JP2020052936A - Method, computer and program for generating training data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method or the like capable of automatically generating high-quality training data. A data generation device 10 generates a first trained model M1 by performing machine learning using verification data D1 to D3, and classifies verification data D1 to D3 using the first trained model M1. Then, the reference data D1 is selected from the verification data that are not correctly classified, the converted data T1 to T3 are generated from the reference data D1, the converted data T1 to T3 are classified, and the correctly classified converted data T3. By applying the inverse conversion of the conversion according to the above to the training data K1 and K2, new training data N2 and N3 are generated. [Selection diagram] Fig. 4

Description

  The present invention relates to a method for generating training data and the like.

  In machine learning, not only the quality but also the quantity of training data is important. In particular, in order to generate a trained model for classifying (clustering) or predicting data with high accuracy, a large amount of training data to which a correct answer is given is required.

  In order to prepare such training data, it is necessary to classify a large amount of data manually to give a correct answer, or to prepare a predicted value manually, which requires a great deal of labor. For this purpose, an example of a method for automatically generating training data is described in Patent Document 1.

JP 2014-178229 A

  However, the conventional technique has a problem that it is difficult to generate high-quality learning data.

  For example, in the configuration of Patent Literature 1, training data itself cannot be generated, and new training data is created based only on feature amounts independently of the original training data. For this reason, there is a possibility that data (for example, data that is not in an image data format or data that cannot be recognized by a human being as a photograph) that is far from actual data (for example, a photograph) may be generated. A similar problem may occur not only with image data but also with, for example, audio data.

  The present invention has been made to solve such a problem, and an object of the present invention is to provide a method, a computer, and a program capable of automatically generating high-quality training data.

In order to solve the above-described problems, a method according to the present invention is a method of generating training data included in learning data used for machine learning,
Generating a first learned model by the computer performing machine learning using the training data;
A step of: using the first learned model to obtain an output value for at least one verification data;
A step in which the computer selects at least one reference data from the verification data in which the output value of the first learned model is inappropriate;
A computer performing a plurality of different conversions on the reference data to generate a plurality of different converted data;
A step of: using the first learned model to obtain an output value for each of the converted data;
A computer generates at least one new training data by performing an inverse transformation of the transformation applied to the converted data having an appropriate output value to at least one training data. Steps to
Is provided.
Further, the method according to the present invention is a method of generating training data included in learning data used for machine learning,
Generating a first learned model by the computer performing machine learning using the training data;
A step of: using the first learned model to obtain an output value for at least one verification data;
A step in which the computer selects at least one reference data from the verification data whose output value by the first learned model is appropriate;
A computer performing a plurality of different conversions on the reference data to generate a plurality of different converted data;
A step of: using the first learned model to obtain an output value for each of the converted data;
A step of generating at least one new training data by performing, on the at least one training data, the conversion that has been performed on the converted data having an inappropriate output value among the converted data; When,
Is provided.
According to a specific mode, the training data and the verification data included in the learning data are two-dimensional image data, and the conversion is a geometric linear conversion or a color value conversion, and a feature amount extracted by a neural network is added. And deletion, and emphasizing and weakening the feature amount extracted by the neural network.
According to a specific mode, the training data and the verification data included in the learning data are audio data, and the conversion includes changing a pitch, changing a reproduction speed of a sound, adding noise, removing noise, This includes application of a low-pass filter or application of a high-pass filter, addition or deletion of a feature extracted by a neural network, and enhancement or weakening of a feature extracted by a neural network.
Further, a computer according to the present invention executes the above method.
Further, a program according to the present invention causes a computer to execute the above method.

  According to the method and the like according to the present invention, new learning data is generated by performing conversion on actual learning data, so that higher-quality learning data can be generated.

