KR20210088421A - Machine learning method for incremental learning and computing device for performing the same - Google Patents

Abstract

A machine learning method for incremental learning of the present invention constructs a model using training data, and the constructed model is incrementally updated by using only new weights generated based on the new training data. The machine learning method comprises: an encoding step; a generating step; a determining step; a constructing step; an encoding step; a normalizing step; and an updating step.

Description

A machine learning method for progressive learning and a computing device for performing the same {MACHINE LEARNING METHOD FOR INCREMENTAL LEARNING AND COMPUTING DEVICE FOR PERFORMING THE SAME}

The present invention relates to machine learning, and more particularly, to machine learning related to incremental learning.

In order to improve the adaptability and reliability of a supervised machine learning model widely used in the field of artificial intelligence (AI), various studies on incremental learning are being attempted. Progressive learning allows the model to be more adaptable to continuously changing environments.

Artificial Neural Network (ANN)-based machine learning models such as Deep Neural Network (DNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), etc. It is a well-known fact that there is a problem with Catastrophic Forgetting (CF), which limits implementation of incremental or continuous learning, and that it is difficult to explain the model or the results because the internal structure of the model is very complex. .

In the ANN-based machine learning model, when new training data is input, the CF problem of escaping from an optimized state (previously learned state) for the entire previous training data and forgetting previously learned content may occur. It is difficult to incrementally scale the model (either incremental updates or incremental performance improvements).

Although various methods are being studied to improve the CF problem, most of the studies involve degradation of model performance, so a method to effectively improve the CF problem is still insufficient.

For non-image multivariate numeric data (multivariate numeric data or multivariate numeric heterogeneous data), it is an algorithm that outperforms ANN-based algorithms. Gradient Boosting, GB). However, this technique also does not easily provide gradual learning because it optimizes the entire training data in building the model.

An object of the present invention for solving the above problems is to provide a machine learning method and a computing device for easily performing incremental learning without degrading the performance of the model.

The above and other objects, advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings.

A machine learning method for progressive learning according to an aspect of the present invention for achieving the above object comprises: encoding training data labeled with a plurality of class labels; Constructing the variables included in the encoded training data into nodes, and connecting adjacent nodes among the nodes with edges having a weight indicating the connection strength to generate a plurality of variable networks classified by the plurality of class labels step; determining variable networks selected according to performance from among the generated plurality of variable networks as main variable networks; constructing a model by combining the determined main variable networks; encoding new training data; calculating a new weight by using the instance of the encoded new training data and then normalizing the calculated new weight; and gradually updating the built model by updating the weight of each of the determined main variable networks based on the new normalized weight.

A computing device for executing a machine learning method for gradual learning according to another aspect of the present invention comprises: a processor; a repository for storing training data labeled with multiple class labels and new training data; and a machine learning module for building a model using the training data labeled with the plurality of class labels under the control of the processor, wherein the machine learning module comprises: the training data labeled with the plurality of class labels and the new an encoder that encodes the training data; Constructing the variables included in the encoded training data into nodes, and connecting adjacent nodes among the nodes with edges having a weight indicating the connection strength to generate a plurality of variable networks classified by the plurality of class labels variable network generator; Determining variable networks selected according to performance from among the generated plurality of variable networks as main variable networks, calculating a new weight using an instance of the encoded new training data, and normalizing the calculated new weight variable network determiner; a model builder for building a model by combining the determined main variable networks; and based on the new normalized weight and an update unit for progressively updating the built model by updating the weight of each of the determined main variable networks.

According to the present invention, when new training data is input, the previously built model is maintained and the previously built model is learned using only the weights generated based on the new training data, so that the structure of the previously built model is improved. Progressive learning is easy because the model can be updated without changes.

1 is a flowchart illustrating a machine learning method for gradual learning according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining a variable permutation selected in the step of selecting feature sequences shown in FIG. 1 .
FIG. 3 is a diagram for schematically explaining the model building step S400 shown in FIG. 1 .
FIG. 4 is a view for explaining an ensemble configuration of each sub-model shown in FIG. 1 .
5 is a block diagram of a computing device implemented to perform a machine learning method for gradual learning according to an embodiment of the present invention.

