Information processing device, information processing method, and program

Abstract

An information processing device 100 according to the present disclosure comprises: an input unit 121 that receives an input case represented by a vector made up of a plurality of element values, and a prediction model that outputs a prediction with respect to the plurality of element values which are input; a generation unit 122 that generates a correction vector represented by a vector made up of a correction amount with respect to the element values in the input case; and a calculation unit 123 that calculates a prediction, using the prediction model, for a virtual case representing a state resulting from element values on the correction vector associated with the input case, the virtual case being between the input case and a corrected case obtained by correcting the input case with the correction vector.

Description

Information processing device, information processing method, and program

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

[0002] Predictions are made for input data representing cases using machine learning models in various fields. In this context, proposed modifications to input data that change predictions made by machine learning models are also generated. For example, Patent Literature 1 describes the generation of proposed modifications to input data that result in a desired prediction.

[0003] International Publication No. 2022 / 034685

[0004] However, with the technology of Patent Document 1 described above, although it is possible to obtain a proposed revision of input data that may result in a desired prediction, it is not possible to recognize other proposed revisions of input data that may be related to the desired prediction, and this creates a problem in that it is difficult to make further effective use of the proposed revisions and the predictions.

[0005] Therefore, the purpose of the present disclosure is to solve the above-mentioned problem that even if a revision proposal is generated for a case where a prediction may be changed, it is not possible to make further effective use of the revision proposal and the prediction.

[0006] An information processing device according to one aspect of the present disclosure includes: an input unit that receives an input example represented by a vector consisting of a plurality of feature values ​​and a prediction model that outputs a prediction for the plurality of input feature values; a generation unit that generates a correction vector that represents a correction to the feature values ​​of the input example as a vector; and a calculation unit that calculates a prediction using the prediction model for a virtual example that represents a state between the input example and a corrected example obtained by correcting the input example with the correction vector, and that is made up of feature values ​​on the correction vector linked to the input example. Also, an information processing method according to one aspect of the present disclosure includes: receiving an input example represented by a vector consisting of a plurality of feature values ​​and a prediction model that outputs a prediction for the plurality of input feature values; generating a correction vector that represents a vector that represents the correction to the feature values ​​of the input example; and calculating a prediction using the prediction model for a virtual example that represents a state between the input example and a corrected example obtained by correcting the input example with the correction vector, and that is made up of feature values ​​on the correction vector linked to the input example. Furthermore, a program according to one aspect of the present disclosure is configured to cause a computer to execute the following processes: accept an input example represented by a vector consisting of multiple feature values ​​and a prediction model that outputs a prediction for the multiple input feature values; generate a correction vector that represents a correction to the feature values ​​of the input example as a vector; and calculate a prediction using the prediction model for a virtual example that represents a state consisting of the feature values ​​on the correction vector linked to the input example, between the input example and a corrected example obtained by correcting the input example with the correction vector.

[0007] With the above-described configuration, when a correction proposal for a case in which a prediction may be changed is generated, the present disclosure can achieve more effective use of the correction proposal and the prediction.

[0008] FIG. 1 is a block diagram showing a configuration of an information processing device according to the present disclosure. FIG. 2 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 3 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 4 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 5 is a flowchart showing processing operations of an information processing device according to the present disclosure. FIG. 6 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 7 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 8 is a diagram showing a state of processing by an information processing device according to the present disclosure. FIG. 9 is a block diagram showing a hardware configuration of an information processing device according to the present disclosure. FIG. 10 is a block diagram showing a configuration of an information processing device according to the present disclosure.

[0009] First Embodiment The present disclosure will be described with reference to the drawings, which may be relevant to any of the embodiments.

[0010] The information processing device 10 of this embodiment is used to make predictions on input data using a prediction model. In particular, this embodiment makes predictions on examples containing multiple features, which are input data, generates a correction vector consisting of feature correction amounts to improve the prediction, and calculates predictions using the prediction model for virtual cases on this correction vector. At this time, this embodiment also outputs changes in the features of the virtual cases along with changes in the predictions corresponding to changes in the virtual cases on the correction vector.

