KR102039146B1 - Method for performing multi-domain ensemble learning and apparatus thereof - Google Patents

Abstract

Provided are a method for multi-domain ensemble learning capable of guaranteeing the performance improvement effect of multi-domain learning and an apparatus thereof. According to some embodiments of the present disclosure, the method for multi-domain ensemble learning may learn a first output layer specific to a first domain using a first dataset belonging to the first domain, learn a second output layer specific to a second domain using a second dataset belonging to the second domain, and learn a third output layer using the first and second datasets. In addition, the method for multi-domain ensemble may guarantee the performance improvement effect according to multi-domain ensemble learning by performing prediction for each domain by further using a predictive value of the third output layer.

Description

Multi-domain ensemble learning method and apparatus therefor {METHOD FOR PERFORMING MULTI-DOMAIN ENSEMBLE LEARNING AND APPARATUS THEREOF}

The present disclosure relates to a multi-domain ensemble learning method and apparatus thereof. More specifically, in performing multi-domain learning on a neural network including a plurality of output layers, a method and a device for performing the method that can ensure the performance improvement effect according to the multi-domain ensemble learning It is about.

Multi-domain learning is a technique for learning a plurality of domains simultaneously through one model.

The goal of multi-domain learning is to reduce the number of learning parameters and improve the performance of the model when learning the shared features that can work well in multiple domains simultaneously.

In general, multi-domain learning may be utilized when the similarity between domains is high. However, when the data set of the target domain is small, it is not easy to improve the prediction performance of the learning model for the target domain even when multi domain learning is performed using the data set of the source domain.

Korean Patent Publication No. 10-2018-0099119

Technical problem to be solved by some embodiments of the present disclosure, a method that can ensure the performance improvement effect according to the multi-domain ensemble learning in performing multi-domain learning on the neural network including a plurality of output layers It is to provide an apparatus for performing the method.

Another technical problem to be solved through some embodiments of the present disclosure, in performing multi-domain learning on a neural network including a plurality of output layers, by utilizing the similarity between domains to improve the performance of the multi-domain ensemble learning It is to provide a method that can be guaranteed and an apparatus for performing the method.

Another technical problem to be solved by some embodiments of the present disclosure, in performing multi-domain learning on a neural network including a plurality of layers, by reflecting the inference result of each domain as a weight according to the multi-domain ensemble learning The present invention provides a method capable of ensuring a performance improvement effect and an apparatus for performing the method.

The technical problems of the present disclosure are not limited to the technical problems mentioned above, and other technical problems that are not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above technical problem, a multi-domain ensemble learning method according to an embodiment of the present disclosure, in a method for performing multi-domain ensemble learning for a neural network including a plurality of output layers in a computing device, the first domain Training a first output layer of the plurality of output layers using a first dataset belonging to; training a second output layer of the plurality of output layers using a second dataset belonging to a second domain. And learning a third output layer from the plurality of output layers using the first data set and the second data set.

In some embodiments, the first dataset may include more data than the second dataset.

In some embodiments, the first dataset may include a 2D image, and the second dataset may include a 3D image. In this case, the first data set may include an FFDM image, and the second data set may include a DBT image.

In some embodiments, the first dataset may comprise a single layer image, and the second dataset may comprise a multi-layer image. At this time. The training of the second output layer may include extracting a first layer image from the multilayer image, training the second output layer using the first layer image, and a second layer from the multilayer image. And extracting an image and learning the second output layer using the second layer image.

In some embodiments, the neural network further comprises a feature extraction layer shared by the first to third output layers, wherein the feature extraction layer is learned by the first dataset and the second dataset. It may be a layer.

In some embodiments, the method may further include predicting a label of prediction data belonging to the first domain using the neural network, wherein the predicting may include a first prediction value and a third prediction value output from the first output layer. Predicting the label based on the second prediction value output from the output layer.

In some embodiments, predicting the label based on a first prediction value output from the first output layer and a second prediction value output from the third output layer may include: applying the first prediction value and the second prediction value to each of the first prediction value and the second prediction value. The method may further include assigning a weight to the first weight value, synthesizing the first predicted value and the second predicted value by reflecting the weighted weight, and predicting the label based on the synthesized predicted value.