Specific structural or functional descriptions of the embodiments according to the concept of the present invention disclosed in this specification are only exemplified for the purpose of explaining the embodiments according to the concept of the present invention, and the embodiments according to the concept of the present invention are It may be implemented in various forms and is not limited to the embodiments described herein.

Since the embodiments according to the concept of the present invention may have various changes and may have various forms, the embodiments will be illustrated in the drawings and described in detail herein. However, this is not intended to limit the embodiments according to the concept of the present invention to specific disclosed forms, and includes changes, equivalents, or substitutes included in the spirit and scope of the present invention.

The terminology used herein is used only to describe specific embodiments, and is not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "have" are intended to designate that the described feature, number, step, operation, component, part, or combination thereof exists, and includes one or more other features or numbers, It should be understood that the possibility of the presence or addition of steps, operations, components, parts or combinations thereof is not precluded in advance.

The present invention relates to a supervised learning algorithm that facilitates gradual learning that conventional machine learning cannot efficiently implement. The present invention is a supervised learning method for predicting a label of a target variable with respect to data consisting of a plurality of variables (variables or features) and a target variable, a Significant Feature Network (SFN) , constructs a learning model using the interrelationship of values included in the variable combination from the learning data, and uses the constructed learning model for classification prediction on new data.

In the present invention, when an existing model is additionally learned using a new data set, incremental learning is very easy because a new model covering the new data set can be built by adding a gradual change to the existing model.

Hereinafter, a machine learning method for gradual learning according to an embodiment of the present invention will be described in detail with reference to the drawings. And the following embodiment relates to supervised learning for the purpose of classification. However, the present invention is not limited thereto, and those skilled in the art will be able to fully understand from the following description that the present invention can also be applied to supervised learning for the purpose of regression.

1 is a flowchart illustrating a machine learning method for gradual learning according to an embodiment of the present invention.

The machine learning method for gradual learning according to an embodiment of the present invention can be largely divided into a step of performing learning and prediction on a single data set and a step of performing gradual learning on an additional data set.

First, "a step of performing learning and prediction on a single data set" will be described, and then "a step of performing gradual learning on an additional data set" will be described. In this specification, a single data set is

Steps to train and make predictions on a single data set

1 , the step of performing training and prediction on a single data set includes preparing training data sets 101 and 102 and test data sets 110 and 111 ( Preparing) (S100), encoding (encoding) step (S200), discovering (discovering) major variable networks (Significant Feature Networks: SFN) (S300), building a model (S400) and It includes a prediction (prediction) step (S500).

A. Preparation of training data set and validation data set (S100)

The training data set 101 includes a plurality of training data labeled with a number of class labels to build a model 400: 400_1, 400_2, ..., 400_N.

Each training data consists of multi-dimensional feature variables and a class label as a target feature or target variable. Each feature (feature or variable) can be continuous or discrete numeric or character values.

The validation data set 110 has the same configuration as the training data set 101 , but is different from the training data sets 101 and 102 in that it is used for the purpose of verifying the prediction performance of a previously built model.

The training data set 101 and the validation data set 110 can be divided into before and after encoding, and the training data set 101 and the validation data set 110 before encoding are raw training data sets and raw data sets. They may be referred to as raw test data sets.

B. Encoding step (S200)

In the encoding step (S200), the encoder 200 encodes the training data set 101 and the verification data set 110 is performed. The encoding may be processing the training data set 101 into data suitable for training (or learning) of the model 400 , and processing the validation data set 110 into data suitable for verification of the model 400 .

The encoding (S200) converts the value of a certain variable into a discrete value, a discontinuous value, or a categorical value when the value of a variable is continuous, or converts a value consisting of text to an appropriate value. It can be converted to a number.

