Learning support device, learning support method, and learning support program

Abstract

This invention provides a learning support device that can obtain rapid and valid prediction results using operational data, even when data drift occurs in the operational data. [Solution] A learning support device comprising: an acquisition unit that acquires learning data and operational data; a feature extraction unit that extracts features from the learning data and features from the operational data; a prediction result calculation unit that calculates a prediction result from the operational data based on the features from the operational data; a correction unit that corrects the prediction result based on the features from the learning data and features from the operational data; and an output unit that outputs the corrected prediction result.

Description

Technical Field

[0001] It relates to a learning support device, a learning support method, and a learning support program.

Background Art

[0002] In recent years, the development of systems that use machine learning models to classify data, detect product abnormalities, etc. has advanced. Examples of such systems include those that classify feedback reports on products from users into predetermined classes, and those that detect product abnormalities from sensor data such as image data at the production site.

[0003] However, the learning data used for learning the machine learning model and the operation data input to the machine learning model at the actual operation site are different data. Also, it is known that as time passes, the characteristics of the operation data change, resulting in so-called data drift where the characteristics of the learning data during learning are deviated from, and the prediction accuracy of the machine learning model decreases. Specific causes of data drift include, for example, changes in the background in images, changes in the materials used for products, changes in the appearance specifications of products, aging deterioration of the imaging camera, changes in the writing style (e.g., terms used) of articles due to changes in the report authors, changes in the data format, etc.

[0004] Related to a technique for preventing a decrease in the prediction accuracy of a machine learning model due to data drift, Patent Document 1 below discloses the following prior art. Based on the data values for each data item of the learning data, range information indicating the values that the data values for each data item can take is calculated, and among the operation data to be subjected to anomaly detection, the operation data in which the data value for each data item is included in the range information is stored as in-range data, the operation data not included in the range information is stored as out-of-range data, and it is determined whether retraining of the machine learning model is necessary based on the comparison result between the tendency of the in-range data and the tendency of the out-of-range data.

Prior Art Documents

Patent Documents

[0005] [Patent Document 1] Japanese Patent Publication No. 2021-193497 [Overview of the project] [Problems that the invention aims to solve]

[0006] However, the aforementioned prior art can determine whether data drift has occurred and whether the machine learning model needs to be retrained based on operational data, but it has the problem that it cannot provide reasonable prediction results from the machine learning model until retraining is performed.

[0007] This invention was made to solve these problems. Specifically, it aims to provide a learning support device, a learning support method, and a learning support program that can obtain rapid and valid prediction results using operational data, even when data drift occurs in the operational data. [Means for solving the problem]

[0008] The above-mentioned problems of the present invention are solved by the following means.

[0009] (1) A learning support device comprising: an acquisition unit that acquires learning data and operational data; a feature extraction unit that extracts feature quantities from the learning data and the operational data, respectively; a prediction result calculation unit that calculates a prediction result from the operational data based on the feature quantities of the operational data; a correction unit that corrects the prediction result based on the feature quantities of the learning data and the feature quantities of the operational data; and an output unit that outputs the corrected prediction result.

[0010] (2) The learning support device described in (1) above, wherein the correction unit performs a correction that degrades the prediction result.

[0011] (3) The learning support device according to (2) above, further comprising a reliability calculation unit that calculates the reliability of the prediction result based on the features of the learning data and the features of the operational data, wherein the correction unit performs a correction that degrades the prediction result based on the reliability.

[0012] (4) The learning support device according to (3) above, wherein the confidence calculation unit calculates the confidence level based on the distance between the feature quantities of the learning data and the feature quantities of the operational data.

[0013] (5) The learning support device according to (3) above, wherein the prediction result of the operational data is the likelihood of the class with the highest likelihood among predetermined classes, and the correction unit degrades the prediction result from the operational data by reducing the likelihood, which is the prediction result from the operational data, according to the confidence level.

