Arithmetic program, arithmetic method, and information processing device

Abstract

To provide an arithmetic program, an arithmetic method, and an information processing device capable of improving model accuracy.SOLUTION: An arithmetic program causes a computer to repeatedly execute processes of: creating a plurality of Ising models by creating the Ising models for each of a plurality of characteristic values, on the basis of a training data group that includes a plurality of pieces of training data in which data and the plurality of characteristic values for the data are associated with each other; creating a combined Ising model by combining the plurality of Ising models; searching for the data of a recommended point for the combined Ising model; and adding, to the training data group, the training data that includes the data of the recommended point and the plurality of characteristic values for the data of the recommended point.SELECTED DRAWING: Figure 6

Description

[Technical field]

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

[0002] Techniques for performing optimization using models have been disclosed (see, for example, Patent Documents 1 to 3). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-96334 A [Patent Document 2] Special Publication No. 2023-507139 [Patent Document 3] Special Publication No. 2023-524236 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, it is possible to optimize the model by sampling using a QUBO format Ising model. In this case, when dealing with multiple characteristic values, the characteristics of each characteristic value cannot be reflected in the model, and the model accuracy may be reduced.

[0005] In one aspect, the present invention aims to provide a calculation program, a calculation method, and an information processing device that can improve model accuracy. [Means for solving the problem]

[0006] In one aspect, the calculation program causes a computer to repeatedly execute the following processes: a process of creating a plurality of Ising models by creating an Ising model for each of a plurality of characteristic values ​​based on a learning data group including a plurality of learning data in which data is associated with a plurality of characteristic values ​​for the data; a process of creating a composite Ising model by combining the plurality of Ising models; a process of searching for recommendation point data for the composite Ising model; and a process of adding learning data including the recommendation point data and the plurality of characteristic values ​​for the recommendation point data to the learning data group. Effect of the Invention

[0007] The model accuracy can be improved. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 illustrates a solution search in QUBO format. [Diagram 2] 1 is a flowchart showing the execution procedure of a sampling technique using a QUBO format model. [Diagram 3] FIG. 2A is a block diagram illustrating an example of an overall configuration of an information processing device, and FIG. 2B is a block diagram illustrating an example of a hardware configuration of the information processing device. [Figure 4] 11 is a flowchart illustrating an example of an operation of the information processing device. [Diagram 5] FIG. 1A is a diagram illustrating an example of a training data group, and FIGS. 1B and 1C are diagrams illustrating examples of training data groups for each characteristic. [Figure 6] FIG. [Figure 7] 1A is a diagram illustrating an optimization problem for a magnetic shield, FIG. 1B is a diagram illustrating an arrangement of Gaussian basis functions, and FIG. 1C is a diagram illustrating magnetic flux lines generated from a coil. [Figure 8] FIG. 1 illustrates a positive and negative configuration. [Figure 9] FIG. 13 is a diagram illustrating a shape function. [Figure 10] FIG. 13 is a diagram showing a simulation result. [Figure 11] FIG. 13 is a diagram showing a simulation result. [Figure 12] 11 is a diagram illustrating an example in which a UI for assisting input has a data display function. [Figure 13] (a) and (b) are examples of graphs showing the frequency distribution of each characteristic value. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Binary variable sampling technology is used as a technique for searching for a good solution with a high evaluation value from a large number of combinations and sequences. Examples of binary variable sampling technology include sampling technology that performs random sampling and sampling technology that uses the Ising model in the QUBO format.

[0010] Although the random sampling technique allows easy sampling, it has the drawback that sampling efficiency is low and a large number of samplings must be performed to obtain a good solution with high accuracy.

[0011] An example of a sampling technique that uses a QUBO-format model is FMQA (Factorization Machine with Quantum Annealing). FMQA is a method that combines QA (Quantum Annealing) and FM (Machine Learning). FMQA creates a QUBO-format FM model from training data, finds a good answer through QA, analyzes the evaluation value of the good answer with a solver, adds the result to the training data, and performs interactive sampling. Another sampling technique that uses a QUBO-format model is FMDA technology. FMDA replaces the QA part of FMQA with DA (Digital Annealer).