In order to solve the above technical problem, a multi-domain ensemble learning apparatus according to another embodiment of the present disclosure executes a memory for storing one or more instructions and the stored one or more instructions, thereby performing a first belonging to a first domain. A first output layer of the plurality of output layers is trained using a dataset, a second output layer of the plurality of output layers is trained using a second dataset belonging to a second domain, and the first dataset. And a processor configured to learn a third output layer from the plurality of output layers using the second data set.

In order to solve the above technical problem, a computer program according to another embodiment of the present disclosure is coupled to a computing device and includes a first output of a neural network including a plurality of output layers using a first dataset belonging to a first domain. Training a layer, training a second output layer of the plurality of output layers using a second dataset belonging to a second domain, and learning the plurality of output data using the first dataset and the second dataset The program may be a program stored in a computer-readable recording medium for executing the learning of the third output layer.

1 and 2 are diagrams for describing a machine learning apparatus and a learning environment according to some embodiments of the present disclosure.
3 is a flowchart of a multi-domain ensemble learning method according to some embodiments of the present disclosure.
FIG. 4 is a flowchart illustrating a detailed process of the dataset acquiring step S100 illustrated in FIG. 3.
5 is a flowchart illustrating a detailed process of an output layer learning step S300 illustrated in FIG. 3.
6 is a flowchart of a method of predicting a label through multi-domain ensemble learning according to some embodiments of the present disclosure.
FIG. 7 is a flowchart illustrating a detailed process of the label predicting step S800 illustrated in FIG. 6.
8 is a hardware diagram illustrating an example computing device that may implement an apparatus in accordance with various embodiments of the present disclosure.

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Advantages and features of the present disclosure, and methods of accomplishing the same will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the technical spirit of the present disclosure is not limited to the following embodiments, and may be implemented in various forms, and only the following embodiments make the technical spirit of the present disclosure complete and in the technical field to which the present disclosure belongs. It is provided to fully inform those skilled in the art the scope of the present disclosure, and the technical spirit of the present disclosure is only defined by the scope of the claims.

In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even though they are shown in different drawings. In addition, in describing the present disclosure, when it is determined that a detailed description of a related known configuration or function may obscure the subject matter of the present disclosure, the detailed description thereof will be omitted.

Unless otherwise defined, all terms used in the present specification (including technical and scientific terms) may be used in a sense that can be commonly understood by those skilled in the art to which the present disclosure belongs. In addition, the terms defined in the commonly used dictionaries are not ideally or excessively interpreted unless they are specifically defined clearly. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. In this specification, the singular also includes the plural unless specifically stated otherwise in the phrase.

In addition, in describing the components of the present disclosure, terms such as first, second, A, B, (a), and (b) may be used. These terms are only for distinguishing the components from other components, and the nature, order or order of the components are not limited by the terms. If a component is described as being "connected", "coupled" or "connected" to another component, that component may be directly connected to or connected to that other component, but there may be another configuration between each component. It is to be understood that the elements may be "connected", "coupled" or "connected".

As used herein, “comprises” and / or “comprising” refers to a component, step, operation and / or element that is mentioned in the presence of one or more other components, steps, operations and / or elements. Or does not exclude additions.

Prior to the description herein, some terms used herein will be clarified.

In the present specification, a task refers to a task to be solved through machine learning or a task to be performed through machine learning. For example, when face recognition, facial expression recognition, gender classification, pose classification, and the like are performed from face data, each of face recognition, facial expression recognition, gender classification, and pose classification may correspond to an individual domain. As another example, when an abnormality is recognized, classified, or predicted from medical image data, each of the abnormality recognition, the abnormality classification, and the abnormality prediction may correspond to an individual task. And the task may be referred to as the destination task.

In the present specification, multi-domain learning means performing learning on different domains using one model.

In this specification, a neural network is a term encompassing all kinds of machine learning models designed to mimic neural structures. For example, the neural network may include all kinds of neural network based models such as an artificial neural network (ANN), a convolutional neural network (CNN), and the like.

Instructions herein refer to a series of computer readable instructions, bound by function, that refer to components of a computer program and to be executed by a processor.

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

1 is a diagram illustrating a machine learning apparatus 10 and a learning environment according to some embodiments of the present disclosure.

