Selective learning method and device of media sample, electronic equipment and storage medium

Abstract

The invention provides a selective learning method and device of a media sample, electronic equipment and a storage medium. The method comprises the steps that a first media sample and a second media sample are determined from a media sample set, and the noise content of the first media sample is smaller than that of the second media sample; constructing a first cross entropy loss of the first media sample according to the real label and the predicted label of the first media sample; according to the prior label probability distribution and the predicted label probability distribution of the second media sample, constructing regularization loss of the second media sample; according to the first cross entropy loss and the regularization loss, constructing total loss of a sample; and training the model based on the total loss of the samples, so that the model selectively learns the first media sample and the second media sample.

Description

Technical Field

[0001] The present disclosure relates to the field of machine learning, and more specifically, to a selective learning method, device, electronic device, and storage medium for media samples. Background Art

[0002] In practical applications, the collected data is usually incomplete due to irregularities and imperfections in the data collection process, as well as privacy issues. This problem is particularly serious in areas such as medicine. These incomplete data sets often contain many complex missing features, which may hinder the effectiveness of machine learning models and seriously affect the performance of machine learning. Summary of the invention

[0003] In view of this, the present disclosure provides a method, device, electronic device and storage medium for selectively learning media samples.

[0004] One aspect of the present disclosure provides a method for selective learning of media samples, including: determining a first media sample and a second media sample from a media sample set, wherein the noise content of the first media sample is less than the noise content of the second media sample; constructing a first cross entropy loss of the first media sample based on a true label and a predicted label of the first media sample; constructing a regularization loss of the second media sample based on a priori label probability distribution and a predicted label probability distribution of the second media sample; constructing a total sample loss based on the first cross entropy loss and the regularization loss; and training a model based on the total sample loss so that the model can achieve selective learning of the first media sample and the second media sample.

[0005] Another aspect of the present disclosure provides a selective learning device for media samples, including: a media sample determination module, used to determine a first media sample and a second media sample from a media sample set, wherein the noise content of the first media sample is less than the noise content of the second media sample; a first cross entropy loss module, used to construct a first cross entropy loss of the first media sample according to the true label and the predicted label of the first media sample; a regularization loss module, used to construct a regularization loss of the second media sample according to the prior label probability distribution and the predicted label probability distribution of the second media sample; a sample total loss module, used to construct a sample total loss according to the first cross entropy loss and the regularized loss; and a selective learning module, used to train a model based on the sample total loss so that the model can achieve selective learning of the first media sample and the second media sample.

[0006] Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the selective learning method of media samples of the present disclosure.

[0007] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the selective learning method of media samples of the present disclosure when executed.

[0008] According to the above-mentioned embodiments of the present disclosure, a simple and effective selective learning method for incomplete data is proposed, which is capable of selectively learning from interpolated samples. Specifically, the selective learning method can dynamically identify the quality of interpolated samples during the training process. For high-quality samples with high confidence, by adopting empirical risk minimization, better performance can be ensured on these samples. For low-quality interpolated samples with high confidence, a regularization term is introduced to prevent the model from overfitting them, thereby maintaining model performance. In addition, by dynamically applying regularization to low-quality samples, selective learning can be performed on the interpolated data, so that a more reliable model can be trained using the interpolated samples to obtain an accurate and robust classification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0010] Figure 1A Schematically showing the change of reconstruction accuracy with respect to different missing rates when the input data contains a large number of inaccurate features;

[0011] Figure 1B A schematic diagram of the optimal decision boundary is schematically shown;

[0012] Figure 1C A schematic diagram of a biased decision boundary is schematically shown;

[0013] Figure 2 An exemplary system architecture to which a selective learning method of media samples can be applied according to an embodiment of the present disclosure is schematically shown;

[0014] Figure 3 A flowchart schematically illustrates a method for selectively learning media samples according to an embodiment of the present disclosure;

[0015] Figure 4 The schematic diagram shows an overview of the selective learning method according to an embodiment of the present disclosure;

[0016] Figure 5 A block diagram schematically shows a selective learning device for media samples according to an embodiment of the present disclosure; and

[0017] Figure 6 A block diagram of an electronic device suitable for implementing the selective learning method of media samples of the present disclosure is schematically shown according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0019] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0020] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0021] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0022] In the embodiments of the present disclosure, the collection, updating, analysis, processing, use, transmission, provision, disclosure, storage, etc. of the data involved (for example, including but not limited to user personal information) are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security, and national security.