[0014] (6) The learning support device according to (3) above, wherein the acquisition unit acquires a plurality of the learning data as a learning data group, and the confidence calculation unit calculates the confidence based on the distance between the distribution of a plurality of features extracted from each learning data in the learning data group and the features of the operational data.

[0015] (7) The learning support device according to (1) above, wherein the output unit further outputs the prediction result before correction.

[0016] (8) The learning support device according to (1) above, wherein the prediction result is the prediction result predicted from the operational data by a neural network trained on the learning data.

[0017] (9) The training data and the operation data are image data or text data, The learning support device according to (1) above, wherein the prediction result calculation unit calculates the prediction result using a supervised machine learning model that has been trained with the combination of the learning data and the correct label corresponding to the learning data as training data.

[0018] (10) A learning support method comprising the steps of: (a) acquiring training data and operational data; (b) extracting features of the training data and features of the operational data from the training data and the operational data, respectively; (c) calculating a prediction result from the operational data based on the features of the operational data; (d) correcting the prediction result based on the features of the training data and the features of the operational data; and (e) outputting the corrected prediction result.

[0019] (11) The learning support method described in (10) above, wherein in step (d), a correction is made to degrade the prediction result.

[0020] (12) The learning support method according to (11) above, further comprising step (f) of calculating the reliability of the prediction result based on the features of the learning data and the features of the operational data, wherein in step (d), a correction is made to degrade the prediction result based on the reliability.

[0021] (13) The learning support method according to (12) above, wherein in step (f), the confidence level is calculated based on the distance between the features of the learning data and the features of the operational data.

[0022] (14) The learning support method according to (12) above, wherein the prediction result of the operational data is the likelihood of the class with the highest likelihood among the predetermined classes, and in step (d), the prediction result of the operational data is degraded by reducing the likelihood, which is the prediction result from the operational data, in accordance with the confidence level.

[0023] (15) The learning support method according to (12) above, wherein in step (a), a plurality of the learning data is acquired as a group of learning data, and in step (f), the confidence level is calculated based on the distance between the distribution of a plurality of features extracted from each of the learning data in the group of learning data and the features of the operational data.

[0024] (16) In the above step (e), the learning support method according to (10) further outputs the prediction result before correction.

[0025] (17) The prediction result is the prediction result predicted from the operation data by a neural network learned by the learning data, and the learning support method according to (10) above.

[0026] (18) The learning data and the operation data are image data or text data. In step (c), the prediction result is calculated using a machine learning model that has been supervised-learned with a combination of the learning data and the correct label corresponding to the learning data as teacher data, and the learning support method according to (10) above.

[0027] (19) A learning support program for causing a computer to execute the learning support method according to any one of claims 10 to 18.

Advantages of the Invention

[0028] Extract the feature amounts of the learning data and the operation data, calculate the prediction result predicted from the operation data based on the feature amount of the operation data, and correct and output the prediction result based on the feature amount of the learning data and the feature amount of the operation data. Thereby, even when data drift occurs in the operation data, a quick and appropriate prediction result that utilizes the operation data can be obtained.

Brief Description of the Drawings

[0029] [Figure 1] It is a diagram showing a schematic configuration of an information processing system. [Figure 2] It is a block diagram showing the hardware configuration of an information processing apparatus. [Figure 3] It is an explanatory diagram for explaining the function of the control unit during learning using learning data. [Figure 4] It is an explanatory diagram for explaining the function of the control unit that extracts feature amounts from learning data. [Figure 5] This is an explanatory diagram illustrating the function of the control unit, which extracts features from operational data, calculates prediction results, and corrects those prediction results. [Figure 6] This is an explanatory diagram illustrating how to calculate the reliability of prediction results based on operational data. [Figure 7] This figure shows the results of the confidence level calculation. [Figure 8] This is an explanatory diagram to explain the correction of prediction results based on confidence levels. [Figure 9] This is a flowchart showing the operation of an information processing device. [Modes for carrying out the invention]

[0030] Hereinafter, a learning support device, a learning support method, and a learning support program according to embodiments of the present invention will be described with reference to the drawings. In the drawings, the same elements are denoted by the same reference numerals, and redundant explanations are omitted. Also, the dimensional ratios in the drawings are exaggerated for illustrative purposes and may differ from actual ratios.

