, where Model_id is TDA, is the target data. That is, it is assumed that the user is using the information processing device 1A to efficiently determine hyperparameters to be used in the target analysis project TDA.

[0065] Table 602 shows data on executed machine learning similar to the target machine learning, identified by the identification unit 13 using the target data and executed data. As an example, the identification unit 13 identifies machine learning similar to the TDA in the names of input data and analysis projects. As a result, in table 602, analysis projects TDA_2, TDA_3, TDA_4, TDA_5, and TDA_6 are identified for the target TDA data, in which the names of analysis devices and input data are similar. Note that if related analysis projects are linked to a certain analysis project, similar analysis projects may be identified by referring to the related analysis projects.

[0066] In this way, the identifying unit 13 uses the target data and the executed data to identify one or more executed machine learning processes (executed data shown in Table 602) that are similar to the target machine learning process.

[0067] (Priority calculation process) FIG. 7 is a diagram showing the results of the priority calculation process executed by the extraction unit 16 and the priority calculation unit 14. Table 701 shows TDA_2, TDA_3, and TDA_4 extracted by the extraction unit 16 from among the analysis projects identified by the identification unit 13. Here, the extraction unit 16 extracts analysis projects with a loss < 2.0 by referring to the condition that the loss specified by the user is less than 2.0. In this way, the extraction unit 16 extracts one or more machine learning processes associated with parameters that satisfy a predetermined condition from among one or more executed machine learning processes identified by the identification unit 13.

[0068] The priority calculation unit 14 calculates an index indicating the priority of the hyperparameter by referring to TDA_2, TDA_3, and TDA_4, and assigns the priority. As an example, the priority calculation unit 14 identifies the degree of contribution to reducing the loss value based on the comparison result between the loss values ​​and the comparison result between the hyperparameters. Then, the priority calculation unit 14 assigns a priority to each hyperparameter based on its contribution to improving the loss value.

[0069] As an example, when comparing two machine learning models in which only a certain hyperparameter is changed, if the difference in loss value is 1 or more, 1 is added to the priority score of that parameter. This type of comparison is performed for each hyperparameter, and priorities can be assigned according to the magnitude of the obtained priority score. The difference in loss value does not need to be 1 or more; any arbitrary number can be set to calculate the priority score.

[0070] As an example, the priority calculation unit 14 may create simultaneous equations in which each hyperparameter is a variable, multiplied by a coefficient, and the sum of the multiplied values ​​is used as a loss value, determine the combination of coefficients that best fits the simultaneous equations, and assign priorities based on that ranking by regarding the coefficients as contributions. Since a smaller hyperparameter may result in a smaller loss value, a simultaneous equation may be evaluated in which the absolute value of the difference between the hyperparameters of two analyses is used as a variable, multiplied by a coefficient, and the sum of the multiplied values ​​is used as the difference in loss values. Furthermore, if the magnitudes of the hyperparameters themselves differ significantly, the absolute value of the difference between the hyperparameters of the two analysis results, normalized to the ratio of the difference between the maximum and minimum values ​​of the hyperparameters, may be used.

[0071] Table 702 is a table in which the priority calculation unit 14 assigns priorities to hyperparameters. As shown in Table 702, the priority calculation unit 14 assigns the highest priority (priority 1) to the learning rate. Next, the priority calculation unit 14 assigns priority 2 to the number of epochs and priority 3 to the batch size.

[0072] That is, Table 702 indicates that the priority calculation unit 14 has estimated that changing the learning rate has the greatest effect on changing the loss value. Specifically, it can be seen from Tables 701 and 702 that decreasing the learning rate has the greatest contribution to improving (reducing) the loss value. It can also be seen that increasing the number of epochs has the second greatest contribution, and increasing the batch size has the third greatest contribution.

[0073] As described above, the priority calculation unit 14 calculates the priority of each of one or more machine learning related parameters extracted by the extraction unit 16, and assigns the calculated priority to the parameter.

[0074] By referring to this table 702, the user can determine which hyperparameters should be adjusted to reduce the loss value. Table 702 also displays the minimum and maximum values ​​of the prioritized hyperparameters. By referring to this, the user can determine to what extent each hyperparameter should be set.

[0075] (Generation process) The generation unit 15 can generate output data that displays the above-mentioned Tables 601, 602, 701, and 702. The generation unit 15 may also generate output data that can display the effects of hyperparameters in a way that is easier for the user to understand. Specific examples of the output data generated by the generation unit 15 will be described later.