[0023] In the embodiments of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0024] In order to improve data quality, ensure the accuracy and reliability of analysis, and minimize or eliminate the impact of missing data on model training and testing, it is necessary to process missing data first. The current methods for processing missing data can be divided into two categories: deletion methods and interpolation methods. The deletion method directly removes samples or attributes with missing features from the data set. Although this method is simple and effective for large data sets with fewer missing features, it may lead to information loss and bias when the data is not completely missing at random. The interpolation method estimates and fills in the missing data in the data set to maintain the integrity of the data set. It includes two methods: single imputation (SI) and multiple imputation (MI). In practical applications, the choice of interpolation method depends on data characteristics, missing data patterns, and specific needs. Compared with the deletion method, the interpolation method can reduce information loss. However, the accuracy of the interpolation must be considered during interpolation, because inaccurate interpolation may have a significant impact on the results.

[0025] Data imputation methods are often used to address the challenges of incomplete data during training and testing. These methods estimate and interpolate missing features based on observed data, aiming to improve data utilization and minimize the negative impact on model performance. The imputed data is used to train machine learning models.

[0026] However, when the missing rate of the original data is high, interpolation may introduce noise features, thereby changing the distribution of the original data. To verify this point of view, the reconstruction accuracy (DRA) of the heart disease dataset using different interpolation methods at different missing rates was examined.

[0027] Figure 1A The diagram schematically shows the change of reconstruction accuracy with respect to different missing rates when the input data contains a large number of inaccurate features.

[0028] like Figure 1A As shown in Figure 1, the imputed data contains a large number of inaccurate features, and as the missing rate increases, the reconstruction accuracy gradually decreases. When the imputed data is used to train a machine learning model, these inaccurate features will have a negative impact on the model.

[0029] Figure 1B A schematic diagram of the optimal decision boundary is shown schematically.

[0030] like Figure 1B As shown in Figure 2, when the model is trained using ideal complete data, the model is basically able to learn the ideal classification boundary.

[0031] Figure 1CA schematic diagram of a biased decision boundary is shown schematically.

[0032] exist Figure 1C In the model, the presence of low-quality imputation samples may cause the classification boundary to deviate from the ideal state because the model is overfitted to the low-quality imputation samples. Such errors may lead to serious consequences in practical applications.

[0033] Therefore, a method is needed that enables the model to learn from the imputed data and make more reliable predictions in the face of uncertainty caused by imputation, thereby reducing the negative impact of inaccurate imputation.

[0034] A traditional approach to address this problem is to regularize the model to prevent it from overfitting on low-quality imputed samples. For example, Label Smoothing (LS) can be applied to the original classification loss function by softening the original true labels. Penalizing Confidence (PC) proposes a confidence penalty term to prevent deep neural networks from overfitting. The goal of these methods can be viewed as indiscriminately reducing the classification loss while regularizing all training samples, but this may introduce other problems. On the one hand, for high-quality samples that should be classified with high confidence, these methods also aim to reduce their confidence, preventing the model from overfitting these samples through regularization. On the other hand, for low-quality imputed samples, these methods still try to reduce their classification loss, even though these samples themselves may not be correctly classified due to their low quality.

[0035] Regularization is a widely used technique in machine learning that aims to prevent overfitting and enhance the generalization ability of the model. Common regularization methods are usually divided into implicit regularization and explicit regularization. Implicit regularization methods do not add explicit constraints to the objective function, but indirectly affect the model complexity by modifying the learning process or optimization algorithm, including methods such as label smoothing (LS) and confidence penalty (PC). Explicit regularization methods limit model complexity by imposing additional constraints or penalties on parameters, thereby reducing the risk of overfitting. These methods tend to pay more attention to specific samples in the data, such as samples that are difficult to classify, such as focal loss (FL), and samples with imbalanced categories, such as dual focal loss (DFL).

