Learning device, learning method, and learning program

Abstract

To provide a learning device, a learning method and a learning program for fast learning. [Solution] The learning device includes a data expansion processing unit that creates increased learning candidate data based on original image data, a confidence evaluation unit that assigns a confidence evaluation value Es to the learning candidate data, a learning data extraction unit that extracts learning data from the learning candidate data in order of the confidence evaluation value Es, and a learning processing unit that performs learning based on the learning data. The learning data extraction unit extracts learning data until the confidence evaluation value Es assigned to the learning candidate data becomes equal to a preset confidence evaluation threshold value α.

Description

[Technical field]

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

[0002] Conventionally, there has been a learning method based on deep learning using a neural network that selects and expands images that are efficient for learning a model as teacher data from prepared images (for example, Patent Document 1). The conventional learning method has ingenious data expansion and data selection processes, and is suitable for learning large amounts of data.

[0003] Here, in the conventional learning method, by adding a condition that the same image as before the data expansion process is not selected during the data selection process, the overlapping learning of similar images increased by the data expansion process is reduced, and the training images are not biased. However, in the conventional learning method, the number of images used as training data for one learning session is reduced, so it is difficult to learn the features of the images, and the number of learning loops increases, which may result in a long time being required to complete the learning. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2024-002431 A Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present invention is to provide a learning device, a learning method and a learning program that enable fast learning. [Means for solving the problem]

[0006] (1) A learning device according to one aspect of the present invention includes a data expansion processing unit that creates increased learning candidate data based on original image data, a confidence evaluation unit that assigns a confidence evaluation value to the learning candidate data, a learning data extraction unit that extracts learning data from the learning candidate data in order of the confidence evaluation values, and a learning processing unit that performs learning based on the learning data, wherein the learning data extraction unit extracts the learning data until the confidence evaluation value assigned to the learning candidate data becomes equal to a predetermined confidence evaluation threshold. (2) In the above (1), a representative image selection processing unit may be provided that selects, from the candidate learning data, a representative image of a group of extended images having common original image data. (3) In the above (2), the learning data extraction unit may extract the learning data until the number of types of the representative images reaches a preset representative image type number threshold. (4) In the above (2), the learning data extraction unit may include in the learning data the learning candidate data assigned a smaller certainty evaluation value than the learning candidate data assigned the largest certainty evaluation value among the representative images. (5) A learning method according to one aspect of the present invention includes a data expansion processing step of creating increased learning candidate data based on original image data, a confidence evaluation step of assigning a confidence evaluation value to the learning candidate data, a learning data extraction step of extracting learning data from the learning candidate data in order of the confidence evaluation values, and a learning processing step of performing learning based on the learning data, wherein in the learning data extraction step, the learning data is extracted until the confidence evaluation value assigned to the learning candidate data becomes equal to a predetermined confidence evaluation threshold value. (6) A learning program according to one embodiment of the present invention is a learning program that causes a computer to execute a data augmentation processing function that creates increased learning candidate data based on original image data, a confidence evaluation function that assigns a confidence evaluation value to the learning candidate data, a learning data extraction function that extracts learning data from the learning candidate data in order of the confidence evaluation values, and a learning processing function that performs learning based on the learning data, wherein the learning data extraction function extracts the learning data until the confidence evaluation value assigned to the learning candidate data becomes equal to a predetermined confidence evaluation threshold. Effect of the Invention

[0007] According to the present invention, it is possible to provide a learning device, a learning method, and a learning program that enable fast learning. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating an overview of a learning device. [Diagram 2] FIG. 1 is a diagram for explaining an outline of the flow of a learning method. [Diagram 3] FIG. 13 is a diagram showing an image of a learning method in the case where extraction of learning data is limited by the number of representative images. [Figure 4] FIG. 13 is a diagram showing an image of a learning method in the case where extraction of learning data is restricted by a confidence evaluation threshold value. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] (Embodiment) Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram for explaining an overview of a learning device 100. FIG. 2 is a diagram for explaining an overview of a flow of a learning method. FIG. 3 is a diagram showing an image of a learning method in a case where extraction of learning data is limited by the number of representative images. FIG. 4 is a diagram showing an image of a learning method in a case where extraction of learning data is limited by a confidence evaluation threshold value α. Note that, hereinafter, parts having common functions may be given the same reference numerals or symbols.

