JP5785631B2 - Information processing apparatus, control method therefor, and computer program - Google Patents

Description

  The present invention relates to an information processing apparatus, a control method therefor, and a computer program, and more particularly to an inference technique for learning data with known attributes and inferring attributes of data with unknown attributes.

  As one of data processing techniques using a computer, an inference technique for inferring an unknown event based on knowledge extracted from a known event is known. Many inference devices that infer unknown events obtain knowledge used for inference by supervised learning. Supervised learning refers to data between attributes and characteristic values using data given the attributes of the data you want to find along with characteristic values that represent the characteristics of the data (that is, data with known attributes (known data)). This is a method for learning correspondence (knowledge). The characteristic value may be referred to as “observation value”, and the data attribute may be referred to as “class” or “label”. Using the knowledge obtained by supervised learning, the inference apparatus infers an attribute from the characteristic value of the unknown data for the data whose attribute is unknown (unknown data). Therefore, the quality of knowledge obtained by supervised learning greatly affects the inference accuracy of the inference apparatus using the knowledge.

  Conventional reasoning apparatuses perform supervised learning on the assumption that the distribution of known data is equal to the distribution of unknown data. Therefore, it has been considered that if a sufficiently large number of known data can be used to accurately learn the distribution of the known data, the attributes of the unknown data can be accurately inferred.

  Patent Document 1 describes that a reasoning result that is more appropriate than the case of using a single reasoning device is obtained by weighting and merging the reasoning results of a plurality of reasoning devices having different inference methods. ing.

  On the other hand, in recent years, there have been attempts to support diagnosis using an inference device in the medical field. For example, a technique for inferring an attribute (such as a diagnosis name) using a characteristic value of a lesioned part as an input has been studied.

Japanese Patent Laid-Open No. 07-281898

  However, a sufficient number of known data may not be obtained when learning an inference device for inferring the attribute of a lesion. Furthermore, the method of acquiring the characteristic value of the lesion may change due to improvement of the medical device, and the characteristic and occurrence probability of the lesion itself may change with changes in the times and the environment. For these reasons, the learning result of the inference device is not always sufficient, and there is no guarantee that the original inference accuracy will be maintained at a certain level. Therefore, the user (doctor) does not know how much the inference result of the inference device can be trusted, and some doctors think that the inference result of the inference device cannot be trusted, and there is a problem that the inference device cannot be used effectively.

  The problem described above cannot be solved simply by increasing the number of learning data or using a plurality of inference devices to improve the inference accuracy. In order to solve the above-described problem, the inference apparatus needs to present information that allows the user to determine the reliability of the inference result.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a technique for enabling effective use of an inference device.

In order to achieve the above object, an information processing apparatus according to the present invention comprises the following arrangement. That is,
An information processing apparatus that displays on a display means a result of inferring an attribute to which input data belongs under a plurality of inference conditions,
Display control means for displaying, on each of the inference conditions, an inference result obtained by inferring the probability that the input data belongs to the attribute for each of a plurality of predetermined attributes under the inference conditions;
With
The display control means highlights the inference result for each of the plurality of inference conditions according to the value of a parameter that specifies the inference condition.

  ADVANTAGE OF THE INVENTION According to this invention, the technique for enabling effective use of an inference apparatus can be provided.

It is a block diagram which shows the function structure of an inference apparatus. It is a flowchart which shows the control procedure of an inference apparatus. It is a figure which shows the 1st GUI display example of an inference apparatus. It is a figure which shows the 2nd GUI display example of an inference apparatus. It is a figure which shows the example of GUI display of an inference apparatus. It is a functional block diagram which shows the structure of an inference apparatus. It is a flowchart which shows the control procedure of an inference apparatus. It is a block diagram which shows the hardware constitutions of an inference apparatus.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[First Embodiment]
(Functional structure of inference device)
FIG. 1 is a functional block diagram illustrating a functional configuration of the inference apparatus according to the first embodiment.

  The inference apparatus 1 can be realized by executing a computer program for executing control described later on a general computer (information processing apparatus). Further, the inference apparatus 1 can be realized by mounting the below-described control in hardware. Furthermore, the inference apparatus 1 can also be realized by constructing a hybrid system including hardware and a computer program by mounting a part of the below-described control by hardware and mounting the other part as a computer program.