[0036] In the process of implementing the concept of the present disclosure, the inventors found that although the multiple imputation (MI) methods take uncertainty into account in the imputation process and can mitigate the impact of inaccurate imputation to a certain extent, they cannot completely eliminate the impact of low-quality imputation samples in model training. This requires the use of regularization methods to prevent the model from overfitting on low-quality samples. However, implicit regularization methods usually apply regularization indiscriminately to all samples, which leads to reduced prediction confidence of the model for high-quality samples that are easy to classify. On the other hand, explicit regularization methods achieve regularization by increasing the weights of samples with larger losses, which may cause the model to overfit more to low-quality samples, so that when applied to low-quality samples, it may also reduce model performance and cause model bias. Therefore, the current regularization method cannot effectively solve the problem of model bias when the data set contains inaccurate interpolation samples.

[0037] Embodiments of the present disclosure provide a method, device, electronic device and storage medium for selective learning of media samples, the method comprising: determining a first media sample and a second media sample from a media sample set, wherein the noise content of the first media sample is less than the noise content of the second media sample; constructing a first cross entropy loss of the first media sample based on a true label and a predicted label of the first media sample; constructing a regularized loss of the second media sample based on a priori label probability distribution and a predicted label probability distribution of the second media sample; constructing a total sample loss based on the first cross entropy loss and the regularized loss; and training a model based on the total sample loss so that the model can achieve selective learning of the first media sample and the second media sample.

[0038] Figure 2 An exemplary system architecture 200 to which the selective learning method of media samples according to an embodiment of the present disclosure can be applied is schematically shown. It should be noted that FIG. 1 is only an example of a system architecture to which the embodiment of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiment of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0039] like Figure 2 As shown, the system architecture 200 according to this embodiment may include a first terminal device 201, a second terminal device 202, a third terminal device 203, a network 204, and a server 205. The network 204 is used to provide a medium for communication links between the first terminal device 201, the second terminal device 202, the third terminal device 203, and the server 205. The network 204 may include various connection types, such as wired and / or wireless communication links, etc.

[0040] The user may use the first terminal device 201, the second terminal device 202, and the third terminal device 203 to interact with the server 205 through the network 204 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 201, the second terminal device 202, and the third terminal device 203, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (only for example).

[0041] The first terminal device 201, the second terminal device 202, and the third terminal device 203 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0042] The server 205 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 201, the second terminal device 202, and the third terminal device 203. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0043] It should be noted that the selective learning method of media samples provided in the embodiment of the present disclosure can generally be executed by the server 205. Accordingly, the selective learning device of media samples provided in the embodiment of the present disclosure can generally be set in the server 205. The selective learning method of media samples provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 205 and can communicate with the first terminal device 201, the second terminal device 202, the third terminal device 203 and / or the server 205. Correspondingly, the selective learning device of media samples provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 205 and can communicate with the first terminal device 201, the second terminal device 202, the third terminal device 203 and / or the server 205. Alternatively, the selective learning method of media samples provided in the embodiment of the present disclosure can also be executed by the first terminal device 201, the second terminal device 202, or the third terminal device 203, or can also be executed by other terminal devices different from the first terminal device 201, the second terminal device 202, or the third terminal device 203. Correspondingly, the selective learning device for media samples provided in the embodiments of the present disclosure may also be set in the first terminal device 201, the second terminal device 202, or the third terminal device 203, or in other terminal devices different from the first terminal device 201, the second terminal device 202, or the third terminal device 203.

[0044] For example, the media sample set may be originally stored in any one of the first terminal device 201, the second terminal device 202, or the third terminal device 203 (for example, the first terminal device 201, but not limited thereto), or may be stored in an external storage device and may be imported into the first terminal device 201. Then, the first terminal device 201 may locally execute the selective learning method of the media sample provided by the embodiment of the present disclosure, or send the media sample set to other terminal devices, servers, or server clusters, and the other terminal devices, servers, or server clusters that receive the media sample set may execute the selective learning method of the media sample provided by the embodiment of the present disclosure.