[0011] As an example, this embodiment describes a case where the financial condition of a business is used as a case, and the multiple features constituting the case include the business's income, liabilities, labor costs, fixed costs, etc., and a prediction model is used to predict the bankruptcy probability of the business based on these multiple features. Specifically, the system calculates and outputs a change in the bankruptcy probability, which is a prediction associated with changes to the case due to modifications to multiple features, and also outputs changes in features such as income and fraud. This allows the system to recognize changes in the predicted bankruptcy probability, and further recognize changes in features such as income and liabilities associated with changes in the predicted bankruptcy probability. As a result, the system can more effectively utilize proposed modifications to the case and their predictions. Note that the information processing device 10 is not limited to the case and its predictions illustrated in this embodiment, but can also be applied to various predictions for various cases.

[0012] The information processing device 10 is composed of one or more information processing devices each including a calculation device and a storage device. As shown in FIG. 1 , the information processing device 10 includes an input unit 11, a generation unit 12, a calculation unit 13, and an output unit 14. The functions of the input unit 11, the generation unit 12, the calculation unit 13, and the output unit 14 can be realized by the calculation device executing a program for realizing each function stored in the storage device. The information processing device 10 also includes a model storage unit 16 and a case storage unit 17. The model storage unit 16 and the case storage unit 17 are each composed of a storage device. Each component and its operation will be described in detail below.

[0013] The input unit 11 accepts input of a prediction model that outputs predictions for cases and stores the prediction model in the model storage unit 16 (step S1 in FIG. 1 ). The prediction model is a machine learning model that outputs prediction data in response to input of case data. Therefore, the prediction model is generated by machine learning using training data that includes pre-prepared case data and result data. In this embodiment, the case data includes multiple feature data items that represent the business's management status, such as income, liabilities, labor costs, and fixed costs. The result data is data that represents the results of management, such as whether or not the business has gone bankrupt or a calculated probability of bankruptcy. By performing machine learning using such training data, the prediction model is generated so as to output prediction data that is the probability of bankruptcy from case data that includes multiple feature data items that represent the business status.

[0014] The input unit 11 also accepts input of input case data represented by a vector consisting of multiple feature values ​​and stores the input case in the case storage unit 17 (step S1 in FIG. 6 ). As described above, the input case data in this embodiment is composed of multiple feature data representing the business management status, and each feature data is, for example, income, liabilities, labor costs, fixed costs, etc. An example of input case data is shown in FIG. 2 . In this example, the feature data of the input case data are assumed to be income "500,000 yen," liabilities "1.5 million yen," labor costs "1 million yen," etc. Since the input case data is thus composed of multiple feature data, it can be represented as an input case vector with each feature data as an element.

[0015] The input unit 11 also accepts input of conditions that must be satisfied by the prediction made by the prediction model for the input case data. In this embodiment, the prediction made by the prediction model outputs the probability of bankruptcy for the input case data, and conditions for the value of this probability of bankruptcy are input. In this case, the value of the probability of bankruptcy that serves as the condition is a value that is generally considered desirable, and is set to, for example, "10% or less." Note that the input condition may be any value, and when the prediction model predicts whether or not bankruptcy will occur, the condition may be "bankrupt" or "not bankrupt."

[0016] The generation unit 12 generates a correction vector represented by a vector including correction amounts for the values ​​of each feature data of the above-mentioned input example data (step S2 in FIG. 6). In this embodiment, the generation unit 12 generates a correction vector whose elements are correction amounts, which are the correction amounts for each value of the above-mentioned input example data, i.e., income "500,000 yen," debt "1.5 million yen," and labor costs "1 million yen." For example, in the example shown in FIG. 2, a correction vector whose elements are values ​​such as the income correction amount "+300,000 yen," the debt correction amount "-600,000 yen," and the labor cost correction amount "-200,000 yen" is generated.

[0017] In this case, for example, the generation unit 12 inputs corrected case data obtained by correcting the input case data into the prediction model, and generates a correction vector so that the bankruptcy probability predicted by the prediction model for the corrected case data satisfies the condition input as described above (e.g., 10% or less). In the example of FIG. 2 , the generation unit 12 generates corrected case data by correcting each feature data of the input case data, such as income of "800,000 yen," debt of "900,000 yen," and labor costs of "800,000 yen." If the bankruptcy probability predicted by inputting such corrected case data into the prediction model is 10% or less, the generation unit 12 generates a correction vector whose elements are the correction amounts of each feature data at this time: income of "+300,000 yen," debt of "-600,000 yen," and labor costs of "-200,000 yen." However, the generation unit 12 may generate the correction vector by any method. For example, the generation unit 12 may generate a correction vector by setting an arbitrary correction amount for each feature data, regardless of whether the prediction by the prediction model satisfies the condition.