[0010] (Learning device) A learning device 100 according to this embodiment will be described with reference to Fig. 1. As shown in Fig. 1, the learning device 100 according to this embodiment includes a data augmentation processing unit 10, a confidence factor evaluation unit 20, a representative image selection processing unit 30, a learning data extraction unit 40, a learning processing unit 50, and a control unit 60.

[0011] The learning device 100 is a device that allows a machine learning model (hereinafter, referred to as a model) to learn so that it can perform appropriate evaluation of target data (e.g., image data). Here, the model is a mechanism (algorithm) in machine learning that receives input values ​​that can be discriminated by a computer and outputs an evaluation value.

[0012] The model recognizes the contents of the target data and performs evaluation. For example, if the target data is image data, the model evaluates the quality of the subject (whether it is good or bad, etc.).

[0013] The model has a structure including a neural network and parameters. The neural network has a structure in which multiple neurons are connected. The neural network may be, for example, a hierarchical multi-layer neural network in which layers in which multiple neurons are grouped are connected. The neural network is defined by the number of neurons and the connection relationship. The connection strength between neurons or layers is defined using parameters (weighting coefficients, etc.). In the neural network, target data is input, and the target data is evaluated based on the calculation results of the multiple neurons and the parameters. The model has, for example, a function of classifying target data (learning data) into either a good product label or a bad product label, or a function of calculating a good product score.

[0014] The learning device 100 includes a computer (not shown) as its main hardware. The computer includes a processor such as a CPU or GPU, a storage device such as a RAM, a ROM, a hard disk or a solid-state drive (SSD), a communication unit for connecting to a network such as the Internet, and the like. The storage device of the computer stores a learning program for operating the computer as the learning device 100 of this embodiment. The computer includes a cloud computing system. Each functional unit of the learning device 100 is realized by the computer as hardware and the learning program as software.

[0015] The data augmentation processing unit 10 is a functional unit that generates increased learning candidate data (learning candidate data set) based on the original image data. This augmentation process can improve the quality and quantity of image data, so-called padding, and can generate a large amount of data that is efficient and high quality for learning a machine learning model from a small amount of data, thereby improving the performance of the model. The original image data is augmented by, for example, a flip process that creates data by flipping the original image data horizontally or vertically or rotating it 180 degrees, a shift process that creates data by shifting the coordinates of the original image data in the X direction or Y direction, a minute rotation process that creates data by rotating the original image data by a minute angle in the right direction or left direction, and a filter process that creates data by applying a filter such as a median filter or a Gaussian filter to the original image data.

[0016] The confidence evaluation unit 20 is a functional unit that assigns a confidence evaluation value Es, calculated by a predetermined formula, to the learning candidate data. The confidence evaluation value Es is an index that indicates the expected degree of contribution of learning the data to improving the judgment performance of the model. In other words, the confidence evaluation value Es is an index that indicates how confidently the model evaluates (judges) the data at the time of learning the data.

[0017] [Confidence rating value] In obtaining the confidence evaluation value Es, the learning device 100 (certainty evaluation unit 20) calculates, for each learning candidate data, a good-quality distance, which is a distance in a feature space between the learning candidate data and the good-quality data, which is learning data to which a good-quality label has been assigned (teacher data that is the basis of machine learning of the model), based on the corresponding feature amount. In this way, the confidence evaluation value Es may be a parameter calculated based on the distance in a feature space between the learning candidate data and the learning data. Note that, as an example, the "distance" used to express the good-quality distance and the defective distance hereinafter may be the Euclidean distance between data projected in the feature space. As long as the distance in the feature space can be calculated, it is not limited to the Euclidean distance, and the Mahalanobis distance or the like may also be used. The distance between the learning data, which is one data of the learning data group, and the learning candidate data, which is one data of the learning candidate data group, is calculated, for example, by the following formula 1.