[0045] It should be understood that Figure 2 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0046] Figure 3 The flowchart of the selective learning method of media samples according to the embodiment of the present disclosure is schematically shown.

[0047] like Figure 3 As shown, the method includes operations S301 to S305.

[0048] In operation S301 , a first media sample and a second media sample are determined from a media sample set, wherein a noise content of the first media sample is smaller than a noise content of the second media sample.

[0049] According to an embodiment of the present disclosure, the media samples in the media sample set may include any one of the following: text samples, image samples, audio samples, etc., and may not be limited thereto. In the case where the media sample set includes media samples with different noise contents that are directly collected (without interpolation), the first media sample with less noise content may be determined as a high-quality sample, and the second media sample with more noise content may be determined as a low-quality sample, according to the amount of noise content.

[0050] According to an embodiment of the present disclosure, a media sample set may include incomplete media samples when initially collected, that is, media samples to be interpolated that have missing features. In this case, it is necessary to first interpolate the missing features for the media samples to be interpolated that have missing features. The method may include: obtaining a training sample set, the training sample set including the media samples to be interpolated, wherein the media samples to be interpolated are represented by a multidimensional feature vector and a multidimensional mask vector corresponding to the multidimensional feature vector, the media samples to be interpolated also have a true label, and each mask element in the multidimensional mask vector uses a first value or a second value to represent whether the corresponding feature vector in the multidimensional feature vector is missing or not missing; obtaining a multidimensional weight determined for the multidimensional feature vector; based on the multidimensional weight, calculating the weighted distance between the media samples to be interpolated and other media samples, wherein the other media samples represent other media samples in the media sample set that have the same true label as the media samples to be interpolated and are other than the media samples to be interpolated; determining a preset number of target other media samples that are closest to the weighted distance to the media samples to be interpolated; interpolating the feature vector to be interpolated of the media samples to be interpolated according to the target feature vector of the target other media samples to obtain the interpolated media samples, wherein the target feature vector has a corresponding relationship with the feature vector to be interpolated.

[0051] For example, consider a training sample ,in is a D-dimensional feature vector, is the corresponding true label. When some features of the sample are missing, a D-dimensional mask vector is used to indicate whether each feature is missing. Specifically, Representation characteristics Available, while Representation characteristics Missing, is the first feature vector dimensions. The obtained training sample set can be expressed as ,in is the number of training samples in the training sample set, is the first training samples, the goal of incomplete data interpolation is to estimate the missing features by finding the data generating function. After interpolating all incomplete sample features, the complete media sample set can be obtained. Used for model training.

[0052] In this method, a multiple imputation method can be used to estimate missing data by using neighbor information, which is called Neighbor-Based Imputation (NBI). As an example, this method first uses a random forest model to obtain the feature importance of training samples. , represents the D-dimensional weight determined for the D-dimensional feature vector. Then, using the D-dimensional weight Calculate training samples With training sample set The weighted distance between other samples in the training set is calculated to ensure that the calculated distance more accurately reflects the true relationship between samples. Based on the calculated weighted distance, NBI recognition and training samples Belonging to the same category The nearest neighbor samples. For the training samples For each missing feature in The corresponding values ​​of the samples of the nearest neighbors are used to construct a distribution representing the missing features, thereby realizing interpolation for the real features.

[0053] According to an embodiment of the present disclosure, there may be multiple target feature vectors. Interpolating the feature vector to be interpolated of the media sample to be interpolated based on the target feature vector of the target other media sample may include: in response to determining that the feature to be interpolated is a continuous feature, interpolating according to the mean of multiple target feature vectors; in response to determining that the feature to be interpolated is a discrete feature, interpolating according to the mode of multiple target feature vectors.

[0054] Specifically, for continuous features, a Gaussian distribution can be constructed and interpolated with its mean. For discrete features, a multinomial distribution can be constructed and interpolated with its mode. After interpolating all missing features, the interpolated samples can be obtained. .