[0031] In this specification, the training data used during training in machine learning, and the operational data targeted for prediction during operation (prediction), include image data and / or text data. Text data may, for example, be text data of sentences. The training data and operational data also include data other than image data and text data, such as audio data, temperature data, and odor data.

[0032] Figure 1 is a diagram showing the schematic configuration of the information processing system 10 according to the embodiment.

[0033] The information processing system 10 includes an information processing device 100 and a shooting device 200. The information processing device 100 and the shooting device 200 are connected to each other so as to be able to communicate with each other via a network or the like. The information processing device 100 constitutes a learning support device. The shooting device 200 may be used when the learning data and operational data are image data. When the learning data and operational data are text data, the shooting device 200 may be replaced by a scanner or the like that scans paper and reads the characters on the paper. When the learning data and operational data are audio data, temperature data, or odor data, the shooting device 200 may be replaced by an audio recognition device that converts audio to text data, a temperature sensor, or an odor sensor, respectively.

[0034] The imaging device 200 captures images of products (including parts) and other items to be inspected, and outputs them as image data. Products include, for example, bolts, flat plates, semiconductor chips, semiconductor wafers, and vehicle bodies. If the imaging device 200 is replaced by a scanner, the scanner scans documents (paper), for example, converts them into text data, and outputs them as document data. Documents include, for example, user feedback reports on products, failure reports, and care records. If the documents are data (electronic data), the scanner can be omitted, and the documents, which are data, can be directly input into the information processing device 100 as document data.

[0035] The imaging device 200 is comprised of, for example, a visible light camera. The imaging device 200 is installed, for example, in a production site (factory) so as to have an appropriate imaging range and direction depending on the size and position of the product to be inspected.

[0036] Image data may be a still image, for example, but it may also be a moving image.

[0037] The imaging device 200 may be used to obtain image data to be used as training data and image data to be used as operational data. The image data to be used as training data may be image data captured by imaging devices other than the imaging device 200.

[0038] The imaging device 200 transmits image data to the information processing device 100.

[0039] The information processing device 100 receives image data from the imaging device 200. If the training data and operational data are in the form of text data, the information processing device 100 may receive text data from the scanner.

[0040] The information processing device 100 extracts features from the training data and the operational data, respectively, using machine learning. Based on the features of the operational data, the information processing device 100 calculates a prediction result from the operational data using machine learning. The information processing device 100 corrects the calculated prediction result based on the features of the training data and the features of the operational data. The information processing device 100 outputs the corrected operational data. The prediction result includes the likelihood of a predetermined class. Specifically, the prediction result is, for example, the likelihood of the class with the highest likelihood among the predetermined classes. The predetermined classes include, for example, product defect classes and document classification classes. Product defect classes include, for example, scratches, chips, cracks, stains, and good products. Document classification classes include, for example, user-induced failures, equipment-induced failures, and no failures.

[0041] The functions of the information processing device 100 will be explained further. For simplicity, the following explanation will use the example where image data is product image data and text data is failure report data.

[0042] Figure 2 is a block diagram showing the hardware configuration of the information processing device 100.

[0043] The information processing device 100 includes a control unit 110, a storage unit 120, a display unit 130, an input unit 140, and a communication unit 150. These components are interconnected via a bus. The information processing device 100 may be configured as a computer.

[0044] The control unit 110 is composed of a CPU (Central Processing Unit) and performs control and calculation processing on each part of the information processing device 100 according to the program. Details of the functions of the control unit 110 will be described later.