[0012] Here, the QUBO format stands for Quadratic Unconstrained Binary Optimization, which is a format that allows binary optimization without quadratic constraints. The QUBO format can be expressed, for example, as in the following formula. Note that x i = 0 or 1 for i = 1, ..., N. ij x i and x j is the coupling coefficient of i x i is the bias coefficient. The first term on the right hand side is a quadratic term and represents the interaction. The second term on the right hand side is a linear term and represents the bias action. The third term on the right hand side is a constant term. In the QUBO format, a good solution x is searched for to minimize E(x), which represents the energy, according to the following formula, as illustrated in Figure 1.

number

[0013] Figure 2 is a flowchart showing the execution procedure of the sampling technique using the Ising model in the QUBO format. Here, we explain the sampling technique when optimizing multiple characteristic values. A training data set is generated by randomly generating the initial points of the training data. The training data is x=(x1, x2, …, x M ) format as data. Each value of the vector data is a binary value of 0 or 1. The number of training data to be generated is determined by the user's settings. Next, using a solver, N ) is calculated. Next, weights are assigned to each characteristic value and added together. For example, the objective function f=w1y1+w2y2+…+w N y N The evaluation value is calculated as follows. The evaluation value is an index for judging whether the initial point and the recommendation point described later are good or not. By the above steps, multiple initial learning data sets each consisting of the data of the initial point and the evaluation value are generated and used as a learning data group.

[0014] Next, an FM model is created using the training data group, and an Ising model in QUBO format is created by converting it to QUBO. Since FM is in QUBO format, creating an FM is equivalent to creating an Ising model in QUBO format. Other machine learning models can also be used as long as they can create an Ising model in QUBO format.

[0015] Next, DA is used to optimize the created QUBO format Ising model, and a good solution (DA recommendation point) with the best evaluation value is generated.

[0016] Next, a solver is used to analyze the DA recommendation points and calculate the characteristic values ​​(y1, y2, …, y N ) is calculated.

[0017] Next, using the procedure described above, weights are assigned to each characteristic value and added together to obtain the objective function f=w1y1+w2y2+…+w N y N is calculated as the evaluation value.

[0018] Next, the evaluation result (a set of recommendation points and evaluation value) is added to the learning data group as learning data. By repeating this series of processes, the learning data group is updated and an optimal solution can be obtained.

[0019] In addition, although FMDA has been explained in Fig. 2, the DA part of FMDA may be replaced with QA, and other methods may be used as long as the Ising machine can solve QUBO. In addition to using an Ising machine dedicated to solving QUBO, the Ising model may be solved using software.

[0020] In such processing, the sampling performance depends on the accuracy of the model used in the sampling. Specifically, in the case of a problem that handles multiple characteristic values ​​as described above, if the characteristic values ​​are added together to form a single objective variable value and a model is created for the generated objective variable value, the features of each characteristic value may not be reflected in the model. In this case, the accuracy of the model may decrease, and as a result, the number of samplings may increase.

[0021] Therefore, in the following embodiment, an example in which the model accuracy can be improved will be described. EXAMPLES

[0022] Fig. 3(a) is a block diagram illustrating an example of the overall configuration of an information processing device 100. As illustrated in Fig. 3(a), the information processing device 100 includes a storage unit 10, an initial point generation unit 20, an evaluation unit 30, an FMDA execution unit 40, a learning data update unit 50, an output unit 60, and the like.

[0023] Fig. 3(b) is a block diagram illustrating an example of a hardware configuration of the information processing device 100. As illustrated in Fig. 3(b), the information processing device 100 includes a CPU 101, a RAM 102, a storage device 103, an input device 104, a display device 105, and the like.