  The inference device 1 includes a control unit 10, an input unit 20, an inference unit 30, a storage unit 40, and a display unit 50. The control unit 10 controls the entire inference apparatus 1 according to a control procedure described later. Although not shown, the control unit 10 is connected to all processing units and processing elements. The control unit 10 can be realized by a CPU 990 described later, for example.

  The input unit 20 sends information (control parameters) such as a characteristic value of inference target data (unknown data) and an inference result display rule related to the control of the inference device 1 from the outside to the inference device 1 according to the control of the control unit 10. Enter.

  The inference unit 30 has a plurality of inference elements that have been learned in advance using different learning data or learning methods. In the example of FIG. 1, the inference unit 30 has inference elements 31 to 34. Each of the inference elements 31 to 34 infers the attribute of the inference object data based on the characteristic value of the inference object data input from the input unit 20 (and the control parameter of the inference apparatus 1) according to the control of the control unit 10, and the inference result Is output to the display unit 50. At this time, since each of the inference elements 31 to 34 is learning different from each other, even if the same characteristic value is input, different attributes (inference results) may be output. That is, each of the inference elements 31 to 34 has a different inference condition, and thus the inference accuracy is also different from each other. Note that the inference elements 31 to 34 do not have to be physically different. For example, the inference elements 31 to 34 may be realized by using a single computer or hardware in a time-sharing manner to perform different inference results. Good.

  The storage unit 40 includes a plurality of storage elements corresponding to the plurality of inference elements. In the example of FIG. 1, the storage unit 40 includes storage elements 41 to 44 corresponding to the inference elements 31 to 34, respectively. . The storage elements 41 to 44 store learning data or information related to the learning technique used when learning each of the inference elements 31 to 34. In addition, the storage unit 40 includes a storage element 45 that stores a control parameter for specifying an inference condition of the inference device 1. The storage unit 40 can be realized by a data storage device such as an external storage device 995 or a RAM 992 described later. The storage elements 41 to 45 do not have to be physically different, and may be different storage areas secured on one memory.

  The display unit 50 reads the information stored in each of the storage elements 41 to 44 based on the display rule of the inference result that is the control parameter of the inference device 1 according to the control of the control unit 10, and based on this, the inference element 31 is read out. The display method of the inference result input from each of .about.34 is determined. And based on the determined display method, display control which displays each inference result on a display apparatus is performed.

For example, it is assumed that each of the inference elements 31 to 34 learns in advance using the following learning data.
・ Inference element 31: Learning data obtained from cases accumulated in the past 2 years at hospital A ・ Inference element 32: Learning data obtained from cases accumulated in the past 10 years at hospital A ・ Inference element 33: In hospital B Learning data / inference element 34 obtained from cases accumulated in the past two years: Learning data obtained from cases accumulated in the past 10 years at hospital B In this case, each of the storage elements 41 to 44 includes, for example, Remember the information. :
Memory element 41: Number of learning data N ₃₁ obtained from cases accumulated in A hospital in the past two years, and reasoning accuracy A ₃₁ of reasoning element ₃₁
Memory element 42: Number of learning data N ₃₂ obtained from cases accumulated in the past 10 years at hospital A, and reasoning accuracy A ₃₂ of reasoning element ₃₂
Memory element 43: Number of learning data N ₃₃ obtained from cases accumulated in hospital B in the past two years, and reasoning accuracy A ₃₃ of reasoning element ₃₃
Memory element 44: Number of learning data N ₃₄ obtained from cases accumulated in the past 10 years at hospital B, and reasoning accuracy A ₃₄ of reasoning element ₃₄
Here, the inference accuracy A _{31 to} A ₃₄ is calculated in advance for each of the inference elements 31 to 34 using a known evaluation method such as a cross-validation method or a data division method.

(Hardware configuration of inference device)
FIG. 8 is a block diagram schematically illustrating a hardware configuration example of the inference apparatus 1 according to the present embodiment. The inference apparatus 1 according to the present embodiment is realized by, for example, a personal computer (PC), a workstation (WS), a mobile terminal, a smartphone, or the like.

  In FIG. 8, a CPU 990 is a central processing unit, and controls the overall operation of the inference apparatus 1 in cooperation with other components based on an operating system (OS), application programs, and the like. The ROM 991 is a read-only memory, and stores programs such as basic I / O programs, data used for basic processing, and the like. A RAM 992 is a writable memory and functions as a work area for the CPU 990.