[0045] The memory unit 120 may consist of, for example, RAM (Random Access Memory), ROM (Read Only Memory), and SSD (Solid State Drive). The RAM temporarily stores programs and data as a working area for the control unit 110. The ROM stores various programs and data in advance. The SSD stores various programs and data, including the operating system.

[0046] The display unit 130 is, for example, a liquid crystal display, and displays various information.

[0047] The input unit 140 is composed of, for example, a touch panel and various keys. The input unit 140 is used for various operations and inputs.

[0048] The communication unit 150 is an interface for communicating with external devices. Network interfaces conforming to standards such as Ethernet®, SATA, PCI Express, USB, and IEEE1394 may be used for communication. In addition, wireless communication interfaces such as Bluetooth®, IEEE802.11, and 4G may be used for communication.

[0049] The functions of the control unit 110 of the information processing device 100 will now be described in detail. The functions of the control unit 110 described below are realized when a program stored in the memory unit 120 is executed using various parameters (for example, weighting parameters of a learning model) stored in the memory unit 120.

[0050] Figure 3 is an explanatory diagram illustrating the function of the control unit 110 during training using training data. Figure 4 is an explanatory diagram illustrating the function of the control unit 110 in extracting features from training data. Figure 5 is an explanatory diagram illustrating the function of the control unit 110 in extracting features from operational data, calculating prediction results, and correcting prediction results.

[0051] (Functions of the control unit 110 during learning) As shown in Figure 3, the control unit 110 functions as a feature extractor 111a, a classifier 112a, and a learning unit 113 during training. The feature extractor 111a and the classifier 112a constitute the learning model.

[0052] The feature extractor 111a acquires training data received from the imaging device 200, etc., and stored in the memory unit 120 by reading it from the memory unit 120. The feature extractor 111a extracts features from each training data in a group of training data (hereinafter also referred to as the "training data group"). If the training data is image data, the feature extractor 111a can use a CNN (Convolutional Neural Network) to extract features from each training data in the training data group. If the training data is text data, the feature extractor 111a can use a natural language processing model such as BERT (Bidirectional Encoder Representations from Transformers) to extract features from each training data in the training data group.

[0053] The classifier 112a classifies the features of each training data into one of the predetermined classes and outputs them as prediction results. Specifically, for example, the classifier 112a calculates and outputs the likelihood for each predetermined class for the features of the training data. As a result, each training data is classified into one of the predetermined classes by classifying the features of each training data into the class with the highest likelihood. For example, a fully connected layer of a neural network is used for the classifier 112a.

[0054] The learning unit 113 can train the feature extractor 111a and the classifier 112a by backpropagation so that the difference between the correct label assigned to each training data in the training data set and the prediction result output from the classifier 112a becomes small. The correct labels can be associated with each training data by the user, input into the input unit 140, and stored in the storage unit 120. The correct labels may be a predetermined class that is the correct prediction result by the classifier 112a. The correct labels may also be predetermined class-specific likelihoods, where the likelihood of the class that is the correct prediction result is 1 and the likelihood of the other classes is 0.

[0055] (Functions of the control unit 110 when extracting features from the training data set) As shown in Figure 4, the control unit 110 functions as a feature extractor 111 when extracting features from each training data in the training data set. The feature extractor 111 may use a trained CNN or BERT model that has been trained by the learning process shown in Figure 3. The feature extractor 111 may use only a part, not all, of the trained feature extractor 111a. In this case, the level of abstraction of the features extracted by the feature extractor 111 may be lower than the level of abstraction of the features extracted by the feature extractor 111a. For the sake of simplicity, the following explanation assumes that the feature extractor 111 uses a trained CNN or BERT model that has been trained by the learning process shown in Figure 3. That is, the explanation assumes that the trained feature extractor 111a is used as the feature extractor 111. The feature extractor 111 and the classifier 112 constitute the learning model.

[0056] The feature extractor 111 extracts features from each training data point in the training data set. The feature extractor 111 stores the multiple features extracted from each training data point in the training data set (hereinafter also referred to as the "feature group") in the storage unit 120. The feature group may also be stored in the storage unit 120 as data on the distribution of features in the feature space.