[0018] The calculation unit 13 inputs hypothetical case data representing the business status of the business represented by the values ​​of each element data on the correction vector into the prediction model and calculates a prediction, which is the bankruptcy probability for the hypothetical case data (step S3 in FIG. 6 ). Here, when each case is represented by a vector with each feature data as an element, as shown in the left diagram of FIG. 3 , a corrected case vector corresponding to the corrected case data obtained by correcting the input case data by the correction amount of the correction vector can be expressed by concatenating the correction vector with the input case vector corresponding to the input case data. In other words, the correction vector is located between the input case vector and the corrected case vector. Therefore, it can be said that an arbitrary point on the correction vector represents the business status of the business during the process of change from the business status of the business in the input case data to the business status in the corrected case data. The calculation unit 13 generates hypothetical case data consisting of each feature data representing the business status of the business at an arbitrary point on the correction vector, and inputs the hypothetical case data into the prediction model to calculate the predicted bankruptcy probability. Then, the calculation unit 13 calculates the bankruptcy probability for virtual case data representing the continuously changing business status of the business on the corrected vector, thereby calculating the change in bankruptcy probability, which is a prediction corresponding to the change in the business status of the business from the input case data to the corrected case data.

[0019] At this time, the calculation unit 13 uses continuously changing parameters to change the values ​​of the features that make up the virtual case data on the correction vector, thereby generating continuously changing virtual case data. For example, as shown in the left diagram of FIG. 4, the input case vector corresponding to the input case data is expressed as x=(x １ ,... ,x ｎ ), and the corrected case vector corresponding to the corrected case data is y=(y １ ,... ,y ｎ ), the virtual case vector xy on the corrected vector can be expressed as x+t*(y-x) using a parameter t (0≦t≦1). The calculation unit 13 can calculate a continuous change in the predicted bankruptcy probability by continuously changing the parameter t.

[0020] The output unit 14 outputs the change in bankruptcy probability, which is the prediction calculated as described above, to the display device of a predetermined information processing terminal for display (step S4 in Fig. 6). At this time, as shown in the right diagram in Fig. 3, the output unit 14 displays a coordinate space set up by corresponding the business management state of the business in the hypothetical case data to the horizontal axis (first axis) and the value of the bankruptcy probability, which is the prediction by the prediction model, to the vertical axis (second axis), and also displays a graph on this coordinate space that shows the change in the predicted bankruptcy probability relative to the change in the business management state of the business in the hypothetical case data on the correction vector.

[0021] The output unit 14 also displays graphs showing changes in the value of each feature data item due to changes in the business status of the business in the hypothetical case data on the corrected vector, corresponding to the graph of changes in bankruptcy probability described above (step S4 in Fig. 6 ). Specifically, as shown in the right diagram in Fig. 3 , the output unit 14 displays a coordinate space for each feature data item (income, debt, labor cost) in which the business status of the business in the hypothetical case data corresponds to the horizontal axis (third axis) and the value of the feature data corresponds to the vertical axis (fourth axis), and also displays a graph on this coordinate space showing changes in the value of each feature data item due to changes in the business status of the business in the hypothetical case data on the corrected vector.

[0022] Furthermore, when displaying each graph corresponding to each piece of feature data, the output unit 14 displays each coordinate space corresponding to each piece of feature data in parallel along a horizontal axis set to the business status of the business in the hypothetical case data. At this time, the output unit 14 also displays each coordinate space corresponding to each piece of feature data in parallel so that the horizontal axis of the coordinate space is also parallel to the horizontal axis of the coordinate space displaying the graph showing the change in bankruptcy probability. More specifically, as shown in the right diagrams of Figures 3 and 4 , the output unit 14 displays the coordinate space corresponding to the bankruptcy probability graph and the coordinate spaces corresponding to each piece of feature data in sequential order along the vertical axis so that their vertical axes are aligned on the same line. Furthermore, at this time, the output unit 14 displays the position of the value on the horizontal axis of each coordinate space in which the graph corresponding to each piece of feature data is displayed in correspondence with the position of the value on the horizontal axis of the coordinate interval in which the graph showing the change in predicted bankruptcy probability is displayed. In other words, as shown by the dotted line along the vertical axis in the right diagram of Figure 3, graphs of the bankruptcy probability (prediction), income, debt, and labor costs (characteristic data) are displayed in parallel, from top to bottom, with the same values ​​on the horizontal axis in each coordinate space aligned.