  The external storage drive 993 can access a recording medium, and can load a program or the like stored in the medium (recording medium) 994 into this system. The media 994 includes, for example, a flexible disk (FD), a CD-ROM, a DVD, a USB memory, a flash memory, and the like. The external storage device 995 is a device that functions as a large-capacity memory. In this embodiment, a hard disk device (hereinafter referred to as HD) is used. The HD 995 stores an OS, application programs, and the like.

  The instruction input device 996 is a device that receives an instruction or command input from a user, and corresponds to a keyboard, a pointing device, a touch panel, and the like. The display 997 is a display device that displays a command input from the instruction input device 996, a response output of the inference device 1 to the command, and the like. An interface (I / F) 998 is a device that relays data exchange with an external device. A system bus 999 is a data bus that manages the flow of data in the inference apparatus 1.

  In addition, it can also be comprised as an alternative of a hardware apparatus with the software which implement | achieves a function equivalent to the above each apparatus.

(Control of inference device)
Next, the control method of the inference apparatus 1 will be described using the flowchart of FIG. 2 and GUI (Graphical User Interface) display examples of FIGS. For convenience of explanation, the explanation will be given first from FIG. 3 and FIG.

  3 and 4 are diagrams illustrating first and second GUI display examples of the inference apparatus according to the first embodiment. The GUI is a user interface that combines the functions of the input unit 20 and the display unit 50. In the following, the terms used in “Windows (registered trademark)”, which is widely used as an OS (operating system) of a computer, will be used when describing each component of the GUI.

  The inference apparatus 1 performs a reception process of accepting selection of any one of a plurality of parameters for specifying an inference condition from the user using the GUI. Specifically, the user can use the combo box 510 to input an inference result display rule that is one of the control parameters that specify the inference condition of the inference device 1. In the example of FIG. 3 and FIG. 4, as a criterion for determining the radius of the circle when displaying the inference results 531 to 534 by the inference elements 31 to 34 as a pie chart, “the number of learning data” or “inference accuracy” It is a display rule to use any of “height”. Instead of a combo box, a list box, an edit box, or a button control such as a check box or radio button can be used.

  In addition, the user can input weights for each of the inference results 531 to 534 that are one of the control parameters of the inference apparatus 1 using the sliders 521 to 524. Based on the weight input by the user, the display unit 50 displays a combined display 535 of the inference result. Instead of the slider, the weight can be input as a numerical value by using an edit box, a spin button or the like.

  3 and 4 are examples in which the inference result of the attribute (diagnosis name) of the lesioned part in the image diagnosis support is displayed. The respective probabilities (probability of each diagnosis name) of the diagnosis names 1 to 3 are represented by the central angle of each sector in the pie chart.

(Control procedure of inference device)
FIG. 2 is a flowchart showing a control procedure of the inference apparatus according to the first embodiment. All processes shown in FIG. 2 are controlled by the control unit 10.

  In step S201, the control unit 10 sets initial values of control parameters of the inference apparatus 1. The control parameters include initial values of inference result display rules and initial values of weights for the inference results 531 to 534. Some of the set control parameters are displayed on the display unit 50. That is, the initial value of the inference result display rule is displayed in the combo box 510 described above. In addition, initial values of weights for the inference results 531 to 534 are displayed on the sliders 521 to 524 described above.

  In step S202, the input unit 20 acquires the characteristic value of the inference target data and passes it to each of the inference elements 31 to 34. When inferring an attribute (diagnosis name) of a lesion in image diagnosis support, a characteristic value is an image feature amount or an image finding of a medical image that is inference target data. This image feature amount can be obtained by calculating an existing geometric feature amount, density feature amount, and the like for an abnormal region extracted using an existing lesion region extraction method. The image findings are information describing, using medical terms, the image features of the abnormal area discovered by the doctor during interpretation, and are manually input by the doctor.

In step S <b> 203, each of the inference elements 31 to 34 executes an inference process, obtains inference results 531 to 534, and transmits them to the display unit 50. In this embodiment, it is _assumed that the probability P _3nk for the diagnosis name k (k = 1 to 3) is obtained as the inference result 53n (n = 1 to 4). That is, P _3nk is the _probability that the inference element 3n infers that the diagnosis name is k, and P _3n1 + P _3n2 + P _3n3 = 1. As described above, since each inference element is learned using different learning data or learning methods, the inference results are different from each other.