Converting a continuous value of a variable into a discrete or categorical value, or converting a character value into a numeric value can be converted according to an encoding rule defined (or programmed) in circumstances. Encoding rules may be static or dynamic in the entire process of learning and prediction.

Also, the encoding ( S200 ) may be to reset the interval of discrete or categorical values, or to convert the input value into another value. Here, resetting the interval of discrete or categorical values may be, for example, resetting the values divided by 10 steps to 5 steps, and converting the input value to another value is, for example, , -2,-1,0,1,2 may be converted to 1,2,3,4,5.

C. Step of discovering (discovering) the main variable network (SFN) (S300)

The step of exploring the SFN ( S300 ) is a key feature network that is a core component of the model 400 using the training data set 201 (or the encoded training data set) encoded by the encoder 200 . , SFN) may be discovered. Here, the discovery of the SFN may be detecting, extracting, or calculating the SFN using the encoded training data set 201 .

Specifically, the step of searching for the SFN (S300) is, for example, a step (S301) of generating a variable permutation (feature sequences), a step (S302) of a formation of a node and an edge (node and edge), Weight calculation step (S303), weight normalization step (S304), variable network evaluation (assessing feature network) step (S305), variable network (feature network) ranking step (S306) and SFN selection ( selecting significant feature networks (S307).

The SFN may be obtained (discovered, detected, extracted, or calculated) through the process of iteration of the steps S301 to S306, and in the SFN selection step S307, in the network ranking step S306 A process of selecting high-ranked specific feature sequences as SFN is performed. A model is constructed using the selected SFN. Hereinafter, each step for deriving the SFN will be described in detail.

C-1. Generating feature sequences (S301)

FIG. 2 is a diagram illustrating an example of a variable permutation generated by the step S301 of generating a feature sequence shown in FIG. 1 .

Referring to FIG. 2 , variable permutation is selecting two or more variables (or two or more generated variables) from a multivariate encoded training data set 201 and arranging them in a specific order.

Variable permutation (feature sequence) is, for example, if N variables from among all variables are selected and listed in a specific order, as shown in FIG. 2 , a specific permutation consisting of _{f 1} , f ₂ , f ₃ , ??, f _N this can be made

Generating a specific variable permutation is a method of selecting a variable (feature) without transformation of the variable, and a method of generating a new variable based on the variables included in the encoded training data set 201 (feature generation) method can be divided into

Variable selection methods for generating variable permutations, for example, a random selection method, a method that considers all combinations, a method obtained from other machine learning methods, and mutual information in information theory There may be various methods such as a method of using

As a variable generation method for generating variable permutations, for example, using a variable extraction method based on deep learning such as linear discriminant analysis (LDA), Principal Component Analysis (PCA), and Autoencoder. There may be various methods, such as a method.

C-2 node and edge (node and edge) configuration (formation) step (S302)

When a specific variable permutation is selected by the previous step S301, nodes and edges are defined, and a feature network can be constructed through this.

Node _{(f 11, f 12, ...} f 1i, f 21, f 22, ... f N1, f N2, f NP, ??) is, also, the respective parameters as shown in ₂ (f _1, f 2, f ₃ , ??, f _N ) are defined as encoded values, and edges (w ₁₁ , w ₁₂ , w ₁₃ , w _1α, w _{21 ,} w _22, w _23, w _2β, ??) are adjacent nodes The connections between them are defined. Here, the connection of nodes within the same variable is not considered. For example, the variable f ₂ contains f ₂₁ , f ₂₂ , ?? There are nodes such as f _{2j and they are connected to the nodes of the adjacent variables f 1} and f ₃ by edges (or connecting lines representing weights). Through this connection of nodes and edges, a feature network for the selected variable permutation is constructed.

C-3 weight calculation step (S303)

Edges connecting nodes have a specific value, and this specific value is defined as a weight indicating the connection strength of the nodes. The weights may be obtained from the encoded training data 201 . When an instance of the encoded training data 201 is input, a weight of an edge connecting nodes activated by the instance is calculated. Here, the instance means each case (example) or sample (sample) constituting the data when the data required for the machine learning model for learning or inference (prediction), etc. is given. Accordingly, the instance may be called a training example or a training sample constituting the training data.