[0024] The CPU (Central Processing Unit) 101 is a central processing unit. The CPU 101 includes one or more cores. The RAM (Random Access Memory) 102 is a volatile memory that temporarily stores a program executed by the CPU 101, data processed by the CPU 101, and the like. The storage device 103 is a non-volatile storage device. For example, a ROM (Read Only Memory), a solid state drive (SSD) such as a flash memory, or a hard disk driven by a hard disk drive can be used as the storage device 103. The storage device 103 stores a calculation program. The input device 104 is an input device such as a keyboard or a mouse. The display device 105 is a display device such as an LCD (Liquid Crystal Display). The CPU 101 executes the calculation program to realize a storage unit 10, an initial point generation unit 20, an evaluation unit 30, an FMDA execution unit 40, a learning data update unit 50, an output unit 60, and the like. In addition, the storage unit 10, the initial point generation unit 20, the evaluation unit 30, the FMDA execution unit 40, the learning data update unit 50, the output unit 60, and the like may be implemented using hardware such as dedicated circuits.

[0025] FIG. 4 is a flowchart showing an example of the operation of the information processing device 100. As illustrated in FIG. 4, the initial point generation unit 20 generates an initial learning data group by randomly generating initial points of learning data (step S1). In step S1, the user may specify a pattern to some extent, or a pattern based on a preset rule may be used. For example, a pattern of all 1's, a pattern of alternating 10's, or a pattern of alternating 01's. The number of learning data to be generated is determined by the user's settings. In addition, the initial points of the learning data may be set by the user. The data of the initial points is x=(x1, x2, ..., x M ) format, where each value is a binary value of either 0 or 1.

[0026] Next, the evaluation unit 30 uses a solver to calculate each characteristic value (y1, y2, ..., y N) is calculated (step S2).

[0027] Next, the learning data update unit 50 adds the initial learning data, which is a set of the data of the initial point obtained in step S1 and the characteristic value calculated in step S2, to the learning data group stored in the storage unit 10 (step S3). Fig. 5(a) is a diagram illustrating each learning data included in the learning data group.

[0028] Next, the FMDA execution unit 40 divides the learning data group into data for each characteristic value (step S4). For example, Fig. 5(b) is data for characteristic value y1, and Fig. 5(c) is data for characteristic value y2.

[0029] Next, the FMDA execution unit 40 creates an FM model for each characteristic value obtained in step S4 (step S5). Thus, an FM model for characteristic value y1, an FM model for characteristic value y2, an FM model for characteristic value y3, and so on are created.

[0030] Next, the FMDA execution unit 40 converts each FM model created in step S5 into a QUBO model, and each Ising model E1(x), E2(x), ..., E N (x) is created (step S6). Since the FM is in the QUBO format, creating the FM is equivalent to creating an Ising model in the QUBO format. Other machine learning models can also be used as long as they can create an Ising model in the QUBO format.

[0031] Next, the FMDA execution unit 40 calculates the Ising models E1(x), E2(x), ..., E N Each weight w1, w2, ... w for (x) N is corrected (step S7).

[0032] Next, the FMDA execution unit 40 uses the weights obtained in step S7 to add up multiple QUBOs to create a composite QUBO (step S8). For example, w1E1(x)+w2E2(x)+...+w N E N Let (x) be the composite QUBO. There is no particular restriction on the weight value, but if arithmetic mean is used, 0≦w1,w2,…w N ≦1 and w1+w2+…+w N =1.

[0033] Next, the FMDA execution unit 40 generates DA recommendation points for the number of DA setting recommendations by optimizing the composite QUBO created in step S8 using DA (step S9). The number of DA setting recommendations is set in advance by the user. The DA recommendation points are good solutions (recommendation points) with the best evaluation value. Alternatively, the DA recommendation points are good solutions (recommendation points) with evaluation values ​​equal to or greater than a threshold value. Alternatively, the DA recommendation points are good solutions (recommendation points) with evaluation values ​​in the top specified range.

[0034] Next, the FMDA execution unit 40 determines whether the recommendation point obtained in step S9 is a point that has already been searched (step S10). For example, it determines whether the recommendation point obtained in step S9 is included in the learning data group stored in the storage unit 10.