  In step S <b> 204, the input unit 20 acquires a display rule specified by the user using, for example, the combo box 510 described above. Note that the process of step S204 may be executed at any time in response to an input from the user. After step S204 is executed at an arbitrary timing, the processing after step S205 is executed. Alternatively, step S204 may be omitted if there is no input from the user.

In step S <b> 205, the display unit 50 reads information that matches the display rule from each of the storage elements 41 to 44. In the example of FIGS. 3 and 4, either the learning data number or the inference accuracy of each inference element is read according to the designated display rule. That is, when “large number of learning data” is designated as the display rule as shown in FIG. 3, the learning data numbers N _{31 to} N ₃₄ are read. Further, when “high inference accuracy” is designated as the display rule as shown in FIG. 4, the inference accuracy A _{31 to} A ₃₄ is read.

In step S206, the display unit 50 displays the inference results 531 to 534 of the inference elements 31 to 34 based on the information read in step S205. As described above, the radius of each circle of the inference results 531 to 534 is determined based on the number of learning data N _{31 to} N ₃₄ (example in FIG. 3) or inference accuracy A _{31 to} A ₃₄ (example in FIG. 4). The For example, when “high inference accuracy” is specified as the display rule, the inference accuracy A _{31 to} A ₃₄ (value of 0 to 1) is added to a predetermined reference value (maximum value) r _max of the radius. Multiply each to determine the radius. Similarly, if the "abundance of learning data" is designated as the display rule determines radius by multiplying the respective number of learning data normalized to r _max. Here, the value after normalization of the number of learning data may be calculated as N _3n / N _max using the maximum value N _max in N _3n , for example. Thereby, each display of the inference results 531 to 534 is regularly changed based on the display rule.

  As described above, in the present embodiment, the inference result is highlighted in accordance with the value of the control parameter (for example, the number of learning data and the inference accuracy) selected from the user for a plurality of inference conditions. For this reason, the user can easily grasp the relationship between the selected control parameter and the inference result. In this embodiment, the inference result is displayed in a graph for each inference condition, and for each inference condition, the overall size of the graph corresponding to the inference condition is set according to the value of the control parameter selected by the user. By setting, the inference result is highlighted. For this reason, the user can understand at a glance the relationship between the selected control parameter and the inference result. In this embodiment, for each inference condition, information indicating the inference condition (for example, a hospital name and a period during which learning data is acquired) is displayed in the vicinity of the display of the inference result. For this reason, the user can easily know under what conditions each inference result is inferred.

In step S207, the input unit 20 acquires weights (weights 31 to 34) for the inference results 531 to 534 specified by the user, for example, using the sliders 521 to 524 described above. Then, each weight is normalized so that the sum of the weights becomes 1. In the following description, the normalized value of the weight 3n of each inference result 53n (n = 1 to 4) obtained in this step is expressed as W _3n . Note that the process of step S207 may be executed at any time in response to an input from the user. After step S207 is executed at an arbitrary timing, step S208 is executed. Alternatively, if there is no input from the user, the input of the weight in step S207 may be omitted.

In step S208, the display unit 50 combines the inference results 531 to 534 by the following calculation method and displays the combined result 535. First, the display unit 50 calculates the probability P _k for the diagnosis name k (k = 1 to 3) of the synthesis result 535 using [Equation 1].
[Equation 1]
P _k = W ₃₁ * P _31k + W ₃₂ * P _32k + W ₃₃ * P _33k + W ₃₄ * P _34k
The resulting P _k value is P ₁ + P ₂ + P ₃ = 1. Based on the P _k value thus obtained, the display unit 50 determines the center angle of each sector in the pie chart of the synthesis result 535. The radius of the pie chart of the synthesis result 535 may be a predetermined size (fixed value).

  As described above, in this embodiment, the user selects a weight to be assigned to each inference condition, weights the inference result of each inference condition with this weight, and synthesizes and displays the inference results in a plurality of inference conditions. To do. For this reason, the user can obtain a highly accurate inference result by appropriately selecting the weight of each inference result based on knowledge, experience, know-how, and the like.