As shown in FIG. 1, the machine learning device 10 is a computing device that performs multi-domain ensemble learning for neural networks. Thus, in various embodiments of the present disclosure, machine learning apparatus 10 may be referred to as multi-domain ensemble learning apparatus 10. Hereinafter, for convenience of description, the machine learning apparatus 10 will be abbreviated as the learning apparatus 10.

The computing device may be a laptop, a desktop, a laptop, a server, or the like, but is not limited thereto and may include any type of device having a computing function. An example of the computing device is shown in FIG. 8.

Although FIG. 1 illustrates that the learning device 10 is implemented as one computing device, the first function of the learning device 10 is implemented in the first computing device in an actual physical environment, and the learning device 10 is illustrated in FIG. The second function of may be implemented in the second computing device. In addition, the learning device 10 may be composed of a plurality of computing devices, and the plurality of computing devices may be implemented by dividing the first function and the second function.

Datasets 12 and 13 are training datasets given correct label information. Specifically, the first dataset 12 is a dataset composed of a plurality of training samples belonging to the first domain, and the second dataset 13 is a plurality of training samples belonging to a second domain different from the first domain. It may be a configured dataset. Here, the training sample may mean a unit of data for learning and may be various data. For example, the training sample may be one image and may be various data other than the image according to the learning object or task.

According to various embodiments of the present disclosure, the learning apparatus 10 may perform multi-domain ensemble learning on a neural network using data sets 12 and 13 belonging to a plurality of domains. The neural network may include a plurality of output layers and at least one feature extraction layer shared by the plurality of output layers. In this case, the plurality of output layers may include a first type of layer specialized for a specific domain and a second type of layer associated with the plurality of domains. In addition, the first type of layer may be learned using a dataset belonging to the specific domain, and the second type of layer may be learned using a dataset belonging to each of the plurality of domains.

For example, the neural network can be configured as illustrated in FIG. 2, which illustrates a neural network that can be used for multi-domain ensemble learning for two domains 12, 13. As illustrated in FIG. 2, the neural network may include a feature extraction layer 14 shared with three output layers 15, 16, 17. In this case, the first output layer 15 may be a layer specialized for the first domain, and the second output layer 16 may be a layer specialized for the second domain. In addition, the third output layer 17 may be a layer associated with the first domain and the second domain.

Accordingly, the first output layer 15 is trained based on the first dataset 12 belonging to the first domain, and the second output layer 16 is a second dataset belonging to the second domain. 13) can be learned. In addition, the third output layer 17 may be trained using the data sets 12 and 13 belonging to each domain. Since the feature extraction layer 14 needs to perform an operation of extracting a feature with respect to the multi-domain, the feature extraction layer 14 may be trained based on both the first data set 12 and the second data set 13. A detailed description of a method of performing multi-domain ensemble learning on the neural network will be described later with reference to FIGS. 3 to 8.

So far, the operation and learning environment of the learning apparatus 10 according to some embodiments of the present disclosure have been described with reference to FIGS. 1 and 2. Hereinafter, methods according to various embodiments of the present disclosure will be described.

Each step of the methods may be performed by a computing device. In other words, each step of the methods may be implemented with one or more instructions executed by a processor of the computing device. All steps included in the methods may be performed by one physical computing device, but the first steps of the methods are performed by the first computing device and the second steps of the methods are performed on the second computing device. It may also be performed by. In the following description, it is assumed that each step of the methods is performed by the learning apparatus 10. Thus, it will be appreciated that where the subject of each operation is omitted in the description of the methods, it may be performed by the apparatus illustrated above. In addition, the methods to be described below may change the execution order of each operation within a range in which the execution order may be logically changed as necessary.

3 is a flowchart of a multi-domain ensemble learning method according to some embodiments of the present disclosure. However, this is only a preferred embodiment for achieving the object of the present disclosure, of course, some steps may be added or deleted as necessary. In addition, in the following description, it will be described on the assumption that the number of domains is two, but this is only for convenience of understanding, and the number of domains may vary depending on the embodiment. In addition, it is assumed that the learning target neural network includes a shared feature extraction layer and three output layers as illustrated in FIG. 2.

Referring to FIG. 3, in step S100, a data set for multi-domain ensemble learning is obtained. For example, the dataset may include a first dataset belonging to the first domain and a second dataset belonging to the second domain. Hereinafter, the first data set refers to a data set belonging to the first domain, and the second data set refers to a data set belonging to the second domain.