[0035] If the answer is “Yes” in step S10, the FMDA execution unit 40 calculates the Ising models E1(x), E2(x), . N Each weight w1, w2, ... w for (x) N At least a part of the above is changed (step S11). After that, the process is executed again from step S7.

[0036] If the result of the determination in step S10 is "No," the evaluation unit 30 uses a solver to calculate each characteristic value of the recommendation point data (step S12).

[0037] Next, the learning data update unit 50 adds the evaluation result (a set of recommendation point data and each characteristic value) to the learning data group as learning data (step S13).

[0038] Next, the FMDA execution unit 40 judges whether the number of iterations has reached an upper limit (step S14). The number of times step S14 is executed may be set as the number of iterations. The upper limit of the number of iterations is preset by the user.

[0039] If the determination in step S14 is "No", the process is executed again from step S4. If the determination in step S14 is "Yes", the execution of the flow chart ends.

[0040] The output unit 60 outputs the result of the processing in Fig. 4. The output result is displayed, for example, by the display device 105. For example, the output unit 60 may output the contents of the learning data group, or may output learning data with a high evaluation value from the learning data group as a good answer.

[0041] FIG. 6 is a diagram summarizing this embodiment. As illustrated in FIG. 6, an initial training data group is generated. The initial training data group is generated by using the data x=(x1, x2, ..., x M ) and each characteristic value (y1, y2, …, y N ) are included.

[0042] Next, as shown in Figure 5(b) and Figure 5(c), the training data group is divided into data for each characteristic value, an FM model is created for each characteristic value, and then QUBO is generated to create an Ising model. Each model is weighted and added together to create a composite QUBO.

[0043] A recommendation point is generated by DA optimization using the composite QUBO. If the recommendation point has already been searched, a new composite QUBO is created by changing the weights of each model, and a recommendation point is generated by DA optimization. If the recommendation point has not been searched, the recommendation point is analyzed by a solver, and the analysis result of the recommendation point is added to the training data. The same process is then repeated with the updated training data.

[0044] According to this embodiment, an Ising model is created for each of a plurality of characteristic values, and then a composite Ising model is created. In this way, the characteristics of each characteristic value can be reflected in the composite Ising model. This can improve the model accuracy, and as a result, the number of sampling times can be reduced.

[0045] (Simulation results) Hereinafter, a description will be given of the results of a simulation in which a hypothetical problem was set and the computational processing according to the above embodiment was performed. FIG. 7(a) is a diagram illustrating an optimization problem of a magnetic shield. FIG. 7(a) is a diagram illustrating a magnetic shield analysis model, and illustrating a coil, a design area of ​​a magnetic body, and a target area. The target area is an area where it is desired to prevent the influence of magnetic flux. The design area of ​​the magnetic body is an area for placing the magnetic body. FIG. 7(b) is a diagram illustrating the placement of Gaussian basis functions. As illustrated in FIG. 7(b), 96 magnetic bodies can be placed in the design area of ​​the magnetic body. FIG. 7(c) is a diagram illustrating magnetic flux lines generated from a coil.

[0046] The table in FIG. 8 is a diagram illustrating positive and negative placements. s1 to s96 indicate 96 positive placement locations. "0" in s1 to s96 indicates that no magnetic material is placed, and "1" indicates that a magnetic material is placed. z1 to z96 indicate 96 negative placement locations. "0" in z1 to z96 indicates that no magnetic material is placed, and "1" indicates that a magnetic material is placed.

[0047] This placement problem can be expressed as a shape function y as follows:i ,z i ∈{0,1}, and w i ∈{-1,0,1}. In this way, the optimization problem of the magnetic shield is viewed as a placement problem of positive and negative Gaussian basis functions.

number

[0048] The distribution of shape function y is shown in the upper diagram of Figure 9. From this result, it becomes possible to determine material information from the magnitude of shape function y, as shown in the lower diagram of Figure 9. In the lower diagram of Figure 9, if the shape function y is 0 or more, material is present, and if the shape function y is less than 0, material is not present.