[0057] (Functions of the control unit 110 during operation) As shown in Figure 5, the control unit 110 functions as a feature extractor 111, a classifier 112, a confidence calculation unit 114, and a correction unit 115 during operation. The feature extractor 111 and the classifier 112 may use a trained model that has been trained using the training data set. The feature extractor 111 constitutes the acquisition unit and the feature extraction unit. The classifier 112 constitutes the prediction result calculation unit. The correction unit 115, together with the confidence calculation unit 114, constitutes the correction unit and the output unit.

[0058] The feature extractor 111 acquires operational data by receiving it from the imaging device 200, etc. Alternatively, the feature extractor 111 may acquire operational data by reading it from the storage unit 120, which has been received from the imaging device 200, etc. and stored in the storage unit 120. The feature extractor 111 extracts features from the operational data.

[0059] The classifier 112 classifies the operational data into one of the predetermined classes and outputs the predicted results based on the operational data. Specifically, the classifier 112 calculates the class with the highest likelihood among the predetermined classes and its likelihood, and outputs it as a predicted result. The classifier 112 may also calculate the likelihood for each predetermined class and output it as a predicted result. By classifying the operational data into the class with the highest likelihood, the operational data is classified into one of the predetermined classes.

[0060] The confidence calculation unit 114 calculates the confidence level of the prediction result (hereinafter also simply referred to as "confidence level") based on the features of the training data set (training data) and the features of the operational data.

[0061] Figure 6 is an explanatory diagram illustrating the method for calculating the confidence level of prediction results based on operational data. Figure 7 shows the results of the confidence level calculation.

[0062] Figure 6 shows the distribution of features in the training data set and their relationship to the operational data in the feature space. Black circles represent features in the training data set, and white circles represent features in the operational data. The number (types) of features extracted by the feature extractor 111 is relatively large, and the feature space of the features becomes multidimensional. In the example in Figure 6, for the sake of simplicity, the features of the training data set and the features of the operational data are shown in a two-dimensional feature space that shows the relationship between two features, the first feature and the second feature.

[0063] In Figure 6, each training data point in the training dataset is associated with the same correct label. Therefore, the multiple features (feature sets) extracted from each training data point are limited to a relatively restricted range in the feature space, as indicated by the dashed line. The operational data is also operational data in which the class of the correct label associated with each training data point was predicted as the prediction result (i.e., the class with the highest likelihood among the given classes).

[0064] For operational data with features that are close in distance (more specifically, distance in the feature space) to the feature distribution of the training data set, the degree of data drift is considered to be relatively small. Prediction results from operational data with a relatively small degree of data drift are considered to have relatively high accuracy. On the other hand, for operational data with features that are far from the feature distribution of the training data set, the degree of data drift is considered to be relatively large. Prediction results from operational data with a relatively large degree of data drift are considered to have relatively low accuracy. This is because the feature extractor 111 and classifier 112 that calculate prediction results from operational data are trained on training data before data drift occurred, and not on data after data drift occurred. In Figure 6, the operational data with features shown in the feature space is considered to have relatively low prediction accuracy because the features are relatively far from the feature distribution of the training data set.

[0065] As mentioned above, the closer the distribution of features in the operational data is to the distribution of features in the training data, the higher the accuracy of the prediction results made from the operational data. Therefore, the reliability of the prediction results can be calculated based on the distance between the distribution of features in the training data and the features in the operational data. The reliability of the prediction results can be calculated, for example, using the Mahalanobis distance. The reliability of the prediction results can be calculated, for example, as the reciprocal of the Mahalanobis distance between the features in the training data and the features in the operational data. The reliability of the prediction results may be a value of 1 or less. Hereinafter, the distance between the distribution of features in the training data and the features in the operational data will also be simply referred to as the "distance from the training data".