[0023] Here, each graph corresponding to each feature data item is a linear graph connecting the value of the input example data and the value of the corrected example data. As described above with reference to Figure 4, since the virtual example data on the corrected vector is expressed as x + t * (y - x) using the parameter t (0 ≤ t ≤ 1), the value of each feature data item is expressed as changing linearly with the change in the parameter t (0 ≤ t ≤ 1).

[0024] As described above, by displaying a graph of the predicted bankruptcy probability, it becomes easy to recognize changes in the prediction due to corrections to the case. Furthermore, by displaying a graph of the value of each feature data in correspondence with the graph of the predicted bankruptcy probability, it becomes easy to recognize the value of each feature data at a certain prediction. For example, as shown in FIG. 5 , if the bankruptcy probability in the input case is 80% and the bankruptcy probability in the corrected case is 10%, it is possible to recognize the values ​​of each feature data (income, debt, and labor costs) when the bankruptcy probability between them is 30%.

[0025] Second Embodiment Next, a second embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, another example of use of the information processing device described in the above embodiment will be described.

[0026] The information processing device 10 in this embodiment has the same configuration as the information processing device 10 shown in FIG. 1, and each component further has the following functions.

[0027] First, in this embodiment, the generation unit 12 is not limited to generating a single correction vector for correcting input case data, but generates a correction vector group consisting of a series of correction vectors that sequentially correct the input case data. For example, as shown in the left diagram of FIG. 7 , the generation unit 12 generates a series of correction vector groups consisting of a first correction vector 1 that corrects the input case data and a second correction vector 2 that further corrects the corrected case data to create corrected case data. In this case, the generation unit 12 generates a correction vector group consisting of any number of correction vectors that can be connected, and selects correction vector groups whose bankruptcy probabilities, which are predictions of the corrected case data corrected by the correction vector group, satisfy set conditions. However, the generation unit 12 may generate a correction vector group consisting of a number of correction vectors using any method.

[0028] The calculation unit 13 in this embodiment inputs hypothetical case data representing the business management state of the business represented by the values ​​of each element data on the correction vector group into the prediction model and calculates a prediction, which is the bankruptcy probability, for the hypothetical case data. At this time, the calculation unit 13 generates hypothetical case data consisting of feature data representing the business management state of the business at any point on each correction vector for each correction vector constituting the correction vector group, and inputs the hypothetical case data into the prediction model to calculate the predicted bankruptcy probability. As a result, the calculation unit 13 can calculate, for each correction vector, a change in the bankruptcy probability, which is a prediction corresponding to a change in the business management state on each correction vector, as shown in FIG. 7 .

[0029] At this time, the calculation unit 13 uses continuously changing parameters to change the values ​​of the features that make up the virtual case data on the correction vectors, as described above, to generate continuously changing virtual case data. For example, as shown in the left diagram of FIG. 8, if the correction vector group consists of two correction vectors 1 and 2, and the input case vector corresponding to the input case data is expressed as x=(x １ ,... ,x ｎ ), and the corrected case vector corresponding to the corrected case data is y=(y １ ,... ,y ｎ ), and furthermore, the input example vector is corrected by the correction vector 1 and the example vector is z=(z １ ,... ,z ｎ ) In this case, the hypothetical case vector xz on the correction vector 1 can be expressed as x+2t*(z-x) using the parameter t (0≦t≦½), and the hypothetical case vector zy on the correction vector 2 can be expressed as z+2(t-½)(y-z) using the parameter t (½≦t≦1). Then, the calculation unit 13 can calculate the continuous change in the predicted bankruptcy probability by continuously changing the parameter t within the range of (0≦t≦1).