[0049] The purpose of optimizing the magnetic shield shape is to determine the average magnetic flux density B ave T and the magnetic area S mag Specifically, the following equation is minimized. Note that E1 is the B ave T E2 is the S when the entire design area is magnetic. mag The characteristic value y1 is B ave T / E1, and the characteristic value y2 is S mag / E2.

number

[0050] For this optimization problem, sampling was performed using two methods, a comparative example and this embodiment. As a common setting, the number of initial points was set to 192. The number of iterations was set to 96, the weight w1 of the characteristic value y1 was set to 0.5, and the weight w2 of the characteristic value y2 was set to 0.5. However, as in step S11 of FIG. 4, if the recommendation point has already been searched, each weight may change. In this embodiment, an Ising model was created for each characteristic value from the learning data group, and then synthesized, and optimization was performed on the synthesized Ising model. In the comparative example, an Ising model for the objective function = w1y1+w2y2 was created and optimization was performed.

[0051] 10 and 11 are diagrams showing the results. FIG. 10 shows the best value found by the search. The vertical axis represents the energy E in the QUBO format. Therefore, the smaller the value on the vertical axis, the better the result. The number of CAE analyses on the horizontal axis represents the number of iterations. The lower diagram in FIG. 10 is an enlarged diagram in which the scale of the vertical axis of the upper diagram in FIG. 10 has been changed. As shown in FIG. 10 and FIG. 11, it can be seen that the method of this embodiment can sample more efficiently than the method of the comparative example.

[0052] Next, an example of creating a synthetic Ising model from an Ising model for each characteristic value will be described. For example, the weight of each Ising model may be set based on a hypothesis assumed in advance by a user or on limited characteristic value data in advance. For example, statistical values ​​or mathematical features such as the average value, maximum value, and minimum value of the characteristic value may be used as the weight ratio of each Ising model.

[0053] For example, when the number of execution cycles of steps S4 to S14 in FIG. 4 increases, the number of learning data included in the learning data group increases, and the number of characteristic values ​​to be calculated also increases. In this case, the data group of characteristic values ​​may deviate from the mathematical characteristics assumed in advance. In particular, since multiple characteristic values ​​are used in this embodiment, it may be difficult to assume the weight of each characteristic value in advance compared to the case where one characteristic value is used. Therefore, the user may specify a method of weight correction, so that the weights may be automatically corrected.

[0054] For example, it is possible to change the weights during the execution cycle by rewriting the set value data of the weights referred to by the optimization algorithm during the execution cycle of steps S4 to S14 in Fig. 4. However, in this case, it is preferable for the user to monitor and analyze the characteristic value data as appropriate.

[0055] Therefore, for example, it is possible to realize automatic correction of weights based on a correction method set in advance by a user by automatically correcting weights from the mathematical characteristics of characteristic values ​​without the user having to change the weights. For example, the setting of this automatic correction of weights can be specified using a UI (User Interface) that assists the user in setting weights. In the UI that assists the setting, the user can check the status of characteristic values, and weights and corrections may be changed even during an execution cycle. The UI that assists the setting may include a UI that can display and check stored characteristic values ​​and data distributions using graphs. In addition, a conversational AI technology that assists UI input may be used.

[0056] This section explains the UI that assists in inputting weight settings when creating a synthetic Ising model. For example, the case where a user sets "2" as the weight w2 for y2 based on prior assumptions (the average values ​​of y1 and y2 are the same) is explained. In this example, the user specifies that even if the relationship (ratio) of the average values ​​changes during the execution cycle, the weight "2" is automatically corrected for the average value.

[0057] The specific correction content is as follows. When f=w1y1+w2y2, let the correction value be α, then let w1=1 and w2=2×α. α is the correction value that satisfies p1:αp2=1:x, where p1 and p2 are the average values ​​of y1 and y2. x represents the expected value after correction. For example, when x=1, this indicates that the average value of y2 after correction will be the same as y1. For example, if the user specifies the expected value x=1, when p1 / p2=1.5 during the execution cycle, the correction value α will be 1.5, so the average value of y1 and y2 will be 1×y1+2×1.5y2, and by multiplying the average value of y2 by 1.5, the prior assumption and the assumption for the average value can be weighted similarly.