  In step S209, the control unit 10 determines whether or not to end the processing of the inference apparatus 1 according to the instruction from the user acquired by the input unit 20. And when it determines with not complete | finishing, the process after step S204 is performed again, and when it determines with complete | finishing, the process of the inference apparatus 1 is complete | finished.

  The process of the inference device 1 is executed by the above procedure.

  As described above, the difference between the inference results obtained from a plurality of inference elements and the reason why the difference has occurred can be displayed in an easy-to-understand manner for the user, so that the user can determine the reliability of the inference results of each inference element. Furthermore, since the user can freely weight and synthesize inference results obtained from a plurality of inference elements, a highly reliable synthesis result for the user can be obtained. Therefore, the user can effectively use the inference apparatus according to the present embodiment.

  Furthermore, the cause of the difference in each inference result (difference in learning data or learning method) may be previously notified to the user by a screen display at the time of unillustrated. Alternatively, the user may be able to confirm the cause of the difference between a plurality of inference results at any time by displaying the screen when not shown. Accordingly, the user can deeply consider the cause of the difference between the inference results, and thus can determine the importance of the plurality of inference results in more detail. In the above example, for example, it is assumed that a certain user works at hospital A and places more importance on case data of hospital A than case data of hospital B. Furthermore, it is assumed that the user has determined that the case data of the past two years can be more reliable than the case data of the past ten years because the imaging device was replaced three years ago in Hospital A. In such a case, not only a large weight is given to the inference result that best meets the criteria selected by the display rule, but also a relatively large weight is given to the inference result learned from the case data of the past two years of hospital A. Therefore, deep consideration of the user can be assisted.

[Second Embodiment]
In the first embodiment, information related to the reliability of each inference element (radius of each pie chart in FIGS. 3 and 4) is displayed without being distinguished by the data attribute (diagnostic name). The inference apparatus according to the second embodiment displays information related to the reliability of each inference element for each attribute of learning data of each inference element.

  The configuration of the inference apparatus according to this embodiment is the same as that of the first embodiment. However, the content of the data held in the storage unit 40 and the display process by the display unit 50 are different. Hereinafter, the inference apparatus according to this embodiment will be described only with respect to differences from the first embodiment.

  FIG. 5 is a diagram illustrating a GUI display example of the inference apparatus according to the present embodiment.

  The GUI screen configuration is the same as in FIGS. 3 and 4 is a method of displaying graphs of the inference results 531 to 534 and the synthesis result 535.

In FIG. 5, the fan-shaped radius is changed for each diagnosis name. In order to realize this display method, each of the storage elements 41 to 44 is divided into diagnosis names as learning data or information related to the learning method used when learning each of the inference elements 31 to 34 as follows. Information is stored in advance.
○ Memory element 41: As information obtained from cases accumulated in the past two years at Hospital A,
・ Number of learning data with attribute of diagnosis name 1 N ₃₁₁
・ Number of learning data with attribute of diagnosis name 2 N ₃₁₂
・ Number of learning data with attribute of diagnosis name 3 N ₃₁₃
Inference accuracy A ₃₁₁ of the inference element 31 for test data having the attribute of diagnosis name 1
・ Inference accuracy A ₃₁₂ of the inference element 31 for test data having the attribute of diagnosis name 2
- inference accuracy A ₃₁₃ of the inference elements 31 for the test data with the attributes of diagnosis 3
○ Memory element 42: As information obtained from cases accumulated in the past 10 years at Hospital A,
・ Number of learning data N ₃₂₁ with attribute of diagnosis name 1
・ Number of learning data with attribute of diagnosis name 2 N ₃₂₂
・ Number of learning data with attribute of diagnosis name 3 ₃₂₃
Inference accuracy A ₃₂₁ of the inference element 31 for test data having the attribute of diagnosis name 1
- inference accuracy A ₃₂₂ of the inference elements 31 for the test data with the attribute of the diagnosis 2
The inference accuracy A ₃₂₃ of the inference element 31 for the test data having the attribute of the diagnosis name 3
○ Memory element 43: As information obtained from cases accumulated in the past two years at hospital B,
・ Number of learning data with attribute of diagnosis name 1 N ₃₃₁
・ Number of learning data with attribute of diagnosis name 2 N ₃₃₂
・ Number of learning data with attribute of diagnosis name 3 N ₃₃₃
The inference accuracy A ₃₃₁ of the inference element 31 for the test data having the attribute of diagnosis name 1
Inference accuracy A ₃₃₂ of the inference element 31 for test data having the attribute of diagnosis name 2
- inference accuracy A ₃₃₃ of the inference elements 31 for the test data with the attributes of diagnosis 3
○ Memory element 44: As information obtained from cases accumulated in the past 10 years at hospital B,
_{-Number of} learning data N ₃₄₁ with attribute of diagnosis name 1
・ Number of learning data with attribute of diagnosis name 2 N ₃₄₂
・ Number of learning data with attribute of diagnosis name 3 N ₃₄₃
The inference accuracy A ₃₄₁ of the inference element 31 for the test data having the attribute of diagnosis name 1
The inference accuracy A ₃₄₂ of the inference element 31 for the test data having the attribute of the diagnosis name 2
The inference accuracy A ₃₄₃ of the inference element 31 for the test data having the attribute of the diagnosis name 3
The control procedure of the inference device 1 according to the present embodiment is also shown in the flowchart of FIG. 2 as in the first embodiment. However, steps S205, S206, and S208 are respectively corrected as follows.