[0066] In the example shown in Figure 7, the confidence level of the operational data shown in Figure 6 is calculated as the reciprocal of the Mahalanobis distance between the features of the training data set and the features of the operational data. As shown in Figure 7, the confidence level of the operational data features shown in Figure 6 is 0.01 (the maximum confidence level is 1). Therefore, the accuracy of the prediction results predicted from this operational data is considered to be low.

[0067] The correction unit 115 corrects the prediction results based on the reliability of the prediction results. Since the reliability is calculated based on the features of each training data in the training data set and the features of the operational data, it can be said that the correction of the prediction results is performed based on the features of the training data and the features of the operational data.

[0068] Figure 8 is an explanatory diagram illustrating the correction of prediction results based on confidence levels. The tables shown in Figure 8—the prediction results from each operational data, the confidence level calculation results according to distance, and the corrected prediction results—can be stored in the storage unit 120.

[0069] In the table of prediction results from each operational data, the prediction result from "Operational Data 1" is "Class A with a likelihood of 0.98", the prediction result from "Operational Data 2" is "Class C with a likelihood of 0.80", and the prediction result from "Operational Data 3" is "Class C with a likelihood of 0.66".

[0070] In the confidence level table based on distance, the confidence level for "Operational Data 1" is calculated as "0.95" based on the distance between each ground truth label for "Class A" and the corresponding training data set. The confidence level for "Operational Data 2" is calculated as "0.40" based on the distance between each ground truth label for "Class C" and the corresponding training data set. The confidence level for "Operational Data 3" is calculated as "0.80" based on the distance between each ground truth label for "Class C" and the corresponding training data set.

[0071] As mentioned above, the confidence level of the prediction can be calculated based on the distance between the feature distribution of the training data set and the feature distribution of the operational data. However, the confidence level of the prediction can also be calculated based on the distance between the feature distribution of a single training data point and the feature distribution of the operational data.

[0072] The correction of the prediction results is done by calculating the product of the likelihood (the likelihood of the class with the highest likelihood among a given class) which is the prediction result from the operational data, and its confidence level. In other words, the corrected prediction result is the product of the likelihood of the class with the highest likelihood which is the prediction result from the operational data, and its confidence level.

[0073] Because the prediction results are corrected based on confidence, operational data that is far from the distribution of the training data set in the feature space will have lower confidence in the prediction results. The likelihood of the class with the highest likelihood in the prediction result is corrected by a correction calculation that multiplies it by a confidence value of 1 or less, which degrades the prediction result. This ensures that the prediction results from the machine learning model are appropriately corrected.

[0074] In the table of corrected prediction results, for "Operational Data 1", the predicted result "likelihood of Class A is 0.98" is multiplied by a confidence level of "0.95", resulting in a corrected prediction result of "likelihood of Class A is 0.931". For "Operational Data 2", the predicted result "likelihood of Class C is 0.80" is multiplied by a confidence level of "0.40", resulting in a corrected prediction result of "likelihood of Class A is 0.320". For "Operational Data 3", the predicted result "likelihood of Class C is 0.66" is multiplied by a confidence level of "0.80", resulting in a corrected prediction result of "likelihood of Class A is 0.931".

[0075] Furthermore, if the prediction results derived from operational data are likelihoods for each predetermined class (i.e., likelihoods for each of the predetermined classes), the prediction results may be corrected by calculating the confidence level for each class and multiplying the likelihood level for each class by the respective confidence level.

[0076] The correction unit 115 outputs the corrected prediction result. The output includes transmission to other devices via the communication unit 150 and display on the display unit 130. The correction unit 115 may also output the uncorrected prediction result along with the corrected prediction result.

[0077] The operation of the information processing device 100 will be described below.

[0078] Figure 9 is a flowchart showing the operation of the information processing device 100. This flowchart is executed by the control unit 110 according to the program stored in the storage unit 120.