[0058] FIG. 12 is a diagram illustrating a case where the input-assist UI is provided with a data display (graph display, etc.) function. In the example of FIG. 12, it is specified that automatic correction is not performed for y1. This is an example of a UI in which a correction equivalent to the average value p2 of y2 being corrected to the average value p1 of y1, which satisfies p1:p2=1:1, is performed. For example, it may be possible to display a graph by clicking on the "Graph" column in FIG. 12.

[0059] FIG. 13(a) is an example of a graph displaying the frequency distribution of characteristic value y1. FIG. 13(b) is an example of a graph displaying the frequency distribution of characteristic value y2. In these examples, statistical data is also displayed. FIG. 13(a) displays the frequency distribution of the initial data (y1 init) and all data (y1 all) including the data after learning. FIG. 13(b) displays the data after correction (y2 corrected) in addition to the initial data (y2 init) and all data (y2 all). In FIG. 13(a), the average value p1 of y1 of the initial data (y1 init) is 0.50, and the average value p1 of y1 of all data (y1 all) including the data after learning is 0.54. In Fig. 13(b), the average value p2 of y2 in the initial data (y2 init) is 0.85, the average value p2 of y2 in all data including the learned data (y2 all) is 0.71, and the average value p2 of y2 in the corrected data (y2 corrected) is 0.54. Thus, for y2, the average of the corrected characteristic values ​​is the same as that of y1.

[0060] In the above example, the FMDA execution unit 40 is an example of an execution unit that creates multiple Ising models by creating an Ising model for each of multiple characteristic values ​​from a learning data group including multiple learning data in which data is associated with multiple characteristic values ​​for the data, creates a composite Ising model by synthesizing the multiple Ising models, and executes a search for recommendation point data for the composite Ising model. The learning data update unit 50 is an example of an update unit that adds learning data including recommendation point data and multiple characteristic values ​​for the recommendation point data to the learning data group and updates the learning data group.

[0061] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to such specific embodiments, and various modifications and variations are possible within the scope of the gist of the present invention described in the claims. (Appendix 1) On the computer, A process of creating a plurality of Ising models by creating an Ising model for each of the plurality of characteristic values ​​based on a training data group including a plurality of training data in which data is associated with a plurality of characteristic values ​​for the data; A process of creating a composite Ising model by synthesizing the plurality of Ising models; A process of searching for recommendation point data for the synthetic Ising model; adding learning data including the recommendation point data and the plurality of characteristic values ​​for the recommendation point data to the learning data group. (Appendix 2) The computer includes: 2. The arithmetic program according to claim 1, further comprising a program for executing a process of creating the composite Ising model by weighting and adding up the plurality of Ising models. (Appendix 3) The computer includes: 3. The computing program according to claim 2, further comprising: a program for executing a process of correcting the weights in accordance with the plurality of characteristic values. (Appendix 4) The arithmetic program according to claim 1, characterized in that, when the searched-for recommendation point is already included in the learning data group, the synthetic Ising model is recreated when the multiple Ising models used to search for the recommendation point are synthesized, and the recommendation point is searched for in the recreated synthetic Ising model. (Appendix 5) The arithmetic program according to claim 1, wherein the data is vector data of binary values. (Appendix 6) The computer A process of creating a plurality of Ising models by creating an Ising model for each of the plurality of characteristic values ​​based on a training data group including a plurality of training data in which data is associated with a plurality of characteristic values ​​for the data; A process of creating a composite Ising model by synthesizing the plurality of Ising models; A process of searching for recommendation point data for the synthetic Ising model; adding learning data including the data of the recommendation points and the plurality of characteristic values ​​for the data of the recommendation points to the learning data group. (Appendix 7) The computer, The calculation method according to claim 6, further comprising the step of performing a process of generating the composite Ising model by weighting and adding up the plurality of Ising models. (Appendix 8) The computer, The calculation method according to claim 7, further comprising the step of correcting the weights in accordance with the plurality of characteristic values. (Appendix 9) The calculation method according to claim 6, characterized in that, when the searched recommendation point is already included in the learning data group, the synthetic Ising model is recreated when the multiple Ising models used to search for the recommendation point are synthesized, and the recommendation point is searched for in the recreated synthetic Ising model. (Appendix 10) The calculation method according to claim 6, wherein the data is vector data of binary values. (Appendix 11) An execution unit that creates a plurality of Ising models by creating an Ising model for each of the plurality of characteristic values ​​based on a learning data group including a plurality of learning data in which data and a plurality of characteristic values ​​for the data are associated with each other, creates a composite Ising model by synthesizing the plurality of Ising models, and executes a search for data of recommendation points for the composite Ising model; an updating unit that adds learning data including the recommendation point data and the plurality of characteristic values ​​for the recommendation point data to the learning data group, and updates the learning data group. (Appendix 12) The information processing device according to claim 11, wherein the execution unit creates the composite Ising model by weighting and adding up the multiple Ising models. (Appendix 13) The information processing device according to claim 12, wherein the execution unit corrects the weights in accordance with the plurality of characteristic values. (Appendix 14) The information processing device according to claim 11, wherein, when the searched recommendation point is already included in the learning data group, the execution unit recreates the composite Ising model when synthesizing the multiple Ising models used to search for the recommendation point, and searches for the recommendation point in the recreated composite Ising model. (Appendix 15) The information processing device according to claim 11, wherein the data is vector data of binary values. [Explanation of symbols]