FIG. 2 is a diagram illustrating an example of a configuration of a computer according to Embodiment 1 of the present invention. FIG. 4 is a diagram showing an example of learning data according to the first embodiment. 6 is a flowchart illustrating a flow of a process according to the first embodiment. 9 is a diagram illustrating an example of conversion with respect to reference data according to Embodiment 1. FIG. FIG. 5 is a diagram showing an example of new training data generated in the first embodiment. FIG. 13 is a diagram showing an example of verification data according to the second embodiment. 9 is a flowchart illustrating a flow of a process according to the second embodiment. FIG. 19 is a diagram illustrating an example of conversion with respect to reference data according to the second embodiment. FIG. 14 is a diagram showing an example of new training data generated in Embodiment 2. FIG. 2 is a configuration diagram illustrating a case where each function of the computer in FIG. 1 is implemented by a processing circuit that is dedicated hardware. FIG. 2 is a configuration diagram illustrating a case where each function of the computer in FIG. 1 is realized by a processing circuit including a processor and a memory.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
Embodiment 1 FIG.
FIG. 1 shows an example of the configuration of a data generation device 10 according to Embodiment 1 of the present invention. The data generation device 10 functions as a device (computer) that generates learning data used for machine learning.

  As shown in FIG. 1, the data generation device 10 has a configuration as a known computer, and includes an operation unit 11 that performs an operation and a storage unit 12 that stores information. The arithmetic means 11 includes, for example, a CPU (Central Processing Unit), and the storage means 12 includes, for example, a semiconductor memory and a HDD (hard disk drive).

  The storage unit 12 stores the learning data. The learning data is learning data for generating a learned model for obtaining an output value of the data. The trained model may be, for example, a model for classifying data into any of a plurality of classes. In this case, the output value of the learned model is, for example, a value indicating which class the data belongs to. Further, the learned model may be a model for predicting future data. In this case, the output value of the learned model is, for example, a predicted value of future data. In the following, a model for classifying data will be described. However, the present invention can be applied to a model for predicting data. Although a specific method of using the learning data in the machine learning process can be designed as appropriate, it includes training data and verification data.

  The learning data is a set of data units having a predetermined data format. Each data unit includes a data portion to be classified and a correct answer label. In this specification, the expressions “classify data” and “recognize data” may both mean “estimate the correct label of data”. Hereinafter, a case where the data is image data will be described. However, the present invention is also applicable to a case where the data is audio data.

  FIG. 2 shows an example of verification data according to the first embodiment. The verification data includes verification data D1 to D3. In the present embodiment, in each data unit, the data portion to be classified is a two-dimensional image. As a specific example, the data portion of a certain data unit represents an image of a person, and the correct label of the data unit is “person”. In this case, it can be said that the data unit belongs to the class of [person]. The verification data D1 to D3 in FIG. 2 correspond to this. Another data unit represents an image of an object that is not a person, and the correct label of the data unit is “non-person”. In this case, it can be said that the data unit belongs to the class [non-person]. Three or more classes may be defined.

  The training data is also defined in the same format as the verification data. In this example, these training data can be said to be training data for generating a trained model that classifies the data into a "person" or "non-person" class. That is, the learned model generated by machine learning using these training data is a model for recognizing whether or not the image represents a person.

  The storage unit 12 of the data generation device 10 also stores a program (not shown), and the arithmetic unit 11 executes this program, so that the data generation device 10 realizes the functions described in this specification. That is, this program causes a computer to execute the method described in this specification.

  The data generation device 10 may include other components that are commonly included in a known computer. For example, a display and a printer as output devices, a keyboard and a mouse as input devices, a network interface serving as an input device and an output device for a communication network, and the like may be provided.

  FIG. 3 is a flowchart illustrating a flow of a process performed by data generation device 10 in the first embodiment. In the process of FIG. 3, first, the data generation device 10 generates a first learned model M1 by performing machine learning using the training data (step S1). The first learned model M1 is a model that classifies data into one of a plurality of classes. That is, for the data, a value indicating which class the data belongs to is output as an output value. The first learned model M1 may be any type of model, and may be configured using, for example, a neural network. The specific structure of the first learned model M1 can be appropriately designed by those skilled in the art.