[0055] According to an embodiment of the present disclosure, the media sample set obtained based on the above may include interpolated media samples obtained by interpolating missing features for the media samples to be interpolated that have missing features. In this case, high-quality samples and low-quality samples may be determined based on the interpolation accuracy. Specifically, when it is determined that the interpolation accuracy of the first media sample is greater than the interpolation accuracy of the second media sample, the first media sample with a higher interpolation accuracy may be determined as a high-quality sample, and the second media sample with a lower interpolation accuracy may be determined as a low-quality sample.

[0056] For example, suppose the interpolated media sample set It consists of high-quality samples and low-quality samples. Among them, high-quality samples can include original samples without missing features and accurately interpolated media samples, and low-quality samples can include inaccurately interpolated media samples.

[0057] According to an embodiment of the present disclosure, before performing the above operation S301, the model may also be first trained for e rounds. Specifically, the third cross entropy loss of the media sample may be calculated according to the real label and the predicted label of the media sample in the media sample set, and the model may be trained for a preset round based on the third cross entropy loss.

[0058] For example, given the full dataset after imputation , you can use Empirical Risk Minimization (ERM) to train the model . Formally, it can be expressed as formula (1).

[0059]

[0060] Formula (1)

[0061] In formula (1), The third cross entropy loss function can be expressed as

[0062] It should be noted that in the classification task, It may include cross entropy loss and square loss, etc., which are not limited here.

[0063] Through the above-mentioned embodiments of the present disclosure, low-quality samples can be accurately identified.

[0064] According to an embodiment of the present disclosure, a media sample set may include multiple media samples. High-quality samples and low-quality samples may also be determined by a cross-entropy loss value. Specifically, the above operation S301 may include: for each media sample in the media sample set, according to the real label and the predicted label of the media sample, calculating the second cross-entropy loss value of the media sample, and obtaining multiple second cross-entropy loss values ​​determined for the multiple media samples in the media sample set; according to the multiple second cross-entropy loss values, determining the first media sample and the second media sample, the second cross-entropy loss value of the first media sample is less than the second cross-entropy loss value of the second media sample.

[0065] For example, the formula (1) can be Think of it as the widely used cross entropy loss , establish the second cross entropy loss as formula (2).

[0066] Formula (2)

[0067] In formula (2), Indicates the number of categories of sample labels, Representation sample In category The real label under The corresponding model for the sample Belongs to category The probability of the predicted label.

[0068] Low-quality imputed samples usually have higher cross entropy loss Inspired by , we can automatically identify low-quality samples by using the loss value of training samples. Specifically, during the training process, given samples, the second cross entropy loss value of each sample can be calculated according to formula (2), and when it is determined that the second cross entropy loss value of the first media sample is less than the second cross entropy loss value of the second media sample, the first media sample with the lower second cross entropy loss value is determined as a high-quality sample, and the second media sample with the lower second cross entropy loss value is determined as a low-quality sample.

[0069] According to an embodiment of the present disclosure, the above-mentioned determining the first media sample and the second media sample based on multiple second cross entropy loss values ​​may include: sorting multiple media samples in a media sample set in ascending order according to the multiple second cross entropy loss values ​​to obtain an ascending sorting result; determining a first numerical value according to the product of a predetermined ratio and the total number of samples of the multiple media samples; determining a second numerical value according to the difference between the total number of samples and the first numerical value; determining a media sample corresponding to the first second numerical value elements in the ascending sorting result as the first media sample; and determining a media sample corresponding to the last first numerical value elements in the ascending sorting result as the second media sample.

[0070] According to an embodiment of the present disclosure, formally, the corresponding distribution of the imputed samples can follow Hubler's -Pollution model, that is, satisfying formula (3).

[0071] Formula (3)

[0072] In formula (3), represents the distribution of high-quality samples, Represents the distribution of low-quality samples. is a parameter that represents the proportion of inaccurate imputed data in the overall dataset. For a specific model and dataset, Is a fixed but unknown constant. In order to simplify the calculation in practical applications, the present disclosure may regard it as an adjustable hyperparameter and determine the above predetermined ratio based on it.