In some embodiments, the first dataset may consist of an image generated by a first photographing method, and the second dataset may consist of an image generated by a second photographing method. That is, the multi-domain may be defined based on the photographing method. For example, the first imaging method may be a full-field digital mammography (FFDM) method, and the second imaging method may be a digital breast tomosynthesis (DBT) method. In this case, the neural network may be trained to perform a prediction task (e.g. cancer diagnosis) on the FFDM image and the DBT image.

In some embodiments, the first dataset may comprise more data (ie, training samples) than the second dataset. In this case, a process of increasing the number of samples of the second domain through over-sampling may be further performed.

In some embodiments, the first dataset may include data of a different type (or format) from the second dataset. For example, the first dataset may consist of a 2D image (e.g. FFDM image), and the second dataset may consist of a 3D image (e.g. DBT image). In another example, the first dataset may consist of a single channel or single layer image (e.g. FFDM image), and the second dataset may consist of a multichannel or multi-layer image (e.g. DBT image). In this case, before inputting data into the neural network, a process of adjusting (or converting) the shape of the input data according to the input format of the neural network may be further performed. In this regard will be described in detail with reference to FIG.

4 assumes that the input form of the neural network is implemented according to the form of the first dataset.

Referring to FIG. 4, in step S101, it is determined whether the first data set and the second data set include different types of data. If the shape of the data is different, in step S102, each data included in the second dataset may be adjusted (or converted) to have the same input form as the first dataset. The specific adjustment process may vary depending on the embodiment.

In some embodiments, the first dataset may be FFDM data and the second dataset may be DBT data. The DBT data may be multi-channel or 3D input, but the neural network may be implemented to receive a single channel image as the FFDM data. In this case, the single channel image may be extracted (or sampled) from the multi channel image, and the extracted single channel image may be input to the neural network.

In some embodiments, the first dataset may include a single layer image, and the second dataset may include a multi-layer image. In addition, the neural network may be implemented to receive a single layer image as an input. In this case, a single layer image may be extracted (or sampled) from the multilayer image, and the extracted single channel image may be input to the neural network.

In step S103, it is determined whether the adjusted data satisfy the condition. The condition refers to a criterion for determining whether the adjusted data is suitable as sample data for learning. For example, the sharpness, the ratio with the first data set, whether the specific color is included, the size of the data, and the like may be conditions. The condition may be set through user input or may be set automatically according to the type of task.

In some embodiments, it may be determined whether the number of adjusted second datasets corresponds to a ratio with the number of sample data included in the first dataset included in the first domain. At this time, the ratio between the different domains can be tailored so as not to be biased. For example, if the ratio between the first domain and the second domain is significantly different, such as 10: 1, the second domain may be adjusted to a desired ratio by over sampling the second domain. The ratio may be a preset value and a value that varies depending on the type of task and domain.

This will be described with reference to FIG. 3 again. In step S200, it is determined to which domain the dataset belongs. If it is determined that the dataset belongs to the first domain, step S300 may be performed, and otherwise, step S400 may be performed. Step S200 may be omitted according to an embodiment.

In operation S300, a first output layer is learned using a first dataset belonging to the first domain. The first output layer may mean a layer specialized in the first domain among three output layers. Also, a feature extraction layer shared by a plurality of output layers may be learned using the first dataset. The detailed learning process of this step S300 is shown in FIG.

Referring to FIG. 5, in operation S301, a predicted value output from the first output layer is obtained. In step S302, an error with respect to the obtained prediction value is calculated. In step S303, the error is propagated back to update the weights of the first output layer and the feature extraction layer. That is, the error of the obtained prediction value is propagated again in the reverse direction and the weight is updated in such a manner that the error is minimized.

Referring back to FIG. 3, in operation S400, a second output layer is learned using a second dataset belonging to a second domain. The second output layer may mean a layer specialized in the second domain among three output layers. Also, a feature extraction layer shared by a plurality of output layers may be learned using the second dataset. Since the detailed learning process of this step S400 is similar to the above-described step S300, further description thereof will be omitted.

In operation S500, a third output layer is learned using the first data set and the second data set. The third output layer is a layer associated with the first domain and the second domain, and may be used to improve prediction performance in each domain when predicting a label. In this regard, it will be described in detail later with reference to FIG.