  Next, the data generation device 10 classifies the verification data using the first learned model M1 (Step S2). That is, an output value is obtained for the verification data. Step S2 may be performed on at least one verification data (for example, verification data D1), but may be performed on all verification data.

  Here, among the verification data shown in FIG. 2, verification data D1 is not correctly classified (for example, it is recognized as a person), and verification data D2 and D3 are correctly classified (for example, it is recognized as a person). I do. It is assumed that the reason why the classification of the verification data D1 is incorrect is that, for example, the person in the image is too long in the vertical direction and the color is light.

  Next, the data generation device 10 selects at least one of the verification data that is not correctly classified using the first learned model M1 (that is, the verification data whose output value by the first learned model M1 is inappropriate). One reference data is selected (step S3). Here, “not correctly classified” means, for example, an image that is classified into an incorrect class (eg, an image of a person but classified into a class of [non-person]). May also be included.

  The selection criterion can be designed as appropriate. For example, it is possible to randomly select verification data that has not been correctly classified, or to select all verification data that has not been correctly classified as reference data. Good. Here, it is assumed that the verification data D1 has been selected as the reference data.

Next, the data generation device 10 performs a plurality of different conversions on the reference data to generate a plurality of different converted data (step S4).
FIG. 4 shows an example of conversion for reference data. By performing a conversion for enhancing the contrast on the verification data D1, converted data T1 is generated. Similarly, by performing a conversion for compressing the verification data D1 in the vertical direction, converted data T2 is generated, and a conversion for enhancing the contrast and a conversion for compressing the verification data D1 to the compression in the vertical direction are generated. As a result, converted data T3 is generated.

  Specific arithmetic processing corresponding to such conversion can be appropriately designed using known image processing software or other techniques.

  Here, the conversion related to the converted data T1 and the conversion related to the converted data T3 have a part in common (both include the conversion for enhancing the contrast), but are different conversions as a whole. I have. Thus, it can be said that different conversions have been performed on the converted data T1 to T3. Thus, each converted data is generated by performing different conversions on the reference data.

  Next, the data generation device 10 classifies the converted data T1 to T3 using the first learned model M1 (step S5). That is, using the first learned model M1, output values are obtained for the converted data T1 to T3. Here, it is assumed that converted data T1 and T2 are not correctly classified (for example, recognized as not being a person), and converted data T3 is correctly classified (for example, recognized as being a person).

  Here, when all the converted data are correctly classified, the process may return to step S3, and the process may be repeated from the selection of the reference data.

  Here, the reason why the classification of the converted data T3 generated from the verification data D1 is correctly performed even though the classification of the verification data D1 is incorrect is that the contrast and the length and width of the converted data T3 are different. It is assumed that the reason is that the first learned model M1 can be correctly classified because the ratio has been changed to the same degree as many other training data. In other words, it is considered that an image having a low contrast and a long length in the vertical direction like the verification data D1 was insufficient in the learning data. For this reason, a trained model that can correctly classify the verification data D1 can be generated by adding an image having a low contrast and a long length in the vertical direction like the verification data D1 to the training data and newly performing machine learning again. It is expected that the nature will increase.

  Next, the data generation device 10 generates new training data based on the content of the conversion (Step S6). More specifically, the inverse transformation of the transformation performed on correctly classified data (that is, the one whose output value is appropriate) among the converted data is performed on all the training data. In this example, since the conversion applied to the correctly classified converted data T3 is “conversion for enhancing contrast and conversion for compressing in the vertical direction”, the inverse conversion to this is, for example, “conversion for reducing contrast”. And a conversion that extends in the vertical direction ”. In this example, the order of the elements constituting the transformation is commutative, but if not, the order of the elements may be reversed.

  As described above, at least one training data as a source of new learning data may be used.