[0073] Afterwards, the samples can be sorted in ascending order according to the calculated second cross entropy loss value, for example, formula (4) can be obtained.

[0074] Formula (4)

[0075] In formula (4), It means to sort the second cross entropy loss value in ascending order and return the original index of the sorted element, which can be recorded as . Indicates The second cross entropy loss value of the media sample. Then, according to the Huber's -Contamination acceleration, the proportion of low-quality samples is set to , with low loss media samples can be regarded as high-quality samples and can be expressed as a set as shown in formula (5): .

[0076] Formula (5)

[0077] Accordingly, the high loss The remaining media samples can be regarded as low-quality samples and can be expressed as the set shown in formula (6): .

[0078] Formula (6)

[0079] In this way, we can estimate the distribution of the sample in a non-parametric way. or , dividing the media sample set into high-quality samples and low-quality samples to obtain a first media sample and a second media sample.

[0080] In the process of implementing the concept of the present disclosure, the inventors discovered that the data interpolation method mainly estimates and interpolates missing features based on the observed data. However, the sample set obtained by the interpolation method contains low-quality samples with inaccurate interpolation, and there are often a large number of inaccurate features in the low-quality samples. These features may distort the original distribution of the data and cause the model to learn inaccurate information. On this basis, optimizing all samples using the above-mentioned empirical risk minimization method may cause the model to learn biased information. Specifically, during the model training process, low-quality samples may introduce bias and noise, thereby preventing the model from effectively capturing the potential real information in the data. In this case, the model may overfit these low-quality data during the training process, resulting in deviations in the decision boundary, which in turn leads to a decrease in performance on unseen data.

[0081] To this end, a method is needed to reduce the negative impact of inaccurate imputation samples so that the model can effectively learn from the imputed data.

[0082] An effective way to prevent the model from overfitting low-quality samples is regularization. However, the goal of related regularization methods can usually be regarded as reducing the classification loss of all samples indiscriminately while regularizing all training samples. On the one hand, for simple samples that should be classified with high confidence, these methods will reduce their confidence to prevent overfitting. On the other hand, for low-quality samples that may not be correctly classified, such methods will also reduce their loss, resulting in biased model learning. Therefore, related regularization methods cannot solve the problem of model bias caused by overfitting of inaccurate imputed samples. In order to solve this problem, it is necessary to develop an effective and robust selective learning for incomplete data (SLID) method for imputed samples. For example, regularization can be applied in the loss function to prevent the model from overfitting biased samples.

[0083] In view of the fact that related regularization methods cannot solve the model bias problem caused by inaccurate interpolation data introduced by interpolation technology, the present disclosure proposes a selective learning method for incomplete data based on the quality of interpolation data. The selective learning method proposed in the present disclosure belongs to an explicit regularization method. The method follows the following principles: For high-quality samples, cross entropy loss is used to allow the model to learn them normally. For low-quality samples, regularization is applied to them based on Kullback-Leibler (KL) loss. And operations S302~S303 can be constructed accordingly.

[0084] In operation S302 , a first cross entropy loss of the first media sample is constructed according to the true label and the predicted label of the first media sample.

[0085] According to an embodiment of the present disclosure, after the samples are divided into high-quality samples and low-quality samples, the goal of the method is to enable the model to effectively learn from high-quality samples in , while minimizing the set To this end, for high-quality samples, the first cross entropy loss as shown in formula (7) can be directly used to train the model.

[0086] Formula (7)

[0087] In operation S303 , a regularization loss of the second media sample is constructed according to the prior label probability distribution and the predicted label probability distribution of the second media sample.

[0088] According to an embodiment of the present disclosure, regularization may be applied to low-quality samples.

[0089] In this case, we can first calculate the prior class probability of this type of sample and predicted class probability The Kullback-Leibler (KL) divergence between the two can be used to obtain the prior label probability distribution and the predicted label probability distribution. Then, for the low-quality sample set, Construct the regularization loss as shown in formula (8) .

[0090] Formula (8)

[0091] In formula (8), The KL divergence, which measures the difference between two distributions, can be calculated using formula (9).