[0030] Then, in the present embodiment, the output unit 14 displays and outputs a graph showing the change in bankruptcy probability, which is the calculated prediction, as described above. Also, as described above, the output unit 14 displays each graph showing the change in the value of each feature data, corresponding to the graph showing the change in bankruptcy probability. At this time, the output unit 14 displays each graph so that the graph corresponding to each correction vector can be recognized. For example, as shown in the right diagrams of FIGS. 7 and 8 , a graph may be displayed together with plots or symbols showing the input example x, the correction example y, and the corrected example z between them. Furthermore, the correction vector corresponding to the graph may be displayed clearly. In this way, by generating multiple correction vectors, the correction amount can be set in stages, and different changes in prediction depending on the correction amount can be recognized, thereby enabling effective use of predictions.

[0031] Here, when generating a correction vector group consisting of a plurality of correction vectors, the generation unit 12 in the present embodiment described above may generate a plurality of correction vector groups and select a correction vector group that satisfies a predetermined criterion from among the plurality of correction vector groups. Then, the calculation unit 13 in the present embodiment calculates the change in bankruptcy probability for the correction vector group selected by the generation unit 12 as described above.

[0032] As an example, the generation unit 12 may select a correction vector group from among multiple vector groups based on a norm representing the length of the correction vector included in the correction vector group. In particular, the generation unit 12 may select a correction vector group having the smallest sum of the norms of all correction vectors included in the correction vector group. For example, as shown in FIG. 9 , the generation unit 12 first generates a correction vector group consisting of multiple correction vectors linked from an input example to correction example A and a correction vector group consisting of multiple correction vectors linked from the input example to correction example B. In this case, the generation unit 12 calculates the sum of the norms of each correction vector included in each of the two correction vector groups and selects the correction vector group having the shortest length of the linked correction vectors from the input example to the correction example. In this way, by avoiding correction vectors with long lengths, correction vectors that require fewer corrections for the input example and are easier to achieve are more likely to be adopted.

[0033] Furthermore, the generation unit 12 may select a correction vector group from among multiple correction vector groups based on the number of correction vectors included in the correction vector group. In particular, the generation unit 12 may select the correction vector group with the smallest total number of correction vectors included in the correction vector group. For example, if the generation unit 12 generates two correction vector groups as shown in FIG. 9 as described above, the generation unit 12 calculates the total number of correction vectors included in each of the two correction vector groups and selects the correction vector group with the smallest total number. This improves the readability of each correction vector.

[0034] The generating unit 12 may set a predetermined upper limit on the number of correction vectors that the correction vector group may contain, and generate the correction vector group so that the number of correction vectors is less than the upper limit. Furthermore, when sequentially generating a plurality of correction vector groups, the generating unit 12 may generate further correction vector groups so that the number of correction vectors included in the correction vector groups already generated is smaller than that of the correction vector groups already generated.

[0035] 9 , the generation unit 12 may first generate one correction vector or a group of correction vectors that corrects the input example into a correction example that satisfies a predetermined prediction condition, as described above, and then generate multiple groups of different correction vectors that can be used to correct the input example into such a correction example. That is, multiple groups of correction vectors represented by routes that lead from the input example to a specific correction example may be generated. Then, the generation unit 12 may select a group of correction vectors from the generated groups of correction vectors based on the norm or number of correction vectors included, as described above.

[0036] Furthermore, when generating a group of correction vectors including multiple correction vectors, the generation unit 12 may generate the correction vectors such that each correction vector corrects some of the feature data included in the input example data. For example, as shown in FIG. 10 , if the total number of feature data included in the input example data is four, correction vectors 1 and 2 may be generated such that each correction vector corrects a smaller number of feature data than the total number, for example, two feature data. In this case, the generation unit 12 may further generate the correction vectors such that each correction vector corrects different feature data included in the input example data. For example, as shown in FIG. 10 , if the feature data included in the input example data is income, liabilities, labor costs, and fixed costs, correction vector 1 may correct income and liabilities, and correction vector 2 may correct labor costs and fixed costs, respectively. This may improve the interpretability of the graph corresponding to each correction vector.

[0037] Third Embodiment Next, a third embodiment of the present disclosure will be described with reference to the drawings. In this embodiment, an outline of the configuration of the information processing device described in the above embodiments is shown.