  FIG. 5 shows an example of new training data. New training data N2 is generated by performing "conversion for reducing contrast and conversion for extending in the vertical direction" on training data K1. Similarly, new training data N3 is generated by performing "conversion for reducing contrast and conversion for extending in the vertical direction" on training data K2. Similarly, conversion is performed on all training data.

  Next, the data generation device 10 generates a second learned model by performing machine learning using the training data used in step S1 and the new training data generated in step S6 (step S7). ).

  Here, the new training data N2 and N3 are images having a low contrast and a long length in the vertical direction like the verification data D1, and the second trained model generated by using these images is the verification data D1. It can be said that there is a high possibility that D1 can be correctly classified.

  As described above, the data generation device 10 according to the first embodiment of the present invention can automatically generate high-quality learning data for correctly classifying the verification data D1.

In step S5, the contents of the conversion performed in step S6 are determined by classifying the converted data T1 to T3 obtained by performing different conversions on the verification data D1. All or a part of the verification data not correctly classified is selected, and step S5 is executed. The conversion performed in step S6 may be configured to determine the inverse conversion of the conversion having many data correctly classified in step 5 as the content of the conversion to be performed. At this time, it may be configured to perform only the inverse conversion of the conversion with the most data classified most correctly, or to perform the inverse conversion of the plurality of higher-order conversions.
If the verification data that has been correctly classified like the verification data D2 and D3 in step S2 is not correctly classified in step S5, the number of verification data may be counted. If the number of verification data is larger than a certain value, the recognition rate of the first learned model M1 may not be improved even if the training data is increased by performing the inverse transformation of the transformation in step S6. Therefore, by comparing the number of pieces of verification data correctly classified in step S5 with the number of pieces of verification data not correctly classified in step S5, the difference between the number of pieces of verification data that can be correctly classified in step S5 is determined. A large conversion or a conversion that hardly generates verification data that is not correctly classified in step S5 may be preferentially selected.
Furthermore, even if D1 does not change so much as to be correctly classified, the probability of being a correct class and the probability of being an incorrect class are obtained. Paying attention to the conversion in which the probability of being an incorrect class decreases, conversely, for verification data that has been correctly classified such as D2 and D3, a conversion in which the probability of being a correct class does not decrease or decreases is selected as a candidate. You may.

  In particular, in the present embodiment, since the conversion is performed on the image, the converted data is also in a range that can be recognized as an image, and data that is not an image or data that cannot be recognized as an image by humans may be generated. Sex is low.

  Note that such an effect is not necessarily achieved for any learning data set, but high-quality data can be efficiently added to at least many learning data sets.

Embodiment 2 FIG.
The second embodiment is a modification of the method of selecting reference data and the method of generating new training data described in the first embodiment. Hereinafter, differences from the first embodiment will be described.

  FIG. 6 shows an example of learning data according to the second embodiment. The verification data includes verification data D4 to D6. The verification data D4 to D6 in FIG. 6 are examples of data belonging to the class of [person].

  FIG. 7 is a flowchart illustrating a flow of a process performed by the data generation device 10 according to the second embodiment. Steps S11 and S12 are the same as steps S1 and S2 in FIG.

  Here, of the verification data shown in FIG. 6, verification data D4 and D6 are not correctly classified (for example, recognized as a person), and verification data D5 is correctly classified (for example, recognized as a person). I do.

After step S12, the data generation device 10 selects at least one reference data from the verification data correctly classified using the first learned model M1 (step S13). In the second embodiment, unlike the first embodiment, the reference data is selected from correctly classified verification data. Here, it is assumed that the verification data D5 has been selected as the reference data.
The selection criterion can be designed as appropriate. For example, the verification data may be randomly selected from the correctly classified verification data, or all the correctly classified verification data may be selected as the reference data.