[0062] 10 Storage area 20 Initial point generator 30 Evaluation Section 40 FMDA Executive Division 50 Learning data update unit 60 Output section 100 Information processing device 101 CPU 102 RAM 103 Storage device 104 Input Devices 105 Display device

Claims

1. On the computer, A process of creating a plurality of Ising models by creating an Ising model for each of the plurality of characteristic values ​​based on a training data group including a plurality of training data in which data is associated with a plurality of characteristic values ​​for the data; A process of creating a composite Ising model by synthesizing the plurality of Ising models; A process of searching for recommendation point data for the synthetic Ising model; adding learning data including the recommendation point data and the plurality of characteristic values ​​for the recommendation point data to the learning data group.

2. The computer includes:

2. The computing program according to claim 1, further comprising a step of executing a process of creating the composite Ising model by weighting and adding up the plurality of Ising models.

3. The computer includes:

3. The computing program according to claim 2, further comprising a step of correcting the weights in accordance with the plurality of characteristic values.

4. 2. The arithmetic program according to claim 1, wherein, when the searched-for recommendation point is already included in the learning data group, when the plurality of Ising models used to search for the recommendation point are synthesized, the synthetic Ising model is recreated, and the recommendation point is searched for in the recreated synthetic Ising model.

5. 2. The arithmetic program according to claim 1, wherein the data is vector data of binary values.

6. The computer A process of creating a plurality of Ising models by creating an Ising model for each of the plurality of characteristic values ​​based on a training data group including a plurality of training data in which data is associated with a plurality of characteristic values ​​for the data; A process of creating a composite Ising model by synthesizing the plurality of Ising models; A process of searching for recommendation point data for the synthetic Ising model; adding learning data including the data of the recommendation points and the plurality of characteristic values ​​for the data of the recommendation points to the learning data group.

7. An execution unit that creates a plurality of Ising models by creating an Ising model for each of the plurality of characteristic values ​​based on a learning data group including a plurality of learning data in which data and a plurality of characteristic values ​​for the data are associated with each other, creates a composite Ising model by synthesizing the plurality of Ising models, and executes a search for data of recommendation points for the composite Ising model; an updating unit that adds learning data including the recommendation point data and the plurality of characteristic values ​​for the recommendation point data to the learning data group, and updates the learning data group.

Images