[0079] The control unit 110 acquires the learning data set and the operation data (S101). The control unit 110 can acquire the learning data set by reading it from the storage unit 120. The control unit 110 can acquire the operation data by reading it from the storage unit 120 or by receiving it from the imaging device 200 or the like.

[0080] The control unit 110 extracts the features of each training data from the training data group and the features of the operational data from the training data group and the operational data, respectively (S102). The control unit 110 may also perform a process in advance to extract the distribution of multiple features from each training data from the training data group, store them in the storage unit 120, and use this feature distribution.

[0081] The control unit 110 calculates the prediction result from the operational data (S103).

[0082] The control unit 110 calculates the confidence level of the prediction result based on the distribution of multiple features extracted from each training data in the training data group and the features of the operational data (S104).

[0083] The control unit 110 corrects the prediction result based on the reliability and outputs the corrected prediction result (S105).

[0084] This embodiment provides the following effects.

[0085] The system extracts features from both the training data and the operational data, calculates prediction results based on the operational data's features, and then corrects and outputs the prediction results based on both the training data's features and the operational data's features. This allows for the rapid and valid acquisition of prediction results using the operational data, even when data drift occurs in the operational data.

[0086] Furthermore, the correction of prediction results based on the features of the training data and the operational data is a correction that degrades the prediction results. This allows for obtaining reasonable prediction results that correspond to data drift.

[0087] Furthermore, the reliability of the prediction results is calculated based on the features of the training data and the operational data, and a correction is applied to degrade the prediction results based on the reliability. This allows for obtaining more accurate prediction results.

[0088] Furthermore, confidence is calculated based on the distance between the features of the training data and the features of the operational data. This makes it easy to obtain more reliable prediction results.

[0089] Furthermore, the prediction result from operational data is set to the likelihood of the class with the highest likelihood within a given class, and the likelihood of the prediction result from operational data is reduced according to the confidence level, thereby degrading the prediction result from operational data. This improves the validity of the prediction result.

[0090] Furthermore, multiple training data sets are acquired as a training data set, and confidence is calculated based on the distance between the distribution of multiple features extracted from each training data set in the training data set and the features of the operational data. This makes it easy to further improve the validity of the prediction results.

[0091] Furthermore, the prediction results before correction are also output. This allows for verification of the validity of the prediction results after correction.

[0092] Furthermore, the prediction results are based on predictions made from operational data by a neural network trained on training data. This allows for easier, more flexible, and more accurate prediction results to be obtained.

[0093] Furthermore, the training data and operational data are used as image data or text data, and the combination of training data and the corresponding correct labels is used as training data. A supervised machine learning model is then used to calculate the prediction results. This allows for a reasonable range of operational data to be appropriately expanded while obtaining valid prediction results.

[0094] The configuration of the information processing system 10 described above is intended to illustrate the main configuration in order to explain the features of the above-described embodiment, and is not limited to the above configuration, and can be modified in various ways within the scope of the claims. Furthermore, it does not preclude the configurations that are generally found in information processing systems.

[0095] For example, the information processing device 100 may be composed of multiple devices.

[0096] Furthermore, the flowchart described above may omit some steps, and other steps may be added. Also, some steps may be executed simultaneously, and one step may be divided into multiple steps for execution.

[0097] Furthermore, the means and methods for performing the various processing tasks in the information processing system 10 described above can be implemented using either dedicated hardware circuits or a programmed computer. The program may be provided, for example, on a computer-readable recording medium such as a USB memory stick or a DVD (Digital Versatile Disc)-ROM, or it may be provided online via a network such as the Internet. In this case, the program recorded on the computer-readable recording medium is usually transferred to and stored in a storage unit such as a hard disk. The program may also be provided as a standalone application software, or it may be incorporated as a function into the software of a device such as a detection unit. [Explanation of Symbols]

[0098] 10 Information Processing Systems, 100 Information Processing Devices, 110 Control unit, 111a, 111 Feature extractor, 112a, 112 classifier, 113 Learning Department, 114. Confidence calculation unit, 115 Correction unit, 120 storage section, 130 Display section, 140 Input section, 150 Communications Department, 200 imaging devices.