[0076] In addition, the priority calculation unit 14 may not only assign priorities to multiple hyperparameters, but also calculate optimal values ​​or optimal ranges that further narrow the range of hyperparameters, and the generation unit 15 may generate these as output data.

[0077] As described above, the information processing device 1A according to this exemplary embodiment includes, in addition to the configuration of the information processing device 1 according to exemplary embodiment 1, an extraction unit 16 that extracts one or more machine learning processes associated with parameters that satisfy a predetermined condition from among the one or more executed machine learning processes identified by the identification unit 13, and a priority calculation unit 14 that assigns priorities to multiple parameters related to the one or more machine learning processes extracted by the extraction unit 16. Therefore, according to the information processing device 1A according to this exemplary embodiment, in addition to the effects achieved by the information processing device 1 according to exemplary embodiment 1, an effect is obtained in that the user can determine the type of hyperparameter that should be adjusted to improve the loss value and how to set the hyperparameter.

[0078] (Information processing method S2) Next, the flow of the information processing method S2 executed by the information processing device 1A will be described with reference to the drawings. Fig. 8 is a flowchart showing the flow of the information processing method S2.

[0079] As shown in the figure, the information processing method S2 includes the following steps: That is, in step S21, one or more processors (for example, the first acquisition unit 11) acquire target data that is data related to target machine learning.

[0080] Also, in step S22, one or more processors (for example, the second acquisition unit 12) acquire executed data, which is data relating to multiple executed machine learning processes.

[0081] Next, in step S23, one or more processors (for example, the identifying unit 13) use the target data and the executed data to identify one or more executed machine learning processes similar to the target machine learning process.

[0082] Next, in step S24, one or more processors (e.g., the extraction unit 16) extract one or more machine learning processes associated with parameters that satisfy a predetermined condition from among the identified executed machine learning processes. The predetermined condition is, for example, that the loss value is smaller than a threshold.

[0083] Next, in step S25, one or more processors (for example, priority calculation unit 14) calculate the priority of each of the one or more extracted machine learning related parameters, and assign the calculated priority to the parameter.

[0084] Next, in step S26, one or more processors (e.g., generation unit 15) generate output data including identification information of multiple parameters and value ranges of the multiple parameters, and including priority in an identifiable manner.

[0085] With this configuration, it is possible to derive the types of parameters that satisfy predetermined conditions, their value ranges, and priorities in previously executed machine learning similar to the machine learning to be newly trained. The types of parameters, their value ranges, and priorities can then be output in an identifiable manner. Therefore, the information processing method S2 according to this exemplary embodiment has the same effect as the information processing method S1 according to the first exemplary embodiment, and also has the effect of allowing a user to determine the types of hyperparameters that should be adjusted to improve the loss value and how to set those hyperparameters.

[0086] Exemplary Embodiment 3 A third exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first and second exemplary embodiments are denoted by the same reference numerals, and their description will not be repeated.

[0087] 9 is a block diagram showing the configuration of an information processing system 2 according to this exemplary embodiment. As shown in the figure, the information processing system 2 includes a first acquisition unit 11, a second acquisition unit 12, an identification unit 13, a priority calculation unit 14, and a generation unit 15. The functions of these units are as described in exemplary embodiment 1. The first acquisition unit 11, the second acquisition unit 12, the identification unit 13, the priority calculation unit 14, and the generation unit 15 are connected to each other via an information communication network N so as to be able to communicate information with each other.

[0088] Furthermore, each unit of the information processing system 2 can communicate information with the database 19 and the analysis result database 30 via the information communication network N. The functions of the database 19 and the analysis result database 30 are as described in the exemplary embodiment 2. Some or all of the units of the information processing system 2, the database 19, and the analysis result database 30 may be distributed. Some or all of the units of the information processing system 2, the database 19, and the analysis result database 30 may be located on the cloud.

[0089] As described above, the components of the information processing system 2 are connected to each other via the information communication network N so as to be able to communicate with each other. Furthermore, the components of the information processing system 2 are capable of communicating with the database 19 and the analysis result database 30 via the information communication network N. Furthermore, some or all of the components of the information processing system 2, the database 19, and the analysis result database 30 may be located on the cloud. Therefore, according to the information processing system 2 according to this exemplary embodiment, in addition to the effects achieved by the information processing device 1 according to the exemplary embodiment 1, the effect that some or all of the components of the information processing system 2, the database 19, and the analysis result database 30 can be distributed and located at any desired locations can be obtained.