[0038] 11, the information processing device 100 of this embodiment is configured as a general information processing device, and is equipped with the following hardware configuration, for example: CPU (Central Processing Unit) 101 (arithmetic unit); ROM (Read Only Memory) 102 (storage device); RAM (Random Access Memory) 103 (storage device); programs 104 loaded into RAM 103; a storage device 105 storing the programs 104; a drive device 106 for reading and writing data from and to a storage medium 110 external to the information processing device; a communication interface 107 for connecting to a communication network 111 external to the information processing device; an input / output interface 108 for inputting and outputting data; and a bus 109 for connecting the various components.

[0039] 11 shows an example of the hardware configuration of the information processing device 100, and the hardware configuration of the information processing device is not limited to the above-described case. For example, the information processing device may be configured with only a part of the above-described configuration, such as excluding the drive device 106. Furthermore, the information processing device may use a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point Number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof, instead of the above-described CPU.

[0040] The information processing device 100 can be equipped with an input unit 121, a generation unit 122, and a calculation unit 123 shown in FIG. 12 by having the CPU 101 acquire and execute the program group 104. The program group 104 is stored in advance in the storage device 105 or the ROM 102, for example, and is loaded into the RAM 103 and executed by the CPU 101 as needed. The program group 104 may be supplied to the CPU 101 via the communication network 111, or may be stored in advance in the storage medium 110, and the drive device 106 may read the program and supply it to the CPU 101. However, the input unit 121, the generation unit 122, and the calculation unit 123 described above may be constructed using dedicated electronic circuits for realizing such means.

[0041] The input unit 121 receives input examples represented by vectors consisting of multiple feature values ​​and a prediction model that outputs predictions for the multiple input feature values. The generation unit 122 generates a correction vector represented by a vector consisting of correction amounts for the feature values ​​of the input examples. Furthermore, the calculation unit 123 calculates a prediction using the prediction model for a virtual example that represents a state between the input example and a corrected example obtained by correcting the input example with the correction vector, the state being composed of feature values ​​on the correction vector linked to the input example.

[0042] By configuring the present disclosure as described above, it becomes easier to recognize changes in predictions due to corrections to cases, and more effective use of correction proposals and their predictions can be achieved.

[0043] In addition, at least one or more of the functions of the above-mentioned input unit 121, generation unit 122, and calculation unit 123 may be executed by an information processing device installed and connected anywhere on the network, that is, they may be executed by so-called cloud computing.

[0044] The above-described program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-RWs, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program can also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can be supplied to a computer via wired communication paths such as electric wires and optical fibers, or via wireless communication paths.

[0045] Although the present disclosure has been described above with reference to the above-described embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each of the above-described embodiments can be combined with other embodiments as appropriate.