number

[0018] If the distance between the learning candidate data and the good-quality data OKg, which is one of the good-quality data OK, is d(OKg,s), d(OKg,s) is expressed as the following formula 2 using formula 1. In OKg, OK is a code indicating a good product, and g is an integer equal to or less than the number (m) of data items in the good-quality data OK.

number

[0019] If the set of distances between the learning candidate data and each of the non-defective data OK is a vector d(OK, s), the vector d(OK, s) is expressed as the following formula 3 using formula 2.

number

[0020] The non-defective distance E(OK,s) in the learning candidate data is, for example, the minimum value among the elements of the vector d(OK,s). In other words, the non-defective distance E(OK,s) is the minimum value among the elements of the vector d(OK,s), which is a set of distances between the learning candidate data and each of the non-defective data OK. The non-defective distance E(OK,s) is expressed as the following formula 4 using formula 3. In this case, the smaller the non-defective distance E(OK,s) is, the closer the learning candidate data is to one of the non-defective data OK in the feature space.

number

[0021] The non-defective distance E(OK,s) in the learning candidate data may be calculated by, for example, extracting a smallest elements from the elements of the vector d(OK,s) and averaging the a elements. a is a natural number, for example, 3. In this case, the non-defective distance E(OK,s) is expressed as the following formula 5 using formula 3. In this case, the smaller the non-defective distance E(OK,s), the closer the learning candidate data is to multiple (a pieces of) OK non-defective data in the feature space, and the closer the learning candidate data is to a group of OK non-defective data (non-defective cluster).

number

[0022] Furthermore, the learning device 100 (certainty evaluation unit 20) calculates, for each piece of learning candidate data, a defective distance, which is the distance in the feature space between the learning candidate data and the defective data, which is learning data to which a defective label has been assigned, based on the corresponding feature amount. If the distance between the learning candidate data and the defective data NGh of the defective data NG is d(NGh,s), d(NGh,s) is expressed as the following formula 6 using formula 1. In NGh, NG is a code indicating a defective product, and h is an integer equal to or less than the number (n) of data in the defective data NG.

number

[0023] If the set of distances between the learning candidate data and each defective data NG is a vector d(NG, s), the vector d(NG, s) is expressed as the following formula 7 using formula 6.

number

[0024] The defective product distance E(NG,s) in the learning candidate data is, for example, the minimum value among the elements of the vector d(NG,s). In other words, the defective product distance E(NG,s) is the minimum value among the distances between the learning candidate data and each defective product data NG. The defective product distance E(NG,s) is expressed as the following formula 8 using formula 7. In this case, the smaller the defective product distance E(NG,s) is, the closer the learning candidate data is to one of the defective product data NG in the feature space.

number

[0025] The defective distance E(NG,s) in the learning candidate data may be calculated by, for example, extracting the smallest a elements from the elements of the vector d(NG,s) and averaging the a elements. In this case, the defective distance E(NG,s) is expressed as the following formula 9 using formula 7. In this case, the smaller the defective distance E(NG,s), the closer the learning candidate data is to multiple (a pieces) of defective data NG in the feature space, and the closer the learning candidate data is to a group of defective data NG (defective cluster).

number

[0026] Furthermore, the learning device 100 (certainty evaluation unit 20) calculates a certainty evaluation value Es in the learning candidate data using the calculated good distance E(OK,s) and defective distance E(NG,s). The certainty evaluation value Es is, for example, a value obtained by dividing the good distance E(OK,s) by the defective distance E(NG,s), and is expressed by the following formula 10.

number

[0027] For example, the smaller the confidence evaluation value Es is than 1, the smaller the good-quality distance E(OK,s) is compared to the defective distance E(NG,s), and the learning candidate data is closer to the good-quality cluster than to the defective cluster. Therefore, when the learning candidate data is data with a defective label, the smaller the confidence evaluation value Es is, the more difficult it is for the learning candidate data to be classified into a good-quality label or a defective label in the model based on the learning results of the current learning data group, and the higher the learning effect of the model.