[0090] Exemplary Embodiment 4 A fourth exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first to third exemplary embodiments are denoted by the same reference numerals, and their description will not be repeated.

[0091] In this exemplary embodiment, the output data generated by the generation unit 15, i.e., the content of the output displayed on the display unit 20 described in exemplary embodiment 2, will be described. Fig. 10 is a flow diagram showing the output data generation process S3 executed by the generation unit 15. Note that the content of the generation process shown in Fig. 10 is just an example, and the generation unit 15 may execute various other generation processes.

[0092] As shown in the figure, the generation process S3 includes the following steps: In step S31, one or more processors (e.g., the generation unit 15) calculate the minimum and maximum hyperparameter values ​​of the identified machine learning and store them in a database. The identified machine learning is the machine learning identified by the identification unit 13. The database may be, for example, the database 19 described in the second exemplary embodiment.

[0093] Next, in step S32, one or more processors (for example, the generation unit 15) generate a parallel coordinate graph drawn using the maximum value, minimum value, and priority of each hyperparameter. A parallel coordinate graph is a graph that shows the numerical range (coordinate range) of one hyperparameter on a one-dimensional axis, and is a graph in which the axes of multiple hyperparameters are arranged in parallel. The parallel coordinate graph can be generated using known drawing software.

[0094] Next, in step S33, one or more processors (e.g., generation unit 15) generate drawing data by extracting a portion of the identified machine learning input data. By displaying a portion of the input data, the user can check whether the type of input data, the numerical values ​​of the input data, etc., are similar to the target data. Note that, although step S33 is not essential, when this step is executed, it is preferable to record the identified machine learning input data in database 19 in advance.

[0095] Next, in step S34, various output data are generated based on the user's designation. For example, in step S341, the generation unit 15 may generate output data in which the axes of the hyperparameters are arranged in a predetermined order or in an order designated by the user. The predetermined order may be, for example, an order of priority, an order of the number of times the parameter values ​​have been changed, or the like. The priority order may be set as a default when no designation is made by the user. The output data includes identification information of the hyperparameters and their value ranges.

[0096] In step S342, output data may be generated in which the coordinates of the maximum and minimum values ​​of the ranges displayed on each axis selected by the user are connected by a line. Alternatively, as a default, output data may be generated in which the coordinates of the maximum and minimum values ​​of the ranges displayed on all axes are connected by a line. The coordinates of the maximum and minimum values ​​can be calculated using known coordinate calculation software.

[0097] FIG. 11 is an example of a parallel coordinate graph (output data) in which the axes of hyperparameters are arranged in order of priority. The output data shown in FIG. 11 includes identification information and value ranges of the hyperparameters arranged in order of priority. Specifically, the numerical range of one hyperparameter is displayed on a one-dimensional vertical axis along with its identification information. The vertical axes of multiple hyperparameters are arranged in order of priority from the left, with the learning rate (Learning_rate1), the number of epochs (Training_epochs), the batch size (Batch_size), window_size (the size for extracting some data from all data), and lstm_hidden_dim (the number of hidden units in an LSTM (long short-time memory) architecture) attached as identification information for the hyperparameters. Note that FIG. 11 includes the hyperparameters window_size and lstm_hidden_dim, which are not described in the second exemplary embodiment, but the number and type of hyperparameters are not limited. In this exemplary embodiment, output data including window_size and lstm_hidden_dim is described.

[0098] The numerical range of each hyperparameter is shown as a rectangle with its maximum and minimum values. The coordinates of the maximum and minimum values ​​of adjacent hyperparameters are connected by lines.

[0099] From the graph in Figure 11, we can see that adjusting the learning rate is the most important factor in reducing the loss value. We can also see that the next most important hyperparameters are the number of epochs, followed by the batch size, window_size, and lstm_hidden_dim.

[0100] Figure 12 shows an example in which part of the identified machine learning input data is included in the output data. Part of the input data, X.csv and Y.csv, is plotted below the parallel coordinate graph in Figure 12. As mentioned above, by displaying part of the input data, the user can check whether the type of input data, the numerical values ​​of the input data, etc., are similar to the target data.