[0046] <Supplementary Notes> Some or all of the above embodiments may also be described as in the following supplementary notes. Below, an outline of the configurations of an information processing device, an information processing method, and a program according to the present disclosure will be described. However, the present disclosure is not limited to the following configurations. (Supplementary Note 1) An information processing device comprising: an input unit that accepts input examples represented by a vector consisting of a plurality of feature values ​​and a prediction model that outputs a prediction for the plurality of input feature values; a generation unit that generates a correction vector that represents a correction to the feature values ​​of the input example as a vector; and a calculation unit that calculates a prediction by the prediction model for a virtual example that represents a state consisting of the feature values ​​on the correction vector linked to the input example, between the input example and a corrected example in which the input example is corrected with the correction vector. (Supplementary Note 2) The information processing device according to Supplementary Note 1, wherein the calculation unit calculates a change in prediction by the prediction model in response to a change in the state of the virtual example on the correction vector linked to the input example. (Supplementary Note 3) The information processing device according to Supplementary Note 2, comprising an output unit that outputs a graph showing a change in prediction by the prediction model with respect to a change in the state of the virtual case on the correction vector linked to the input example, in a coordinate space with a first axis corresponding to the state of the virtual case and a second axis corresponding to the value of the prediction by the prediction model. (Supplementary Note 4) The information processing device according to Supplementary Note 3, wherein the output unit sets a coordinate space with a third axis corresponding to the state of the virtual case and a fourth axis corresponding to the value of the feature, corresponding to each of the features, and outputs a graph showing a change in the value of the feature with respect to a change in the state of the virtual case on the correction vector linked to the input example, in the coordinate space corresponding to each of the features. (Supplementary Note 5) The information processing device according to Supplementary Note 4, wherein the output unit outputs the coordinate spaces corresponding to each of the features so as to be displayed in parallel along the third axis.(Supplementary Note 6) The information processing device according to Supplementary Note 5, wherein the output unit outputs a coordinate space corresponding to each of the features so that the first axis and the third axis are parallel, and so that the coordinate spaces are displayed with the values ​​on the first axis and the values ​​on the third axis aligned in correspondence with each other. (Supplementary Note 7) The information processing device according to Supplementary Note 1, wherein the generation unit generates a correction vector group consisting of a series of correction vectors that sequentially correct the input example, and the calculation unit calculates a prediction by the prediction model for the virtual example consisting of feature values ​​on the correction vector group sequentially linked to the input example, between the input example and the correction example obtained by correcting the input example with the correction vector group. (Supplementary Note 8) The information processing device according to Supplementary Note 7, wherein the generation unit generates the correction vector group based on the norm of the correction vector included in the correction vector group. (Supplementary Note 9) The information processing device according to Supplementary Note 8, wherein the generation unit generates the correction vector group so that a sum of norms of the correction vectors included in the correction vector group is small. (Supplementary Note 10) The information processing device according to Supplementary Note 7, wherein the generation unit generates the correction vector group based on the number of correction vectors included in the correction vector group. (Supplementary Note 11) The information processing device according to Supplementary Note 10, wherein the generation unit generates the correction vector group so that the number of correction vectors included in the correction vector group is small. (Supplementary Note 12) The information processing device according to Supplementary Note 7, wherein the generation unit generates the correction vector group so that the correction vectors included in the correction vector group correct a number of features that is smaller than the total number of features included in the input example. (Supplementary Note 13) The information processing device according to Supplementary Note 7, wherein the generation unit generates the group of correction vectors such that each of the correction vectors included in the group of correction vectors corrects a different one of the features included in the input example.(Supplementary Note 14) The information processing device according to Supplementary Note 1, wherein the calculation unit calculates a prediction by the prediction model for the virtual case in which the feature values ​​on the correction vector are changed using a continuously changing parameter. (Supplementary Note 15) An information processing method comprising: receiving an input example represented by a vector consisting of a plurality of feature values ​​and a prediction model that outputs a prediction for the plurality of input feature values; generating a correction vector that represents a correction to the feature values ​​of the input example as a vector; and calculating a prediction by the prediction model for a virtual case that represents a state consisting of the feature values ​​on the correction vector linked to the input example, between the input example and a corrected example in which the input example is corrected with the correction vector. (Supplementary Note 16) The information processing method according to Supplementary Note 15, wherein a change in prediction by the prediction model in response to a change in the state of the virtual case on the correction vector linked to the input example is calculated. (Supplementary Note 17) The information processing method according to Supplementary Note 16, comprising outputting a graph showing a change in the prediction by the prediction model relative to a change in the state of the virtual case on the correction vector linked to the input example, in a coordinate space defined by a first axis corresponding to the state of the virtual case and a second axis corresponding to the value of the prediction by the prediction model. (Supplementary Note 18) A computer-readable storage medium storing a program causing a computer to execute processes of: accepting an input example represented by a vector consisting of multiple feature values ​​and a prediction model that outputs predictions for the multiple input feature values, generating a correction vector that represents a correction to the feature values ​​of the input example as a vector, and calculating a prediction by the prediction model for a virtual case that represents a state consisting of the feature values ​​on the correction vector linked to the input example, between the input example and a corrected example obtained by correcting the input example with the correction vector.

[0047] REFERENCE SIGNS LIST 10 Information processing device 11 Input unit 12 Generation unit 13 Calculation unit 14 Output unit 16 Model storage unit 17 Case storage unit 100 Information processing device 101 CPU 102 ROM 103 RAM 104 Program group 105 Storage device 106 Drive device 107 Communication interface 108 Input / output interface 109 Bus 110 Storage medium 111 Communication network 121 Input unit 122 Generation unit 123 Calculation unit

Claims

1. An input unit that receives an input case represented by a vector composed of values of a plurality of features and a prediction model that outputs a prediction for the input values of the plurality of features; a generation unit that generates a correction vector representing, in vector form, a correction to the values of the features of the input case; and a calculation unit that calculates a prediction by the prediction model for a virtual case representing a state composed of the values of the features on the correction vector connected to the input case, between the input case and a corrected case obtained by correcting the input case with the correction vector. An information processing apparatus comprising the above components.