In step S <b> 205, the display unit 50 reads information that matches the display rule from each of the storage elements 41 to 44. In the example of FIG. 5, since “high inference accuracy” is designated as a display rule, inference accuracy A _{311 to} A ₃₁₃ , A _{321 to} A ₃₂₃ , A _{331 to} A ₃₃₃ , and A _{341 to} A ₃₄₃ are read out. . On the other hand, when “abundance of learning data” is designated as the display rule, the number of learning data N _{311 to} N ₃₁₃ , N _{321 to} N ₃₂₃ , N _{331 to} N ₃₃₃ , and N _{341 to} N ₃₄₃ can be read out. That's fine.

In step S206, the display unit 50 displays the inference results 531 to 534 of the inference elements 31 to 34 based on the information read in step S205. The fan-shaped radius (referred to as r _3nk ) of the diagnosis name k (k = 1 to 3) of the inference result 53n (n = 1 to 4) is determined based on the inference accuracy A _3nk . Alternatively, the fan-shaped radius of the diagnosis name k may be determined based on the learning data number N _3nk depending on the display rule. The central angle of the diagnostic name k of the inference result 53n is determined based on the probability P _3nk of the diagnosis name k of the inference result 53n. Thereby, each display of the inference results 531 to 534 is regularly changed based on the display rule.

  In step S208, the display unit 50 combines the inference results 531 to 534 by the following calculation method and displays the combined result 535.

The probability P _k of the combined result 535 for the diagnosis name k is calculated using the above [Equation 1]. Based on the P _k value obtained in this way, the display unit 50 determines the center angle of each sector of the synthesis result 535. On the other hand, the fan-shaped radius (r _k ) for the diagnosis name k in the synthesis result 535 is calculated using [Equation 2].
[Equation 2]
r _k = W ₃₁ * r _31k + W ₃₂ * r _32k + W ₃₃ * r _33k + W ₃₄ * r _34k
As described above, in this embodiment, for each inference condition, the inference result is highlighted for each attribute according to the value of the control parameter given in advance corresponding to the attribute. For this reason, the difference between the inference results obtained from multiple inference elements and the reason for the difference can be displayed to the user in an easy-to-understand manner for each data attribute (diagnostic name). Can judge for themselves. Furthermore, since the user can freely weight and synthesize inference results obtained from a plurality of inference elements, a highly reliable synthesis result for the user can be obtained. Therefore, the user can effectively use the inference apparatus according to the present embodiment.

[Third Embodiment]
In the first and second embodiments, the information relating to the reliability of each inference element (the radius of each pie chart in FIGS. 3 and 4 or the radius of each sector in FIG. 5) is stored in the display rule and each storage element. It was decided based on information (number of learning data or inference accuracy). The information stored in each storage element is information obtained when learning each inference device. Therefore, if the display rule is not changed and the re-learning of each inference device is not performed, the information related to the reliability of the inference element is not changed regardless of the characteristic value of the inference target data. The inference apparatus according to the third embodiment is characterized in that information relating to the reliability of each inference element is dynamically obtained according to inference object data and displayed. Specifically, learning data similar to the input data is determined among the learning data used for inferring the attribute to which the input data belongs. Then, the number of the similar learning data or the inference accuracy of the inference using the learning data is used as a control parameter to highlight the inference result.