[0101] Returning to step S343 of step S34, the generation unit 15 may generate output data in which the display mode of the line connecting the coordinates of the maximum or minimum values ​​is changed according to the value.

[0102] The generation unit 15 may generate output data that displays combinations of multiple parameter values ​​by changing the display mode depending on the value of the combination. FIG. 13 illustrates, as an example, output data in which the thickness of the lines representing combinations of multiple hyperparameters is changed depending on the value. In this exemplary embodiment, the value refers to, for example, the magnitude of the effect of reducing the loss value. That is, the generation unit 15 determines that a combination of hyperparameter values ​​that has a relatively large effect of reducing the loss value has a higher value than a combination that has a relatively small effect of reducing the loss value. The generation unit 15 then generates output data that displays a combination of hyperparameter values ​​with a higher relative value using a thicker line. The manner in which the value is changed is not limited to the thickness of the line. For example, the color or type of the line may be changed. The value may also be changed in three or more ways. For example, the relative values ​​may be high, medium, and low, respectively, using extra-thick, thick, and thin lines.

[0103] Figure 13 shows output data when the user selects the number of epochs and batch size. In this case, the combination of 500 epochs and 2000 batch sizes is displayed in bold. This indicates that this combination of values ​​is highly effective in reducing the loss value. In the displays in Figures 11 and 12, the range is shown by connecting the maximum values ​​and the minimum values ​​with lines, so it is not possible to determine which combinations within that range are most effective in reducing the loss value. However, by displaying the data as in Figure 13, the user can determine how to combine multiple hyperparameters.

[0104] Figure 14 shows an example of output data when the user selects the number of epochs, batch size, and window_size. In this case, the combination of 500 epochs, 2000 batch size, and 10 window_size is displayed in bold. This indicates that this combination of the hyperparameter values ​​is highly effective in reducing the loss value.

[0105] FIG. 15 shows an example of output data when a user selects non-adjacent epoch numbers and window_size. In this way, the user may select non-adjacent hyperparameters. In this case, output data is generated in which combinations of the selected hyperparameter values ​​that have a high value (great effect on reducing the loss value) are connected with a thick line. In the example shown in FIG. 15, the combination of 500 epoch numbers and 10 window_size is more valuable than other combinations, so it is connected with a thick line in the output data.

[0106] The generation unit 15 may generate output data that displays the ranges of values ​​of multiple selected parameters from multiple parameters. FIG. 16 is a modified example of the output data shown in FIG. 15 , showing an example of output data that displays multiple hyperparameters selected by a user from multiple hyperparameters and their numerical ranges. Furthermore, as shown in the figure, the generation unit 15 may generate output data in which, among the combinations of the values ​​of the multiple selected hyperparameters, combinations with relatively high values ​​are connected by a thick line. The expand / hide button shown in the upper left of the figure is a button for selecting whether to display only the hyperparameters selected by the user and their numerical ranges. In this way, by displaying only the ranges of the hyperparameters selected by the user, the value of the combination desired by the user can be more clearly recognized. Alternatively, the same effect can be achieved by displaying the selected hyperparameters more prominently instead of displaying the selection.

[0107] The generation unit 15 may generate output data including identification information of hyperparameters arranged in an arrangement order specified by the user and parameter value ranges. FIG. 17 shows an example of output data including identification information of hyperparameters and parameter value ranges when a user specifies a priority order. As shown in the figure, a selection button is displayed in the upper left of the display screen, allowing the user to specify conditions for the arrangement order of the hyperparameters. Note that FIG. 17 shows an example in which the user selects "priority order," and also selects the number of epochs and window_size, which are highlighted with a thick line.

[0108] The figure shows the case where "Order by priority" is selected, but other options are also available, such as "Order by number of changes" (in order of the number of times the hyperparameter value has been changed). Hyperparameters that have been changed many times are likely to have been judged to be valuable by users who have performed analyses in the past and have changed the value to repeat the analysis. Therefore, sorting the hyperparameters in order of the number of times the value has been changed can be used as a basis for deciding which type of hyperparameter to change.