FIG. 3 shows that step S500 is performed after steps S300 and S400, but this is only for convenience of understanding, and some processes (ie, a learning process associated with the first domain) of step S500 are performed in step S300. And some other processes (ie, a learning process associated with the second domain) may be performed with step S400.

In operation S500, the feature extraction layer may also be learned using the second dataset.

So far, the multi-domain ensemble learning method according to some embodiments of the present disclosure has been described with reference to FIGS. 3 to 5. According to the above-described method, the neural network to be taught further includes an additional output layer (eg, a third output layer) in addition to the layers specific to each domain (eg, a first output layer and a second output layer), and a plurality of additional output layers. Learning about the domain takes place. In addition, the additional output layer can be used to assist the prediction process in each domain, thereby improving the prediction performance in each domain.

In particular, the additional output layer may be used to improve the prediction performance in a domain without much data, and the higher the similarity between the two domains, the more the prediction performance may be improved.

In the above description, it has been described assuming that the learning target neural network is based on a convolutional neural network. However, since the technical idea of the present disclosure may be equally applied to other types of neural networks, such as artificial neural networks, the technical scope of the present disclosure is specific to neural networks. It is not limited to.

Hereinafter, a method of label prediction using a learned neural network will be described with reference to FIGS. 6 to 7.

6 is a flowchart of a label prediction method according to some embodiments of the present disclosure.

Referring to FIG. 6, in step S600, prediction data that is not given a label is obtained.

In step S700, it is determined whether the obtained prediction data belongs to the first domain. If it is determined that the prediction data belongs to the first domain, step S800 is performed. In the opposite case, step S900 is performed. Step S700 may be omitted according to an embodiment.

In step S800, label prediction for the first domain is performed. The detailed process of this step S800 is shown in FIG. It demonstrates with reference to FIG.

In operation S801, a first prediction value is obtained through the first output layer of the neural network. Specifically, the prediction data is input to the neural network, and a value output from the first output layer among a plurality of output layers constituting the neural network is obtained as the first prediction value. As described above, the first output layer means a layer specialized for the first domain.

In operation S802, a second prediction value is obtained through the third output layer of the neural network. Specifically, the prediction data is input to the neural network, and a value output from the third output layer of the plurality of output layers constituting the neural network is obtained as the second prediction value. As described above, the third output layer refers to a layer on which learning of the first domain and the second domain is performed.

In step S803, a label is predicted based on the obtained first prediction value and the second prediction value. In this case, a weight may be assigned to each of the first predicted value and the second predicted value, and prediction may be performed on the label in consideration of the assigned weight (e.g. sum of weights, etc.). Weights for each prediction value may be equally given to each other, and may be differently applied according to various criteria such as the type of task, the similarity between domains, the degree to which each output layer is learned, or the type of domain.

Referring back to FIG. 6, in step S900, a label of prediction data belonging to the second domain is predicted. In this case, a prediction value may be obtained in each of the second and third output layers, and a label may be predicted based on the prediction values of the two layers.

So far, the label prediction method according to some embodiments of the present disclosure has been described with reference to FIGS. 6 and 7. According to the above-described method, a label for the specific domain may be predicted by using an output layer specialized for a specific domain and a prediction value for an additional output layer that has learned all of the plurality of domains together. Accordingly, prediction performance in the specific domain may be improved. In particular, even if the specific domain does not have much data, the additional output layer may improve prediction performance in the specific domain.

Hereinafter, an exemplary computing device 100 that can implement an apparatus (e. G. Learning device 10) according to various embodiments of the present disclosure will be described.

8 is a hardware diagram illustrating an example computing device 100 that may implement an apparatus (e.g., learning apparatus 10) according to various embodiments of the present disclosure.

As shown in FIG. 8, the computing device 100 may include one or more processors 110, a bus 150, a communication interface 170, and a memory for loading a computer program executed by the processor 110. 130 and a storage 190 for storing the computer program 191. However, FIG. 8 illustrates only the components related to the embodiment of the present disclosure. Accordingly, those skilled in the art may recognize that other general purpose components may be further included in addition to the components illustrated in FIG. 8.

The processor 110 controls the overall operation of each component of the computing device 100. The processor 110 is configured to include a central processing unit (CPU), a micro processor unit (MPU), a micro controller unit (MCU), a graphics processing unit (GPU), or any type of processor well known in the art. Can be. In addition, the processor 110 may perform an operation on at least one application or program for executing a method / operation according to embodiments of the present disclosure. Computing device 100 may have one or more processors.