(Configuration of inference device)
FIG. 6 is a functional block diagram showing the configuration of the inference apparatus according to the third embodiment.

  The control unit 10, the input unit 20, the inference unit 30, and the display unit 50 have the same functions as those described with reference to FIG. Hereinafter, the inference apparatus according to this embodiment will be described only with respect to differences from the first embodiment.

  The storage unit 40 includes storage elements 41 to 44 corresponding to the inference elements 31 to 34, respectively. Each storage element stores information related to known data used when learning the corresponding inference element. The information relating to the known data includes the characteristic value and attribute of each known data, and the inference result of the known data by the corresponding inference element (information on whether or not the attribute of the known data has been correctly inferred). In addition, the storage unit 40 includes a storage element 45 that stores control parameters and the like of the inference apparatus 1. The storage elements 41 to 45 do not have to be physically different, and may be different storage areas secured on one memory. Alternatively, instead of the storage elements 41 to 45, a database or external storage device (not shown) connected to the inference device 1 via a communication line or network (not shown) may be used.

  The similar data search unit 60 includes similar data search elements 61 to 64 corresponding to the inference elements 31 to 34 and the storage elements 41 to 44, respectively. Each of the similar data search elements 61 to 64 searches for known data having a characteristic value similar to the characteristic value of the input inference target data.

  The calculation unit 70 includes calculation elements 71 to 74 corresponding to the similar data search elements 61 to 64, respectively. The calculation unit 70 calculates information based on the display rule (control parameter) from the retrieved known data.

(Control procedure of inference device)
Next, the control procedure of the inference apparatus 1 according to the present embodiment will be described using the flowchart of FIG. However, the processing (S201, S203, S204, and S206 to S209) in which the same step numbers as in FIG. 2 perform the same control as described with reference to FIG. Hereinafter, S701 to S703, which are processes different from those in FIG. 2, will be described.

  In step S701, the input unit 20 acquires the characteristic value of the inference target data and passes it to each of the inference elements 31 to 34 and each of similar data search elements 61 to 64 described later.

  In step S702, the similar data search elements 61 to 64 obtain the characteristic values of all the known data stored in the corresponding storage elements 41 to 44 and the characteristic values of the inference target data passed from the input unit 20, respectively. Compare. Then, a plurality of similar data is obtained by using a known similar data search technique. In the known similar data search technique, the distance between the characteristic value of the inference target data and the characteristic value of the known data to be compared (the distance between vectors when each characteristic value is regarded as a multidimensional vector) is Close data is set as similar data. For the method of selecting similar data, for example, the distance (similarity) between the inference target data and the method of selecting similar data until the predetermined number is reached in order of the distance (similarity) from the inference target data is predetermined. There is a method of selecting all similar data below a threshold. In the following description, the similar data search elements 61 to 64 are all described as having obtained a plurality of similar data using the latter method (a method of selecting all similar data at a distance equal to or less than a threshold).

  In step S703, the calculation elements 71 to 74 calculate information based on the display rule from the plurality of similar data obtained by the corresponding similar data search elements 61 to 64, respectively. For example, the first display rule is a large number of similar data, and the second display rule is a high inference accuracy for similar data. When the first display rule is selected, each of the calculation elements 71 to 74 counts the number of a plurality of similar data obtained as search results. Then, this value is normalized using a predetermined value.

  On the other hand, when the second display rule is selected, each of the calculation elements 71 to 74 first calculates the inference result (correct attribute of known data) for each of a plurality of similar data obtained as a search result. Examine information about whether or not inference was possible. Here, the inference result of the similar data is read from the storage elements 41 to 44 corresponding to the calculation elements 71 to 74, respectively. Next, each of the calculation elements 71 to 74 calculates the ratio of correct answers in the inference result of the plurality of read similar data, that is, the inference accuracy for the similar data.

  Thereafter, by executing the processing from step S206 onward, the display as illustrated in the GUI of FIGS. 3, 4 and 5 is performed. Note that the information related to the reliability of each inference element (the radius of each pie chart in FIGS. 3 and 4 or the radius of each sector in FIG. 5) is information derived from the display rules and inference target data (and known data) ( It is displayed as information that changes according to the number of similar data or the inference accuracy for similar data.