Next, the data generation device 10 generates a plurality of different converted data by performing a plurality of different conversions on the reference data, similarly to step S4 (step S14).
FIG. 8 shows an example of conversion for reference data. The converted data T4 is generated by performing a conversion to reduce the contrast on the verification data D5. Similarly, converted data T5 is generated by performing a conversion that extends in the vertical direction on the verification data D5, and converted data T6 is generated by performing a conversion that inverts the verification data D5 from side to side. You.

  Next, the data generation device 10 classifies the converted data T4 to T6 using the first learned model M1, as in step S5 (step S15). Here, it is assumed that the converted data T5 is not correctly classified (for example, recognized as a person), and the converted data T4 and T6 are correctly classified (for example, recognized as a person).

  Here, the reason why the classification of the converted data T4 and T6 was correctly performed is that an image having a low contrast such as the converted data T4 or an image which is horizontally inverted such as the converted data T6 in the training data. Suppose that the reason may be that many similar images existed in the training data. Nevertheless, the reason why the classification of the converted data T5 generated from the same verification data D5 was erroneous was that a vertically long image such as the converted data T5 was included in the training data. Suppose that the reason was that there was a shortage. Therefore, if a vertically long image such as the converted data T5 is added to the training data and the machine learning is newly performed, a trained model that can correctly classify an image similar to the converted data T5 is generated. It is expected that the possibility will increase.

  Next, the data generation device 10 generates new training data based on the content of the conversion (Step S16). More specifically, at least one piece of new training data is generated by applying the transformation that has been performed on the converted data that is not correctly classified to at least one piece of training data. In this example, new training data is generated by the conversion “conversion that extends in the vertical direction” performed on the converted data T5 that has not been correctly classified.

  It should be noted that the training data serving as the basis of the new training data may be at least one as described above, but new training data may be similarly generated for all the training data.

  FIG. 9 shows an example of new training data. By performing "vertical expansion" on the training data K3, new training data N4 is generated. Similarly, by performing “vertical expansion” on the training data K4, new training data N5 is generated. By performing “vertical expansion” on the training data K5, New training data N6 is generated.

  Next, similarly to step S7, the data generation device 10 performs machine learning using the training data used in step S11 and the new training data generated in step S16, thereby obtaining the second learned model. Is generated (step S17).

  Here, since the new training data N4 to N6 are vertically long images like the converted data T5, the second trained model generated using them is like the converted data T4. It can be said that there is a high possibility that not only an image having a low contrast and an image similar to an image which is inverted left and right like converted data T6, but also an image long in the vertical direction like converted data T5 can be correctly classified. . That is, the second learned model is a robust model that can appropriately classify more various data.

  As described above, according to the data generation device 10 according to Embodiment 2 of the present invention, high-quality training data for correctly classifying verification data or the like similar to the converted data T5 is automatically generated. be able to.

In step S15, the contents of the conversion performed in step S16 are determined by classifying the converted data T4 to T5 obtained by performing different conversions on the verification data D5. All or part of the correctly classified verification data is selected, and step S15 is executed. The conversion performed in step S16 may be configured to determine, in step 15, a conversion having many data that is not correctly classified as the content of the conversion to be performed. At this time, only the conversion with the most data that is not correctly classified may be performed, or a plurality of higher-order conversions may be performed.
If the verification data that was not correctly classified like the verification data D4 and D6 in step S12 is correctly classified in step S15, the number of verification data may be counted. If the number of verification data is larger than a certain value, the recognition rate of the first learned model M1 may not be improved even if the training data is increased by performing the conversion in step S16. Therefore, the number of verification data that is not correctly classified in step S12 is compared with the number of verification data that is correctly classified in step S15, and the difference in the number of verification data that is not correctly classified in step S15 is large. A conversion that hardly generates verification data that can be correctly classified is considered to be effective.
Furthermore, even when D5 does not change so large that it is not correctly classified, the probability of being a correct class and the probability of being an incorrect class can be obtained. Focusing on the transformation in which the probability of being a class decreases, conversely, for the verification data that has not been correctly classified such as D4 and D6, a transformation in which the probability of being an incorrect class does not decrease or decreases little is selected as a candidate. You may.