The weight may be calculated according to a predefined weight calculation rule. One of the weight calculation methods is to divide the network by class and update the weight.

For example, given training data with 3 class labels from 1 to 3, 3 variable networks are created with the same variable permutation, and the training data with class 1 label gives the weight of network 1, 2 The training data with the class label No. 2 is used to calculate the weight of the second network, and the training data with the class label No. 3 is used to calculate the weight of the third network. This means that for one variable permutation, variable networks with different weights are created according to the class label.

C-4 weight normalization step (S304)

When an edge weight is calculated by a plurality of instances included in the encoded training data set 1 201, a normalization process is performed on the calculated weight.

The normalization process may be performed according to a predefined weight normalization rule. Here, the weight normalization rule may be, for example, a rule that sets the sum of edges between two adjacent variables to be 1.

C-5 variable network evaluation (assessing feature network) step (S305)

The variable network evaluation step S305 is a step of calculating a network evaluation index indicating how good the variable network has in classifying a class by using the variable network and weight information generated by the above steps.

There are two main methods for evaluating a variable network.

The first method is a method of mathematically extracting a figure of merit from the characteristics contained in the weight information of the variable network. The second method may be to evaluate the performance of variable networks by calculating class discrimination accuracy using a plurality of variable networks, the normalized weight, and an instance labeled with a class label not used (or used) in weight calculation. have. Any method can arithmetically evaluate the variable network.

C-6 feature network ranking step (S306)

The variable network rank may be determined based on the evaluation index of the variable network arithmetically derived by the result of the previous step ( S305 ). In the initial execution, the first selected variable network ranks first, but if another variable network is selected in step S301 and the process up to S306 is iterated, the ranking may be changed. The rank is indicated by a subscript such as _{SFN 1} , SFN ₂ , SFN _{3 , ??.}

C-7 SFN selection (selecting significant feature networks) step (S307)

In the previous step S306, a predetermined number of variable networks ranked higher are selected. The selected variable networks are used for model building as principal variable networks (SFNs).

D. Step of building a model (S400)

FIG. 3 is a diagram for schematically explaining the model building step S400 shown in FIG. 1 , and FIG. 4 is a diagram for explaining an ensemble configuration of each sub-model shown in FIG. 1 .

Referring to FIG. 3 , the model building step ( S400 ) is a step of constructing a model using the main variable network selected through the previous step ( S307 ). The entire model 400 is composed of a plurality of sub-models that are distinguished for each class.

As shown in FIG. 1, the model constructed to distinguish N classes is configured to include N submodels 400_1, 400_2, ??, and 400_N. And each sub-model may be composed of an ensemble in which the main variable networks selected in the previous step ( S307 ) are combined, as shown in FIG. 4 .

The most basic way to construct an ensemble is to build all sub-models with the same principal variable networks. And when the weights are updated using the training data, the instances of the training data are used to calculate and update the weights of the main variable networks of the submodels corresponding to each class label as shown in FIG. 3 . After the training process, the generated sub-models are composed of the same main variable networks, but have different weight information.

E. Prediction step (S500)

In the prediction step S500, the instance of the verification data set 110 is input to all the sub-models 400_1, 400_2, ... 400_N included in the built model 400, and the sub-model having the largest weight score corresponds to that. It is selected as the prediction class of the instance.

The weight score of a specific sub-model for the instance of the verification data set 110 may be derived using the corresponding weight score of the main variable networks (SFNs) constituting the sub-model.

As shown in Figure 4, the weight score of submodel 1 is calculated as a linear combination of the weight scores of the main variable networks constituting the submodel, such as SFN1 (S311), SFN2 (S312), and SFN3 (S313). can

Let W(Di, SFN _j ) be the weight score of _{SFN j} with respect to the i-th instance D _{i of the verification data set 110 .} At this time, the weight score W ₁ (D _i ) of submodel 1 is calculated as follows.

c _j is a coefficient indicating the contribution according to the ranking of the SFN. For example, if c _j = 1, the weight score is calculated in equal proportion for all SFNs regardless of rank. _{In this case, the value of c j} may be set differently according to j (according to SFN) by giving a difference according to rank.