Claims

1. An acquisition unit that acquires training data and operational data, A feature extraction unit extracts the features of the training data and the features of the operational data from the training data and the operational data, respectively. A prediction result calculation unit calculates a prediction result from the operational data based on the characteristics of the operational data, A correction unit corrects the prediction result based on the features of the training data and the features of the operational data. An output unit that outputs the corrected prediction result, A learning support device having the following features.

2. The learning support device according to claim 1, wherein the correction unit performs a correction that degrades the prediction result.

3. The system further includes a reliability calculation unit that calculates the reliability of the prediction result based on the features of the training data and the features of the operational data. The learning support device according to claim 2, wherein the correction unit performs a correction that degrades the prediction result based on the reliability.

4. The learning support device according to claim 3, wherein the confidence calculation unit calculates the confidence level based on the distance between the feature quantities of the learning data and the feature quantities of the operational data.

5. The prediction result of the operational data is the likelihood of the class with the highest likelihood among the predetermined classes. The learning support device according to claim 3, wherein the correction unit degrades the prediction result from the operational data by reducing the likelihood, which is the prediction result from the operational data, in accordance with the confidence level.

6. The acquisition unit acquires a plurality of the aforementioned training data as a training data group, The learning support device according to claim 3, wherein the confidence calculation unit calculates the confidence level based on the distance between the distribution of a plurality of features extracted from each training data in the training data group and the features of the operational data.

7. The learning support device according to claim 1, wherein the output unit further outputs the prediction result before correction.

8. The learning support device according to claim 1, wherein the prediction result is the prediction result predicted from the operational data by a neural network trained on the training data.

9. The aforementioned training data and operational data are image data or text data. The learning support device according to claim 1, wherein the prediction result calculation unit calculates the prediction result using a supervised machine learning model that has been trained with the combination of the learning data and the correct label corresponding to the learning data as training data.

10. Step (a) to acquire training data and operational data, (b) a step of extracting the features of the training data and the features of the operational data from the training data and the operational data, respectively, (c) A step of calculating a prediction result from the operational data based on the characteristics of the operational data, (d) A step of correcting the prediction result based on the features of the training data and the features of the operational data, Step (e) of outputting the corrected prediction result, A learning support method that has the following characteristics.

11. The learning support method according to claim 10, wherein in step (d), a correction is made to degrade the prediction result.

12. The method further includes a step (f) of calculating the reliability of the prediction result based on the features of the training data and the features of the operational data, The learning support method according to claim 11, wherein in step (d), a correction is made to degrade the prediction result based on the confidence level.

13. The learning support method according to claim 12, wherein in step (f), the confidence level is calculated based on the distance between the features of the learning data and the features of the operational data.

14. The prediction result of the operational data is the likelihood of the class with the highest likelihood among the predetermined classes. In step (d), the learning support method according to claim 12, wherein the likelihood, which is the prediction result from the operational data, is reduced in accordance with the confidence level, thereby degrading the prediction result from the operational data.

15. In step (a) above, multiple training data are acquired as a training data set, The learning support method according to claim 12, wherein in step (f), the confidence level is calculated based on the distance between the distribution of a plurality of features extracted from each training data in the training data group and the features of the operational data.

16. The learning support method according to claim 10, wherein in step (e), the prediction result before correction is further output.

17. The learning support method according to claim 10, wherein the prediction result is the prediction result predicted from the operational data by a neural network trained on the training data.

18. The aforementioned training data and operational data are image data or text data. The learning support method according to claim 10, wherein in step (c), the prediction result is calculated using a supervised machine learning model that has been trained with the combination of the learning data and the correct label corresponding to the learning data as training data.

19. A learning support program for causing a computer to execute the learning support method described in any one of claims 10 to 18.

Images