The memory 130 stores various data, commands, and / or information. Memory 130 may load one or more programs 191 from storage 190 to execute methods / operations in accordance with various embodiments of the present disclosure. The memory 130 may be implemented as a volatile memory such as a RAM, but the technical scope of the present disclosure is not limited thereto.

The bus 150 provides a communication function between components of the computing device 100. The bus 150 may be implemented as various types of buses such as an address bus, a data bus, and a control bus.

The communication interface 170 supports wired and wireless Internet communication of the computing device 100. In addition, the communication interface 170 may support various communication methods other than Internet communication. To this end, the communication interface 170 may comprise a communication module well known in the art of the present disclosure. In some cases, the communication interface 170 may be omitted.

The storage 190 may non-temporarily store the one or more computer programs 191, various data (e.g., a learning data set), a machine learning model, and the like. The storage 190 is well known in the art for non-volatile memory, hard disks, removable disks, or the like to which the present disclosure belongs, such as Read Only Memory (ROM), Eraseable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), Flash Memory, and the like. It may comprise any known type of computer readable recording medium.

Computer program 191 may include one or more instructions that, when loaded into memory 130, cause processor 110 to perform methods / operations in accordance with various embodiments of the present disclosure. That is, the processor 110 may perform methods / operations according to various embodiments of the present disclosure by executing the one or more instructions.

For example, the computer program 191 may train a first output layer of a neural network including a plurality of output layers by using a first dataset belonging to a first domain, and generate a second dataset belonging to a second domain. Instructions to train a second output layer of the plurality of output layers by using; and to train a third output layer of the plurality of output layers by using the first data set and the second data set. Can include them. In addition, the computer program 191 may further include instructions for performing an operation of predicting a label of prediction data belonging to the first domain by using the learned neural network. In this case, the multi-domain ensemble learning apparatus e.g. 10 may be implemented through the computing device 100 in accordance with some embodiments of the present disclosure.

An example computing device 100 that may implement an apparatus in accordance with various embodiments of the present disclosure has been described above with reference to FIG. 8.

So far, various embodiments of the present disclosure and effects according to the embodiments have been described with reference to FIGS. 1 to 8. Effects according to the technical spirit of the present disclosure are not limited to the above-mentioned effects, other effects not mentioned will be clearly understood by those skilled in the art from the following description.

The technical idea of the present disclosure described with reference to FIGS. 1 to 8 may be implemented as computer readable codes on a computer readable medium. The computer-readable recording medium may be, for example, a removable recording medium (CD, DVD, Blu-ray disc, USB storage device, removable hard disk) or a fixed recording medium (ROM, RAM, computer equipped hard disk). Can be. The computer program recorded on the computer-readable recording medium may be transmitted to another computing device and installed in the other computing device through a network such as the Internet, thereby being used in the other computing device.

In the above description, it is described that all the elements constituting the embodiments of the present disclosure are combined or operated as one, but the technical spirit of the present disclosure is not necessarily limited to these embodiments. That is, within the scope of the present disclosure, all of the components may be selectively operated in one or more combinations.

Although the operations are shown in a specific order in the drawings, it should not be understood that the operations must be performed in the specific order or sequential order shown or that all the illustrated operations must be executed to achieve the desired results. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of the various configurations in the embodiments described above should not be understood as such separation being necessary, and the described program components and systems may generally be integrated together into a single software product or packaged into multiple software products. Should be understood.

While the embodiments of the present disclosure have been described with reference to the accompanying drawings, a person of ordinary skill in the art may implement the present disclosure in other specific forms without changing the technical spirit or essential features. I can understand that there is. Therefore, it is to be understood that the embodiments described above are exemplary in all respects and not restrictive. The scope of protection of the present disclosure should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto shall be interpreted as being included in the scope of the technical idea defined by the present disclosure.