Steps to perform incremental training on additional data sets

One of the important features of the present invention is that it is possible to easily perform gradual learning on the newly added training data 102 . First, it is assumed that the model 400 is built based on the training data set 1 101 . Then, a new training data set 2 (102) is input to the encoder (200).

The encoder 200 performs encoding on the new training data set 2 102 , and generates an encoded training data set 2 102 .

Thereafter, instead of performing all a series of processes ( S301 to S307 ) included in the search step ( S300 ) of the SFN for the encoded training data set 2 ( 102 ), a weight scoring process ( S303 ) and weight normalization Progressive learning is performed in such a way that only the process S304 is sequentially performed, and the weights of the already built model 400 are updated with the normalized weights for the encoded training data set 2 102 .

Since such gradual learning is performed in a manner that maintains the previously built model and updates only the state variable called weight when new training data is input, gradual learning can be easily performed.

5 is a block diagram of a computing device implemented to perform a machine learning method for gradual learning according to an embodiment of the present invention.

Referring to FIG. 5 , the computing device 600 includes a storage 610 , a machine learning module 620 , a processor 630 , a memory 640 and a system bus ( 610 , 620 , 630 , 640 ) connecting them. 650) may be included.

Storage 610 stores training data (or training data sets) 102 and validation data (or validation data sets) 110 , 111 labeled with a number of class labels for building a model ( 400 in FIG. 1 ). And, it is a hardware device for storing new training data (or new training data set) 102 for progressively updating the model (400 in FIG. 1) through gradual learning.

Storage 610 is, for example, a computer-readable medium, for example, a hard disk, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs, DVDs, and floppy disks. ), such as magneto-optical media.

The machine learning module 620 builds a model (400 in FIG. 1) under the control or execution of the processor 630, and uses only the new weights generated based on the new training data 102, the built model ( It may be a hardware module or a software module that gradually updates (learns) 400 of FIG. 1 .

The machine learning module 620 may include a plurality of sub-modules classified according to functions, and the plurality of sub-modules are, for example, an encoder 621 and a Feature Networks (FN) generator 622 . , a significant feature networks (SFN) determiner 623 , a model builder 624 , and an update unit 625 .

The encoder 621 is configured to encode training data labeled with a plurality of class labels, and, for example, performs the process of step S200 described in FIG. 1 . The encoder 621 converts continuous values of variables included in the training data into discrete values or categorical values according to a predefined encoding rule.

The encoder 621 also encodes the new training data 102 to generate new weights based on the new training data 102 .

A variable network (FN) generator 622 configures the variables included in the encoded training data into nodes, and connects adjacent nodes among the nodes with edges having a weight indicating the strength of the connection, and the plurality of class labels As a configuration for generating a plurality of variable networks classified as , it may be a configuration for performing steps S301 and S302 described in FIG. 1 .

The variable network (FN) generator 622 generates a variable permutation by listing two or more variables included in the encoded training data in a specific order according to the execution of step S301 .

For example, the variable network (FN) generator 622 randomly selects two or more variables from the encoded training data, and then lists the two or more randomly selected variables in the specific order to generate the variable permutation.

As another example, the variable network (FN) generator 622 performs linear discriminant analysis (LDA), principal component analysis (PCA), and deep learning (deep learning) on two or more variables included in the encoded training data. learning)-based variable extraction method is used to convert new variables, and then the new variables are arranged in a specific order to generate the variable permutations.

When the variable permutation is generated, the variable network (FN) generator 622 configures the values included in each of the listed variables into nodes, and connects adjacent nodes according to the specific order among the configured nodes to the edge. Thus, for the generated variable permutations, a plurality of variable networks classified by the plurality of class labels are generated.