[0109] Returning to step S34 in Fig. 10, the generation unit 15 may generate output data in which axes (hyperparameters) having a predetermined relationship designated by the user are aligned adjacently, as shown in step S344. The predetermined relationship refers to a proportional, inversely proportional, or other relationship. In other words, when a combination of numerical values ​​in which two hyperparameters have relatively high values ​​has a designated proportional, inversely proportional, or other relationship, the generation unit 15 may generate output data in which those hyperparameters are aligned adjacently.

[0110] FIG. 18 is an example of output data that allows the user to specify a predetermined relationship between combinations of hyperparameter values ​​that have a high value. As shown in the figure, a selection button is displayed in the upper left of the display screen, allowing the user to specify the relationship between combinations of hyperparameter values. The figure shows the case where "proportional" is selected, but "inversely proportional" and other options are also selectable. "Inversely proportional" refers to a relationship in which a combination in which one value is relatively large and the other is relatively small has a high value. Note that in FIG. 18, the bold line indicates that combinations in which the number of epochs and the batch size have a proportional relationship (combinations in which the number of epochs and the batch size are both small have a relatively high value).

[0111] As described above, the output data generated by the generation unit 15 is generated in various forms. By outputting and displaying the output data in such various forms on the display unit 20, the user can visually recognize which hyperparameter and what numerical range are effective in improving the performance of the model, thereby achieving the effect of supporting the efficiency of model generation by machine learning.

[0112] [Software implementation example] Some or all of the functions of the information processing devices 1, 1A and the information processing system 2 (hereinafter referred to as "information processing device 1, etc.") may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0113] In the latter case, the information processing device 1, etc., is realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in FIG. 19. The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for operating the computer C as the information processing device 1, etc. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing device 1, etc.

[0114] The processor C1 may be, for example, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0115] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, mouse, display, and printer.

[0116] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0117] [Appendix 1] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiments are also included in the technical scope of the present invention.

[0118] [Appendix 2] Some or all of the above-described embodiments can also be described as follows: However, the present invention is not limited to the following described aspects.

[0119] (Appendix 1) An information processing device comprising: a first acquisition means for acquiring target data, which is data related to a target machine learning; a second acquisition means for acquiring executed data, which is data related to multiple machine learnings that have been executed; an identification means for using the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning; a priority calculation means for calculating a priority for each of multiple parameters related to the one or more executed machine learnings identified by the identification means and assigning the calculated priority to the parameter; and a generation means for generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0120] According to the above configuration, it is possible to support the improvement of the efficiency of model generation by machine learning.

[0121] (Appendix 2) The information processing device described in Appendix 1 is characterized in that it includes an extraction means for extracting one or more machine learning processes associated with parameters that satisfy predetermined conditions from the one or more executed machine learning processes identified by the identification means, and the priority calculation means calculates a priority for each of multiple parameters related to the one or more machine learning processes extracted by the extraction means and assigns the calculated priority to the parameters.

[0122] According to the above configuration, machine learning that includes parameters that satisfy specified conditions is extracted from the identified executed machine learning, thereby more efficiently supporting the improvement of the efficiency of model generation.

[0123] (Appendix 3) The information processing device described in Appendix 2, characterized in that the extraction means extracts one or more machine learning processes that output a loss parameter below a predetermined threshold from the one or more executed machine learning processes identified by the identification means.

[0124] According to the above configuration, machine learning that outputs loss parameters equal to or less than a predetermined threshold is extracted, thereby making it possible to more efficiently support the improvement of the efficiency of model generation.

[0125] (Appendix 4) The information processing device according to claim 3, wherein the priority calculation means uses the loss parameters output by each of the plurality of machine learning methods extracted by the extraction means and parameters other than the loss parameters to assign a priority to each of the parameters other than the loss parameters according to a degree of contribution of the parameters other than the loss parameters to an improvement of the loss parameters.

[0126] According to the above configuration, parameters are prioritized according to their contribution to the improvement of the loss parameter, which makes it possible to more efficiently support the improvement of the efficiency of model generation.

[0127] (Appendix 5) The information processing device described in Appendix 4, characterized in that the priority calculation means identifies the degree of contribution based on a comparison result between loss parameters output by each of a plurality of machine learning methods and a comparison result between parameters other than the loss parameters used in each of the plurality of machine learning methods.

[0128] According to the above configuration, it is possible to increase the contribution of parameters that cause large changes in the loss parameter, thereby more efficiently supporting the improvement of the efficiency of model generation.