Claims (20)

A method for performing multi-domain ensemble learning on a neural network including a plurality of output layers in a computing device, the method comprising:
Learning a first output layer and a shared feature extraction layer of the plurality of output layers using a first dataset belonging to a first domain;
Learning a second output layer and the shared feature extraction layer of the plurality of output layers using a second dataset belonging to a second domain; And
Learning a third output layer and the shared feature extraction layer of the plurality of output layers by using the first data set and the second data set;
Multi domain ensemble learning method. According to claim 1,
The first data set includes more data than the second data set,
Multi domain ensemble learning method. According to claim 1,
The first dataset comprises a 2D image,
Wherein the second dataset comprises a 3D image,
Multi domain ensemble learning method. The method of claim 3, wherein
The first data set includes a full-field digital mammography (FFDM) image,
The second dataset includes a digital breast tomosynthesis (DBT) image,
Multi domain ensemble learning method. According to claim 1,
The first dataset comprises a single layer image,
The second dataset comprises a multilayer image,
Multi domain ensemble learning method. The method of claim 5,
Learning the second output layer may include:
Extracting a first layer image from the multi-layer image and learning the second output layer using the first layer image; And
Extracting a second layer image from the multilayer image and learning the second output layer using the second layer image;
Multi domain ensemble learning method. According to claim 1,
Wherein the shared feature extraction layer is shared by the first output layer, the second output layer, and the third output layer.
Multi domain ensemble learning method. According to claim 1,
Predicting a label of prediction data belonging to the first domain by using the neural network;
The predicting step,
Predicting the label based on a first prediction value output from the first output layer and a second prediction value output from the third output layer.
Multi domain ensemble learning method. The method of claim 8,
Predicting the label based on a first prediction value output from the first output layer and a second prediction value output from the third output layer,
Weighting each of the first prediction value and the second prediction value;
Synthesizing the first prediction value and the second prediction value by reflecting the assigned weight value; And
Predicting the label based on the synthesized prediction value;
Multi domain ensemble learning method. Memory for storing one or more instructions and
By executing the stored one or more instructions,
Train a first output layer and a shared feature extraction layer among a plurality of output layers included in the neural network by using a first dataset belonging to the first domain,
Train a second output layer and the shared feature extraction layer among the plurality of output layers using a second dataset belonging to a second domain,
And a processor configured to learn a third output layer and the shared feature extraction layer of the plurality of output layers using the first data set and the second data set.
Multi Domain Ensemble Learning Device. The method of claim 10,
The first data set includes more data than the second data set,
Multi Domain Ensemble Learning Device. The method of claim 10,
The first dataset comprises a 2D image,
Wherein the second dataset comprises a 3D image,
Multi Domain Ensemble Learning Device. The method of claim 12,
The first data set includes a full-field digital mammography (FFDM) image,
The second dataset includes a digital breast tomosynthesis (DBT) image,
Multi Domain Ensemble Learning Device. The method of claim 10,
The first dataset comprises a single layer image,
The second dataset comprises a multilayer image,
Multi Domain Ensemble Learning Device. The method of claim 14,
The processor,
Extracting a first layer image from the multi-layer image, learning the second output layer using the first layer image,
Extracting a second layer image from the multi-layer image, and learning the second output layer using the second layer image;
Multi Domain Ensemble Learning Device. The method of claim 10,
Wherein the shared feature extraction layer is shared by the first output layer, the second output layer, and the third output layer.
Multi Domain Ensemble Learning Device. The method of claim 10,
The processor,
The neural network is further used to predict a label of prediction data belonging to the first domain, wherein the label is based on a first prediction value output from the first output layer and a second prediction value output from the third output layer. Predicted,
Multi Domain Ensemble Learning Device. The method of claim 17,
The processor,
Integrating the first prediction value and the second prediction value by reflecting weights assigned to each of the first prediction value and the second prediction value, and predicting the label based on the combined prediction value,
Multi Domain Ensemble Learning Device. In conjunction with computing devices,
Training a first output layer and a shared feature extraction layer among a plurality of output layers included in the neural network by using a first dataset belonging to the first domain;
Learning a second output layer and the shared feature extraction layer of the plurality of output layers using a second dataset belonging to a second domain; And
Stored in a computer-readable recording medium for executing the step of learning a third output layer and the shared feature extraction layer of the plurality of output layers using the first data set and the second data set;
Computer programs. The method of claim 19,
Predicting a label of prediction data belonging to the first domain by using the neural network;
The predicting step,
Predicting the label based on a first prediction value output from the first output layer and a second prediction value output from the third output layer.
Computer programs.

Images