  As described above, the difference between the inference results obtained from multiple inference elements is displayed to the user in an easy-to-understand manner based on the number of similar data in the inference target data or the inference accuracy for the similar data. The user can judge the reliability of the user himself. Furthermore, since the user can freely weight and synthesize inference results obtained from a plurality of inference elements, a highly reliable synthesis result for the user can be obtained. Therefore, the user can effectively use the inference apparatus according to the present embodiment.

(Modification 1)
In FIGS. 3, 4, and 5, the difference between a plurality of inference results and the cause thereof are displayed in an easy-to-understand manner for the user by changing the radius of the pie chart or the sector graphic based on the display rule, but the present invention is not limited thereto. That is, even if other graph display or character display is used, the display method based on the display rule can be changed. For example, when a bar graph is used, the bar attributes (maximum values of width and height, etc.) of the bar graph may be regularly changed based on the display rules. Further, for example, when character display is used, character attributes (font size, character color, etc.) may be regularly changed based on display rules. Furthermore, in an arbitrary graph display or character display, the inference results may be arranged in the order matching the display rules.

  As described above, according to the inference device according to the present invention, the difference between the inference results obtained from a plurality of inference elements and the reason why the difference has occurred can be displayed in an easy-to-understand manner for the user. There is an effect that the user can judge the reliability himself. Therefore, there is an effect that the user can effectively use the inference device.

<< Other Embodiments >>
The present invention can also be realized by executing the following processing. That is, software (program) that realizes the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and a computer (or CPU, MPU, or the like) of the system or apparatus reads the program. It is a process to be executed.

Claims (11)

An information processing apparatus that displays on a display means a result of inferring an attribute to which input data belongs under a plurality of inference conditions,
Display control means for displaying, on each of the inference conditions, an inference result obtained by inferring the probability that the input data belongs to the attribute for each of a plurality of predetermined attributes under the inference conditions;
With
The information processing apparatus, wherein the display control unit highlights the inference result for each of the plurality of inference conditions according to a value of a parameter that specifies the inference condition. The display control means includes
Displaying the inference result in a graph for each of the plurality of inference conditions;
For each of the plurality of inference conditions, the inference result is highlighted by setting the overall size of the graph corresponding to the inference condition according to the value of the parameter specifying the inference condition. The information processing apparatus according to claim 1. Selection means for allowing a user to select a weight to be assigned to each of the plurality of inference conditions;
Further comprising: a combining unit that weights each of the inference results inferred for each of the plurality of inference conditions with the selected weight, and synthesizes the inference results in the plurality of inference conditions;
The information processing apparatus according to claim 1, wherein the display control unit further displays the synthesized inference result on the display unit.   4. The display control unit according to claim 1, wherein the display control unit displays, for each of the plurality of inference conditions, information for specifying the inference condition in the vicinity of the display of the inference result. 5. Information processing device.   The display control means highlights the inference result for each of the plurality of inference conditions for each of the plurality of attributes according to a value of a parameter given in advance corresponding to the attribute. The information processing apparatus according to any one of claims 1 to 4. Wherein the inference condition to identify the Rupa parameters, according to any one of claims 1-5, wherein the input data includes the number of training data used to infer attributes belonging Information processing device. Wherein the inference condition to identify the Rupa parameter, the information processing apparatus according to any one of 6 claim 1, characterized in that includes the inference accuracy of the inference by the inference condition. The display control means includes
Either the number of learning data similar to the input data among the learning data used to infer the attribute to which the input data belongs in the inference condition, or the inference accuracy of the inference using the learning data The information processing apparatus according to claim 1, wherein the inference result is highlighted with the parameter as the parameter. A plurality of inference means for inferring attributes to which the input data belongs under different inference conditions,
9. The information processing apparatus according to claim 1, wherein the display control unit displays an inference result inferred by each of the plurality of inference units on the display unit. 10. A control method for an information processing apparatus that displays on a display means a result of inferring attributes to which input data belongs under a plurality of inference conditions ,
Table示制control means, for each of a plurality of attributes previously determined, the input data is displayed an inference result of the inference, respectively with probability the plurality of inference condition belonging to the attributes, to the display means for each inference condition A display control process,
In the display control step, for each of the plurality of inference conditions, the inference result is highlighted in accordance with a parameter value specifying the inference condition.   The computer program for functioning a computer as each means with which the information processing apparatus of any one of Claim 1 to 9 is provided.

Images