[0129] (Appendix 6) 6. The information processing device according to any one of claims 1 to 5, wherein the output data includes identification information of the parameters sorted in order of priority and ranges of parameter values.

[0130] According to the above configuration, the priority and numerical range of the parameters can be visually displayed to the user, thereby supporting more efficient model generation.

[0131] (Appendix 7) 6. The information processing device according to any one of claims 1 to 5, wherein the output data includes identification information of the parameters arranged in an order specified by a user and ranges of parameter values.

[0132] According to the above configuration, the user can recognize the effect of parameters from multiple perspectives, which can more effectively support the improvement of the efficiency of model generation.

[0133] (Appendix 8) 8. The information processing device according to claim 6, wherein the output data is data that displays ranges of values ​​of a plurality of parameters selected from the plurality of parameters.

[0134] According to the above configuration, the user can visually recognize the relationship between a plurality of parameters, which can more efficiently support the generation of a model.

[0135] (Appendix 9) An information processing device described in any one of Appendices 6 to 8, characterized in that the output data is data that displays combinations of values ​​of the multiple parameters by changing the display mode depending on the value of the combination.

[0136] According to the above configuration, the user can visually recognize combinations of parameter values ​​that have great value, thereby more efficiently supporting the generation of models.

[0137] (Appendix 10) 10. The information processing device according to any one of claims 6 to 9, further comprising a display means for displaying the output data.

[0138] According to the above configuration, it is possible to provide visual support to a user in generating a model through machine learning.

[0139] (Appendix 11) An information processing system comprising: a first acquisition means for acquiring target data, which is data related to a target machine learning; a second acquisition means for acquiring executed data, which is data related to multiple machine learnings that have been executed; an identification means for using the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning; a priority calculation means for calculating a priority for each of multiple parameters related to the one or more executed machine learnings identified by the identification means and assigning the calculated priority to the parameter; and a generation means for generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0140] According to the above configuration, it is possible to support the improvement of the efficiency of model generation by machine learning.

[0141] (Appendix 12) An information processing method including the steps of: one or more processors acquiring target data, which is data related to a target machine learning; acquiring executed data, which is data related to multiple machine learning methods that have been executed; using the target data and the executed data to identify one or more executed machine learning methods that are similar to the target machine learning; calculating a priority for each of multiple parameters related to the identified one or more executed machine learning methods and assigning the calculated priority to the parameters; and generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0142] According to the above configuration, it is possible to support the improvement of the efficiency of model generation by machine learning.

[0143] (Appendix 13) A program for causing a computer to execute a first acquisition process for acquiring target data, which is data related to a target machine learning; a second acquisition process for acquiring executed data, which is data related to multiple machine learning processes that have been executed; an identification process for using the target data and the executed data to identify one or more executed machine learning processes that are similar to the target machine learning; a priority calculation process for calculating the priority of each of multiple parameters related to the identified one or more executed machine learning processes and assigning the calculated priority to the parameters; and a generation process for generating output data that includes identification information of the multiple parameters and value ranges of the multiple parameters, the output data including the priorities in an identifiable manner.

[0144] According to the above configuration, it is possible to support the improvement of the efficiency of model generation by machine learning.

[0145] [Appendix 3] Some or all of the above-described embodiments can also be expressed as follows.

[0146] An information processing device comprising at least one processor, the processor executing a first acquisition process for acquiring target data, which is data related to a target machine learning; a second acquisition process for acquiring executed data, which is data related to multiple executed machine learnings; an identification process for using the target data and the executed data to identify one or more executed machine learnings that are similar to the target machine learning; an assignment process for calculating a priority for each of multiple parameters related to the identified one or more executed machine learnings and assigning the calculated priority to the parameters; and a generation process for generating output data that includes identification information for the multiple parameters and value ranges of the multiple parameters, the output data including the priority in an identifiable manner.