[0028] On the other hand, for example, the greater the confidence evaluation value Es is, the smaller the defective distance E(NG,s) is compared to the good distance E(OK,s), and the learning candidate data is closer to the defective cluster than to the good cluster. Therefore, when the learning candidate data is data with a good label, the greater the confidence evaluation value Es is, the more difficult it is for the learning candidate data to be classified into a good label or a bad label in the model based on the learning results of the current learning data group, and the greater the learning effect of the model.

[0029] The representative image selection processing unit 30 is a functional unit that selects, from among the learning candidate data, a representative image of a group of extended images having common original image data. For example, as shown in Fig. 3 and Fig. 4, three circular marks in the learning candidate data constitute one extended image group, and the one with the smallest certainty evaluation value Es (when sorted in ascending order) among the three circular marks is the representative image of the extended image group. When the learning candidate data are sorted in ascending order of the certainty evaluation value Es, the representative image selection processing unit 30 selects the first circular shape, the third triangular shape, the sixth diamond shape, and the nth chestnut shape in the order of the learning candidate data.

[0030] The learning data extraction unit 40 is a functional unit that performs an extraction process of calculating a confidence evaluation value Es for a learning candidate data set as a population and extracting, as learning data, learning candidate data whose confidence evaluation value Es is within a predetermined range (for example, equal to or less than a predetermined confidence evaluation threshold value α). In other words, the learning data extraction unit 40 extracts learning data from the learning candidate data in the order of the confidence evaluation value Es.

[0031] The learning processing unit 50 is a functional unit that causes the model to perform learning processing at a predetermined learning rate. The learning processing is processing for adjusting the parameters of the model so as to approach optimal values.

[0032] The control unit 60 is a functional unit that controls the learning data extraction unit 40 and the learning processing unit 50 to perform effective learning processing for the model.

[0033] (How to learn) Next, the flow of the learning method according to the embodiment will be described. The learning method can be implemented using the learning device 100. Each step (process or process) in the learning method can be executed by a corresponding unit (each module) in the learning device 100. FIG. 2 is a diagram for explaining an overview of the flow of the learning method. FIG. 3 is a diagram showing an image of the learning method when the extraction of learning data is limited by the number of representative images. FIG. 4 is a diagram showing an image of the learning method when the extraction of learning data is limited by a confidence evaluation threshold value α.

[0034] (1) In advance, the learning device 100 acquires a dataset that will serve as the original image data.

[0035] (2) Next, as shown in Fig. 2, the learning device 100 creates increased learning candidate data (learning candidate data set) based on the original image data by the data augmentation processing unit 10 (data augmentation processing step) (processing A). This process makes it possible to improve the quality and quantity of image data, thereby padding the data, and generate a large amount of data that is efficient and high quality for learning a machine learning model from a small amount of data, thereby improving the performance of the model.

[0036] (3) Next, the learning device 100 assigns a confidence evaluation value Es calculated by a predetermined formula to the learning candidate data by the confidence evaluation unit 20 (confidence evaluation step). Then, the learning device 100 selects a representative of the extended image group having common original image data from the learning candidate data set as a representative image by the representative image selection processing unit 30 using the confidence evaluation value Es assigned to each learning candidate data (image) (processing B). Here, the representative image may be an image to which the smallest confidence evaluation value Es is assigned among a plurality of image data (extended image group) extended from the common original image data (an image to which no image having the same extended source is selected during processing B). In other words, in processing B, a process is performed in which the image with the smallest evaluation value when images are arranged in ascending order of evaluation value is set as the representative image, and subsequent images with the same original image data are not set as the representative image. For example, as shown in FIG. 3, the round image in the "learning candidate data" column in the first row is selected as the representative image, and the round images in the second and fourth rows, which have the same original image data, are not selected as the representative image.

[0037] (3-1) Here, the learning data extraction unit 40 may extract learning data until the certainty evaluation value Es assigned to the learning candidate data becomes equal to a preset certainty evaluation threshold value α (learning data extraction step).