After the processing described in the first embodiment is performed, the processing described in the second embodiment is performed, and a series of the flow of returning to the processing described in the first embodiment is repeated until there is no more reference data. Is also good.
In the first and second embodiments, the following modifications can be made.
Steps S1, S7, S11 and S17 may be executed by a computer other than the data generation device 10. Also, the new training data N2 to N6 may be used for other purposes without executing step S7.

  The specific contents of the data conversion can be appropriately designed, and may include, for example, any geometric linear conversion. The geometric linear transformation means, for example, a transformation in which a 2 × 2 matrix acts on coordinates in an orthogonal two-dimensional coordinate system. Examples of the geometric linear transformation include rotation, translation, enlargement / reduction, inversion (for example, left / right inversion), a combination thereof, and the like. The enlargement / reduction includes enlargement / reduction in one or two directions. The enlargement or reduction in one direction includes vertical expansion, vertical compression, horizontal expansion, horizontal compression, diagonal expansion, diagonal compression, and the like. The expansion or contraction in two directions is, for example, expansion or compression in the vertical and horizontal directions (including those having different magnifications in the vertical and horizontal directions), but may also include expansion or compression in an oblique direction.

  Further, the conversion may include a conversion of a color value. Examples of color value conversion include change in brightness, change in color tone, enhancement or mitigation of contrast, and the like. Further, addition or deletion of a feature extracted by a neural network, enhancement or weakening of a feature extracted by a neural network, or the like may be used.

The training data and the verification data included in the learning data may be data of a type other than the image data. For example, it may be audio data. In this case, the data conversion may include a change in the pitch of the sound or a change in the reproduction speed of the sound. Also, converting the data may include adding or removing noise. The addition and removal of noise and their respective inverse transforms can be arbitrarily designed by those skilled in the art. This can be done using the techniques that are available. In the second embodiment, the data conversion further includes application of a low-pass filter (that is, removal of frequency components equal to or higher than a predetermined frequency), application of a high-pass filter (that is, removal of frequency components equal to or lower than a predetermined frequency), and the like. May be included. The application of the low-pass filter and the high-pass filter and their respective inverse transforms can be arbitrarily designed by those skilled in the art. .html>.
The data conversion may include addition or deletion of a feature extracted by a neural network, and enhancement or weakening of a feature extracted by a neural network.

  Any model can be used as a model for performing machine learning. For example, a neural network may be used, or another model may be used.

  In the first and second embodiments, the learned model, the training data, and the verification data for classifying the data have been described. However, the learning model, the training data, and the verification data for performing data prediction can be similarly performed. It is possible. In that case, the output value of the learned model is a predicted value. Also, whether or not the output value is appropriate can be determined based on whether or not the prediction accuracy is equal to or more than a predetermined threshold, instead of whether or not the classification is correctly performed. For example, a square error between a predicted value and a correct value is calculated, and if the square error is less than a threshold value, it is determined that the output value is appropriate (correctly predicted), and the square error is equal to or greater than the threshold value. Then, it may be determined that the output value is inappropriate (it was not correctly predicted).

  In Steps S7 and S17, if a part of the training data is expanded instead of the entire training data, the new training data and the training data based on the new training data are set so as not to affect the tendency of the entire data. Alternatively, the amount of change in weight at the time of learning may be smaller than usual. For example, when the normal weight change amount (or coefficient) is η and n new training data are generated from one original training data, the original training data and the new training data are generated. For the data, the weight change amount (or coefficient) may be η / (n + 1). As a specific example, in the first embodiment, with respect to the training data K1 and the training data K2 and the new training data N2 and N3 generated from the training data K1 and the training data K2, the weight at the time of learning using these data in step S7. The change amount may be １ ／ of the normal amount.