The principal variable network (SFN) determiner 623 determines variable networks selected according to performance from among the generated plurality of variable networks as principal variable networks.

For example, the principal variable network (SFN) determiner 623 calculates the weight of each of the plurality of variable networks using the instance of the encoded training data 201 ( S303 in FIG. 1 ), and the After processing the normalization of the calculated weights (S304 of FIG. 1), the process of evaluating the performance of each of the variable networks using the plurality of variable networks and the normalized weights (S305 of FIG. 1) is processed .

Additionally, the principal variable network (SFN) determiner 623 calculates a new weight using the instance of the new training data 202 encoded by the encoder 200 through step 303 of FIG. 1 , A process of normalizing the calculated new weight is processed through S304.

Then, the main variable network (SFN) determiner 623 determines the rank of the plurality of variable networks according to the evaluated performance (S306 of FIG. 1 ), and then ranks higher according to a preset number among the plurality of variable networks. A process (S307 of FIG. 1 ) of determining the variable networks ranked in , as the main variable networks is processed.

The normalization of the weights calculated by the principal variable network (SFN) determiner 623 may include, for example, that the plurality of class labels include a first class label and a second class label, and the plurality of variables When networks include a first variable network and a second variable network, calculating a weight of the first variable network using an instance of the training data labeled with the first class label, with the second class label Calculating a weight of the second variable network that is different from the weight of the first variable network using the labeled instances of the training data, and normalizing the weight of the first variable network and the weight of the second variable network may include.

The process of evaluating the performance of each variable network by the main variable network (SFN) determiner 623 is to calculate the class discrimination accuracy using the plurality of variable networks, the normalized weight, and an instance labeled with a class label. and evaluating the performance of each variable network based on the calculated class discrimination accuracy.

The model builder 624 processes the process of building the model 400 by combining the principal variable networks determined by the principal variable network (SFN) determiner 623 .

The update unit 625 handles the process of incrementally updating the model 400 built by the model builder 624 based on the new weights normalized by the principal variable network (SFN) determiner 623 . do.

For example, the update unit 625 may incrementally update the built model by adding the new normalized weight to the weight of each of the determined main variable networks.

The processor 630 is a configuration that controls and manages the operations of the storage 610, the machine learning module 620, and the memory 640 through the system bus 650, and includes at least one CPU, at least one GPU, or their It can be a combination.

Although the processor 630 and the machine learning module 620 are illustrated as separate components in FIG. 5 , they may be integrated into one. For example, a machine learning module 620 may be integrated into the processor 630 .

The memory 640 is a hardware device that temporarily or permanently stores intermediate data or result data processed by each configuration in the processor 630 or machine learning module 620, and stores program instructions such as ROM, RAM, flash memory, etc. and hardware devices specifically configured to perform

Examples of program instructions include not only machine codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The protection scope of the present invention is not limited to the description and expression of the embodiments explicitly described above. In addition, it is added once again that the protection scope of the present invention cannot be limited due to obvious changes or substitutions in the technical field to which the present invention pertains.

Claims (13)