[0038] (4) Next, the number of representative images selected in process B is detected. If a representative image is detected, proceed to process D. If a representative image is not detected, perform a process to end learning (process C).

[0039] (5) Here, the selected representative images are sorted in order of the certainty evaluation value Es (e.g., ascending order) and classified into those for learning and those for other, for example, based on a predetermined value, with those with smaller certainty evaluation value Es being used for learning and those with larger certainty evaluation value Es being used for other (Process D).

[0040] (5-1) Here, in process D, the learning device 100 may select representative images until the certainty evaluation value Es reaches a preset certainty evaluation threshold value α (Condition 1). This makes it possible to make it difficult for inefficient candidate learning data for learning to be extracted as learning data (teacher data), and to extract an appropriate amount of learning data that is neither too little nor too much.

[0041] (5-2) Alternatively, in process D, the learning device 100 may execute the classification process until the number of representative images reaches a predetermined specified number (condition 2). In other words, the learning data extraction unit 40 of the learning device 100 continues to extract learning data until the number of types of representative images reaches a preset representative image type number threshold value β. This makes it possible to extract an appropriate amount of learning data that is neither too little nor too much.

[0042] (5-3) Alternatively, in process D, the learning device 100 may execute the process until all of the representative images are classified (condition 3). This makes it possible to extract an appropriate amount of learning data that is neither too little nor too much. The above conditions 1 to 3 may be applied either alone or in any combination.

[0043] (6) Representative images classified for learning by process D are extracted as learning data (process E).

[0044] (6-1) Here, images with a smaller confidence evaluation value Es than the representative image are extracted as learning data (process G). In other words, the learning data extraction unit 40 includes in the learning data learning candidate data with a smaller confidence evaluation value Es than the learning candidate data with the largest confidence evaluation value Es among the representative images. This makes it possible to extract all extended images that have the same original image data as the representative image, so that it is possible to suppress a shortage of learning data and extract a sufficient amount of learning data with a high learning effect, compared to a case in which other learning candidate data that have the same original image data as the representative image are not extracted.

[0045] (7) After process E, learning is performed based on the learning data extracted up to process E (process F). Note that a termination condition is set to stop learning when a predetermined number of times that is expected to result in proper learning is exceeded. After learning in process F, the process returns to process B.

[0046] (8) As appropriate, between process A and process B, a pre-trained model is set as the initial value (process H).

[0047] In this way, the learning method according to the present embodiment uses the learning device 100 to extract learning data until it satisfies the predetermined conditions (conditions 1 to 3 above) based on a representative image in the original image data that is the source of data expansion, so that it is possible to reliably extract high-quality learning data of different types necessary for learning while extracting a sufficient amount of learning data necessary for learning that is not too much. Thus, it is possible to provide a learning device 100 and a learning method that can learn quickly.

[0048] (Study Program) The learning program has a function of making a computer execute each process performed at each step in the above-mentioned learning method. Specifically, the learning program has a data augmentation processing function, a confidence evaluation function, a learning data extraction function, and a learning processing function corresponding to each step in the above-mentioned learning method. This makes it possible to generate high-quality, efficient data and a sufficient large amount of data for learning a machine learning model from a small amount of data, thereby improving the performance of the model. Therefore, a learning program with fast learning can be provided.

[0049] The technical scope of the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention.

[0050] In addition, the components in the above-described embodiment may be replaced with known components as appropriate without departing from the spirit of the present invention. Furthermore, the above-described modified examples may be combined as appropriate without departing from the spirit of the present invention.

[0051] As described above, the learning device 100 according to the embodiment includes a data augmentation processing unit 10 that creates increased learning candidate data based on original image data, a confidence evaluation unit 20 that assigns a confidence evaluation value Es to the learning candidate data, a learning data extraction unit 40 that extracts learning data from the learning candidate data in the order of the confidence evaluation value Es, and a learning processing unit 50 that performs learning based on the learning data. The learning data extraction unit 40 extracts learning data until the confidence evaluation value Es assigned to the learning candidate data becomes equal to a preset confidence evaluation threshold value α. This makes it possible to generate a large amount of data that is efficient and of high quality for learning a machine learning model from a small amount of data, thereby improving the performance of the model. Therefore, a learning device 100 that learns quickly can be provided.