2. The information processing apparatus according to claim 1, wherein the calculation unit calculates a change in the prediction by the prediction model with respect to a change in the state of the virtual case on the correction vector connected to the input case. An information processing apparatus.

3. The information processing apparatus according to claim 2, further comprising an output unit that outputs to display a graph representing a change in the prediction by the prediction model with respect to a change in the state of the virtual case on the correction vector connected to the input case, in a coordinate space defined by a first axis corresponding to the state of the virtual case and a second axis corresponding to the value of the prediction by the prediction model. An information processing apparatus.

4. The information processing apparatus according to claim 3, wherein the output unit sets, for each of the features, a coordinate space defined by a third axis corresponding to the state of the virtual case and a fourth axis corresponding to the value of the feature, and outputs to display a graph representing a change in the value of the feature accompanying a change in the state of the virtual case on the correction vector connected to the input case, in the coordinate space corresponding to each of the features. An information processing apparatus.

5. The information processing apparatus according to claim 4, wherein the output unit outputs to display the coordinate spaces corresponding to each of the features arranged in parallel along the third axis. An information processing apparatus.

6. The information processing apparatus according to claim 5, wherein the output unit arranges the coordinate spaces corresponding to each of the features such that the first axis and the third axis are parallel, and outputs to display the coordinate spaces arranged with the positions of the values on the first axis and the values on the third axis associated with each other. An information processing apparatus.

7. The information processing apparatus according to claim 1, wherein the generation unit generates a group of modification vectors composed of a plurality of successive modification vectors that sequentially modify the input case, and the calculation unit calculates a prediction by the prediction model for a virtual case composed of values of the features on the group of modification vectors sequentially concatenated to the input case, between the input case and the modified case obtained by modifying the input case with the group of modification vectors.

8. The information processing apparatus according to claim 7, wherein the generation unit generates the group of modification vectors based on the norms of the modification vectors included in the group of modification vectors.

9. The information processing apparatus according to claim 8, wherein the generation unit generates the group of modification vectors such that the sum of the norms of the modification vectors included in the group of modification vectors is small.

10. The information processing apparatus according to claim 7, wherein the generation unit generates the group of modification vectors based on the number of the modification vectors included in the group of modification vectors.

11. The information processing apparatus according to claim 10, wherein the generation unit generates the group of modification vectors such that the number of the modification vectors included in the group of modification vectors is small.

12. The information processing apparatus according to claim 7, wherein the generation unit generates the group of modification vectors such that each of the modification vectors included in the group of modification vectors modifies a number of features smaller than the total number of the features included in the input case.

13. The information processing apparatus according to claim 7, wherein the generation unit generates the group of modification vectors such that each of the modification vectors included in the group of modification vectors modifies different features among the features included in the input case.

14. The information processing apparatus according to claim 1, wherein the calculation unit calculates a prediction by the prediction model for a virtual case in which the values of the features on the modification vector are changed using a continuously changing intermediate variable.

15. An information processing method that receives an input case represented by a vector composed of values of a plurality of features and a prediction model that outputs a prediction for the input values of the plurality of features, generates a correction vector representing, in vector form, a correction to the values of the features of the input case, and calculates a prediction by the prediction model for a virtual case representing a state composed of the values of the features on the correction vector concatenated to the input case, between the input case and a corrected case obtained by correcting the input case with the correction vector.

16. The information processing method according to claim 15, wherein the method calculates a change in the prediction by the prediction model with respect to a change in the state of the virtual case on the correction vector concatenated to the input case.

17. The information processing method according to claim 16, wherein the method outputs, for display, a graph representing a change in the prediction by the prediction model with respect to a change in the state of the virtual case on the correction vector concatenated to the input case, in a coordinate space defined by a first axis corresponding to the state of the virtual case and a second axis corresponding to the value of the prediction by the prediction model.

18. A computer-readable storage medium storing a program that causes a computer to execute a process of receiving an input case represented by a vector composed of values of a plurality of features and a prediction model that outputs a prediction for the input values of the plurality of features, generating a correction vector representing, in vector form, a correction to the values of the features of the input case, and calculating a prediction by the prediction model for a virtual case representing a state composed of the values of the features on the correction vector concatenated to the input case, between the input case and a corrected case obtained by correcting the input case with the correction vector.

Images