[0147] The information processing device may further include a memory that stores a program for causing the processor to execute the first acquisition process, the second acquisition process, the identification process, the assignment process, and the generation process. The program may be recorded on a computer-readable, non-transitory, tangible recording medium. [Explanation of symbols]

[0148] 1,1A···Information processing device 2. Information Processing System 11. First acquisition section 12. Second acquisition section 13...Specific section 14...Priority calculation unit 15...Generation section 16...Extraction part 17. Memory 18. Communications Department 19. Database 20...Display section 30. Analysis results database

Claims

1. a first acquisition means for acquiring target data, which is data related to machine learning; a second acquisition means for acquiring executed data, which is data relating to a plurality of executed machine learning processes; An identification means for identifying one or more executed machine learning processes similar to the target machine learning process using the target data and the executed data; An extraction means for extracting one or more machine learning processes associated with parameters that satisfy a predetermined condition from the one or more executed machine learning processes identified by the identification means; a priority calculation means for calculating a priority of each of one or more machine learning related parameters extracted by the extraction means and assigning the calculated priority to the parameter; a generating unit for generating output data including identification information of the plurality of parameters and value ranges of the plurality of parameters, the output data including the priority in an identifiable manner; The extraction means Among the one or more executed machine learning processes identified by the identifying means, one or more machine learning processes in which the machine learning model outputs a loss parameter equal to or less than a predetermined threshold are extracted. Information processing device.

2. The priority calculation means Using the loss parameters output by the machine learning models included in each of the plurality of machine learning models extracted by the extraction means and parameters other than the loss parameters, a priority is assigned to each of the parameters other than the loss parameters according to the degree of contribution of the parameters other than the loss parameters to the improvement of the loss parameters. The information processing device according to claim 1 .

3. The priority calculation means Identifying the degree of contribution according to a comparison result between loss parameters output by the machine learning models included in each of the plurality of machine learning processes and a comparison result between parameters other than the loss parameters output by the machine learning models included in each of the plurality of machine learning processes. The information processing device according to claim 2 .

4. The output data is The parameter identification information and the parameter value ranges arranged in the order of priority are included. The information processing device according to claim 1 .

5. The output data is The parameter identification information and the parameter value ranges are arranged in the order specified by the user. The information processing device according to claim 1 .

6. a first acquisition means for acquiring target data, which is data related to machine learning; a second acquisition means for acquiring executed data, which is data relating to a plurality of executed machine learning processes; An identification means for identifying one or more executed machine learning processes similar to the target machine learning process using the target data and the executed data; An extraction means for extracting one or more machine learning processes associated with parameters that satisfy a predetermined condition from the one or more executed machine learning processes identified by the identification means; a priority calculation means for calculating a priority of each of one or more machine learning related parameters extracted by the extraction means and assigning the calculated priority to the parameter; a generating unit for generating output data including identification information of the plurality of parameters and value ranges of the plurality of parameters, the output data including the priority in an identifiable manner; The extraction means Among the one or more executed machine learning processes identified by the identifying means, one or more machine learning processes in which the machine learning model outputs a loss parameter equal to or less than a predetermined threshold are extracted. Information processing system.

7. one or more processors obtaining target data, which is data relevant to the targeted machine learning; Obtaining executed data, which is data relating to multiple executed machine learning processes; using the target data and the executed data to identify one or more executed machine learning procedures that are similar to the target machine learning procedure; extracting one or more machine learning processes associated with parameters that satisfy a predetermined condition from the identified one or more executed machine learning processes; Calculating a priority for each of the one or more extracted machine learning parameters and assigning the calculated priority to the parameter; generating output data including identification information of the plurality of parameters and value ranges of the plurality of parameters, the output data including the priority in an identifiable manner; Extracting one or more machine learning processes associated with parameters that satisfy a predetermined condition from the one or more identified executed machine learning processes includes: and extracting one or more machine learning processes from the identified one or more executed machine learning processes in which the machine learning model outputs a loss parameter equal to or less than a predetermined threshold. Information processing methods.

8. On the computer, a first acquisition process for acquiring target data, which is data related to machine learning; a second acquisition process for acquiring executed data, which is data relating to a plurality of executed machine learning processes; An identification process for identifying one or more executed machine learning processes similar to the target machine learning process using the target data and the executed data; An extraction process of extracting one or more machine learning processes associated with parameters that satisfy a predetermined condition from the identified one or more executed machine learning processes; a priority calculation process for calculating a priority of each of the one or more extracted machine learning parameters and assigning the calculated priority to the parameter; a generation process for generating output data including identification information of the plurality of parameters and value ranges of the plurality of parameters, the output data including the priority in an identifiable manner; The extraction process includes: and extracting one or more machine learning processes from the identified one or more executed machine learning processes in which the machine learning model outputs a loss parameter equal to or less than a predetermined threshold. program.

Images