A machine learning method performed by a computing device, comprising:
encoding training data labeled with a plurality of class labels;
Constructing the variables included in the encoded training data into nodes, and connecting adjacent nodes among the nodes with edges having a weight indicating the connection strength to generate a plurality of variable networks classified by the plurality of class labels step;
determining variable networks selected according to performance from among the generated plurality of variable networks as main variable networks;
constructing a model by combining the determined main variable networks;
encoding new training data;
calculating a new weight by using the instance of the encoded new training data and then normalizing the calculated new weight; and
gradually updating the built model by updating the weight of each of the determined main variable networks based on the new normalized weight.
A machine learning method for progressive learning, including In claim 1,
The encoding of the training data comprises:
Machine learning for progressive learning, which is a step of converting continuous values of variables included in the training data into discrete or categorical values according to a predefined encoding rule Way. In claim 1,
The step of generating a plurality of variable networks classified into the plurality of class labels comprises:
generating a variable permutation by listing two or more variables included in the encoded training data in a specific order; and
By configuring the values included in each of the listed variables into nodes, and connecting adjacent nodes according to the specific order among the configured nodes to the edge, the generated variable permutation is classified into the plurality of class labels creating multiple variable networks
A machine learning method for progressive learning, including In claim 3,
The step of generating the variable permutation comprises:
randomly selecting two or more variables from the encoded training data; and
generating the variable permutation by arranging the randomly selected two or more variables in the specific order
A machine learning method for progressive learning that includes a. In claim 3,
The step of generating the variable permutation comprises:
Two or more variables included in the encoded training data are converted into new variables using a variable extraction method based on linear discriminant analysis (LDA), Principal Component Analysis (PCA), and deep learning. converting; and
generating the variable permutation by arranging the new variables in a specific order
A machine learning method for progressive learning that includes a. In claim 1,
Determining the selected variable networks as main variable networks comprises:
calculating the weight of each of the plurality of variable networks using an instance of the training data, and normalizing the calculated weight;
evaluating the performance of each variable network using the plurality of variable networks and the normalized weight;
ranking the plurality of variable networks according to the evaluated performance; and
determining upper-ranked variable networks as the main variable networks according to a preset number among the plurality of variable networks;
A machine learning method for progressive learning that includes a. In claim 6,
The step of normalizing the calculated weight is,
when the plurality of class labels include a first class label and a second class label, and the plurality of variable networks include a first variable network and a second variable network,
calculating a weight of the first variable network using an instance of the training data labeled with the first class label;
calculating a weight of the second variable network that is different from the weight of the first variable network by using an instance of the training data labeled with the second class label; and
normalizing the weights of the first variable network and the weights of the second variable network;
A machine learning method for progressive learning that includes a. In claim 6,
Evaluating the performance of each of the variable networks comprises:
calculating class discrimination accuracy using the plurality of variable networks, the normalized weight, and an instance labeled with a class label; and
Evaluating the performance of each variable network based on the calculated class discrimination accuracy
A machine learning method for progressive learning that includes a. In claim 1,
The step of gradually updating the built model is,
and adding the new normalized weight to the weight of each of the determined main variable networks, and gradually updating the built model. A computing device for executing a machine learning method for progressive learning, comprising:
processor;
a repository for storing training data labeled with multiple class labels and new training data; and
A machine learning module for building a model using training data labeled with the plurality of class labels under the control of the processor;
The machine learning module is
an encoder for encoding training data labeled with a plurality of class labels and the new training data;
Constructing the variables included in the encoded training data into nodes, and connecting adjacent nodes among the nodes with edges having a weight indicating the connection strength to generate a plurality of variable networks classified by the plurality of class labels variable network generator;
Determining variable networks selected according to performance from among the generated plurality of variable networks as main variable networks, calculating a new weight using an instance of the encoded new training data, and normalizing the calculated new weight variable network determiner;
a model builder for building a model by combining the determined main variable networks; and
An update unit for progressively updating the built model by updating the weight of each of the determined main variable networks based on the new normalized weight
A computing device comprising a. In claim 10,
The variable network generator is
A first process of generating a variable permutation by listing two or more variables included in the encoded training data in a specific order, and configuring values included in each of the listed variables into nodes, and in the specific order among the configured nodes and connecting adjacent nodes to the edge according to a second process of generating a plurality of variable networks classified into the plurality of class labels with respect to the generated variable permutation. In claim 10,
The main variable network determiner,
A first processor that calculates the weight of each of the plurality of variable networks using an instance of the training data and normalizes the calculated weight, each variable network using the plurality of variable networks and the normalized weight A second process for evaluating the performance of the variables, a third process for determining a rank of the plurality of variable networks according to the evaluated performance, and a third process for determining the rank of the plurality of variable networks according to a preset number among the plurality of variable networks and processing a fourth process of determining with variable networks. In claim 10,
The update unit is
and adding the normalized new weight to the weight of each of the determined main variable networks to handle a process of gradually updating the built model.

Images