[0052] The learning method according to the embodiment includes a data expansion processing step of creating increased learning candidate data based on original image data, a confidence evaluation step of assigning a confidence evaluation value Es to the learning candidate data, a learning data extraction step of extracting learning data from the learning candidate data in the order of the confidence evaluation value Es, and a learning processing step of performing learning based on the learning data. In the learning data extraction step, the learning data is extracted until the confidence evaluation value Es assigned to the learning candidate data becomes equal to a preset confidence evaluation threshold value α. This makes it possible to generate a large amount of data that is efficient and of high quality for learning a machine learning model from a small amount of data, thereby improving the performance of the model. Therefore, a learning method that allows fast learning can be provided.

[0053] The learning program according to the embodiment is a learning program that causes a computer to execute a data augmentation processing function for creating increased learning candidate data based on original image data, a confidence evaluation function for assigning a confidence evaluation value Es to the learning candidate data, a learning data extraction function for extracting learning data from the learning candidate data in the order of the confidence evaluation value Es, and a learning processing function for performing learning based on the learning data. In the learning data extraction function, the learning program extracts learning data until the confidence evaluation value Es assigned to the learning candidate data becomes equal to a preset confidence evaluation threshold value α. This makes it possible to generate a large amount of data that is efficient and of high quality for learning a machine learning model from a small amount of data, thereby improving the performance of the model. Therefore, a learning program that can learn quickly can be provided. [Explanation of symbols]

[0054] 100 Learning Device 10 Data expansion processing section 20 Confidence Evaluation Unit 30 Representative image selection processing unit 40 Learning data extraction unit 50 Learning processing unit 60 Control section α Confidence evaluation threshold β Representative image type number threshold

Claims

1. a data augmentation processing unit that creates increased learning candidate data based on the original image data; a confidence evaluation unit that assigns a confidence evaluation value to the learning candidate data; a learning data extraction unit that extracts learning data from the learning candidate data in order of the confidence evaluation value; A learning processing unit that performs learning based on the learning data; a representative image selection processing unit that selects a representative image from the learning candidate data, the representative image being a representative of a group of extended images having the same original image data, and extracts the representative image as the learning data; The learning data extraction unit extracts the learning data until the certainty evaluation value assigned to the learning candidate data becomes a preset certainty evaluation threshold value. Learning device.

2. The learning data extraction unit extracts the learning data until the number of types of the representative images reaches a preset representative image type number threshold. The learning device according to claim 1 .

3. The learning data extraction unit includes, in the learning data, learning candidate data assigned a smaller certainty evaluation value than the learning candidate data assigned a largest certainty evaluation value among the representative images. The learning device according to claim 1 .

4. A data augmentation processing step for generating increased candidate learning data based on the original image data; a confidence evaluation step of assigning a confidence evaluation value to the learning candidate data; a learning data extraction step of extracting learning data from the learning candidate data in order of the confidence evaluation value; A learning processing step of performing learning based on the learning data; a representative image selection process step of selecting, from the learning candidate data, a representative image of a group of extended images having the same original image data as a representative image and extracting the representative image as the learning data; In the learning data extraction step, the learning data is extracted until the confidence evaluation value assigned to the learning candidate data reaches a preset confidence evaluation threshold value. How to learn.

5. On the computer, A data augmentation processing function for creating increased candidate learning data based on the original image data; a confidence evaluation function that assigns a confidence evaluation value to the learning candidate data; a learning data extraction function for extracting learning data from the learning candidate data in order of the certainty evaluation value; A learning processing function for performing learning based on the learning data; a representative image selection processing function for selecting, from the learning candidate data, a representative image of a group of extended images having the same original image data as a representative image and extracting the representative image as the learning data; A learning program for carrying out the following: In the learning data extraction function, the learning data is extracted until the certainty evaluation value assigned to the learning candidate data reaches a preset certainty evaluation threshold value. Study program.

Images