  Each function in the data generation device 10 according to the first and second embodiments is realized by a processing circuit. The processing circuit that realizes each function may be dedicated hardware or a processor that executes a program stored in a memory. FIG. 10 is a configuration diagram showing a case where each function of the data generation device 10 according to Embodiments 1 and 2 of the present invention is realized by a processing circuit 1000 which is dedicated hardware. FIG. 11 is a configuration diagram showing a case where each function of the data generation device 10 according to the first and second embodiments of the present invention is realized by a processing circuit 2000 including a processor 2001 and a memory 2002.

  When the processing circuit is dedicated hardware, the processing circuit 1000 includes, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), and an FPGA (Field Programmable Gate Array). ) Or a combination thereof. The function of each unit of the data generation device 10 may be realized by an individual processing circuit 1000, or the function of each unit may be realized by the processing circuit 1000 collectively.

  On the other hand, when the processing circuit is the processor 2001, the function of each unit of the data generation device 10 is realized by software, firmware, or a combination of software and firmware. Software and firmware are described as programs and stored in the memory 2002. The processor 2001 reads out and executes a program stored in the memory 2002 to realize the function of each unit. That is, the data generation device 10 includes a memory 2002 for storing a program that results in steps S1 to S7 or S11 to S17 when executed by the processing circuit 2000.

  It can be said that these programs cause a computer to execute the procedures or methods of the above-described units. Here, the memory 2002 includes, for example, a random access memory (RAM), a read only memory (ROM), a flash memory, an erasable programmable read only memory (EPROM), and a non-volatile memory such as an EEPROM. Or volatile semiconductor memory. In addition, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD, and the like also correspond to the memory 2002.

  In addition, about the function of each part mentioned above, you may make it implement | achieve a part by exclusive hardware and implement | achieve a part by software or firmware.

  As described above, the processing circuit can realize the functions of the above-described units by hardware, software, firmware, or a combination thereof.

  10 Data generation device (computer), D1 to D6 verification data (D1, D5 reference data), M1 first learned model, K1 to K5 training data, N2 to N6 new training data, T1 to T6 converted data.

Claims (6)

A method for generating training data included in learning data used for machine learning,
Generating a first learned model by the computer performing machine learning using the training data;
A step of: using the first learned model to obtain an output value for at least one verification data;
A step in which the computer selects at least one reference data from the verification data in which the output value of the first learned model is inappropriate;
A computer performing a plurality of different conversions on the reference data to generate a plurality of different converted data;
A step of: using the first learned model to obtain an output value for each of the converted data;
A computer generates at least one new training data by applying an inverse transformation of the transformation applied to the converted data having an appropriate output value to at least one training data. Steps to
A method comprising: A method for generating training data included in learning data used for machine learning,
Generating a first learned model by the computer performing machine learning using the training data;
A step of: using the first learned model to obtain an output value for at least one verification data;
A step in which the computer selects at least one reference data from the verification data whose output value by the first learned model is appropriate;
A computer performing a plurality of different conversions on the reference data to generate a plurality of different converted data;
A step of: using the first learned model to obtain an output value for each of the converted data;
A step of generating at least one new training data by performing, on the at least one training data, the conversion that has been performed on the converted data having an inappropriate output value among the converted data; When,
A method comprising:   The training data and the verification data included in the learning data are two-dimensional image data, and the conversion includes geometric linear conversion, color value conversion, addition or deletion of a feature extracted by a neural network, or a neural network. The method according to claim 1, further comprising a conversion for enhancing or weakening the extracted feature amount.   The training data and the verification data included in the learning data are voice data, and the conversion includes changing a pitch, changing a voice playback speed, adding noise, removing noise, applying a low-pass filter, and applying a high-pass filter. 3. The method according to claim 1, wherein the method includes applying the formula, adding or deleting a feature extracted by the neural network, or enhancing or weakening the feature extracted by the neural network.   A computer for performing the method according to claim 1.   A program for causing a computer to execute the method according to claim 1.

Images