[0092] Formula (9)

[0093] In formula (9), represents the prior label probability distribution, represents the predicted label probability distribution predicted by the model, and Corresponding to the categories The prior probability and predicted probability of .

[0094] In operation S304 , a sample total loss is constructed according to the first cross entropy loss and the regularization loss.

[0095] According to an embodiment of the present disclosure, based on formula (7) to formula (9), the total sample loss can be expressed as formula (10).

[0096] Formula (10)

[0097] In formula (10), is a hyperparameter that indicates the strength of regularization.

[0098] Through the above embodiments of the present disclosure, a novel regularization method is proposed, which can dynamically distinguish high-quality and low-quality samples and apply regularization only to low-quality samples. This method enables the model to accurately classify high-quality samples while reducing the negative impact of low-quality samples, so that the model can effectively learn from high-quality samples while minimizing the impact of low-quality samples on the model, thereby achieving selective learning of incomplete data.

[0099] In operation S305 , a model is trained based on the total loss of the samples so that the model can selectively learn the first media sample and the second media sample.

[0100] According to the embodiments of the present disclosure, the selective learning method is intended to solve the problem of low-quality samples in the dataset due to inaccurate interpolation. The main goal of selective learning is to effectively learn the interpolated dataset. At the same time, it mitigates the negative impact of low-quality samples.

[0101] Figure 4 The figure schematically shows an overview of the selective learning method according to an embodiment of the present disclosure.

[0102] like Figure 4 As shown, during the model training phase, it is possible to automatically distinguish High-quality samples follow the distribution The low-quality samples are excluded. The model is optimized by empirical risk minimization for high-quality samples. At the same time, regularization is applied to low-quality samples to encourage the model to assign lower confidence scores to these samples.

[0103] Through the above-mentioned embodiments of the present disclosure, a simple and effective selective learning method for incomplete data is proposed, which can selectively learn from interpolated samples. Specifically, the selective learning method can dynamically identify the quality of interpolated samples during the training process. For high-quality samples with high confidence, by adopting empirical risk minimization, better performance can be ensured on these samples. For low-quality interpolated samples with high confidence, a regularization term is introduced to prevent the model from overfitting them, thereby maintaining the model performance and being able to identify these samples during the test process to avoid making decisions. In addition, by dynamically applying regularization to low-quality samples, the interpolated data can be selectively learned, so that a more reliable model can be trained using the interpolated samples to obtain an accurate and robust classification model. A large number of experiments have proved that the selective learning method improves the accuracy and reliability of the model.

[0104] Figure 5 A block diagram of a selective learning device for media samples according to an embodiment of the present disclosure is schematically shown.

[0105] like Figure 5 As shown, the selective learning device 500 for media samples includes a media sample determination module 510 , a first cross entropy loss module 520 , a regularization loss module 530 , a sample total loss module 540 and a selective learning module 550 .

[0106] The media sample determination module 510 is configured to determine a first media sample and a second media sample from the media sample set, wherein the noise content of the first media sample is smaller than the noise content of the second media sample.

[0107] The first cross entropy loss module 520 is used to construct a first cross entropy loss for the first media sample according to the true label and the predicted label of the first media sample.

[0108] The regularization loss module 530 is used to construct a regularization loss for the second media sample according to the prior label probability distribution and the predicted label probability distribution of the second media sample.

[0109] The sample total loss module 540 is used to construct the sample total loss according to the first cross entropy loss and the regularization loss.

[0110] The selective learning module 550 is used to train the model based on the total loss of samples so that the model can selectively learn the first media sample and the second media sample.

[0111] According to an embodiment of the present disclosure, a media sample set includes a plurality of media samples. The media sample determination module includes a second cross entropy loss value calculation unit and a media sample determination unit.

[0112] The second cross entropy loss value calculation unit is used to calculate the second cross entropy loss value of the media sample for each media sample in the media sample set according to the real label and the predicted label of the media sample, and obtain multiple second cross entropy loss values ​​determined for multiple media samples in the media sample set.

[0113] The media sample determination unit is used to determine a first media sample and a second media sample according to a plurality of second cross entropy loss values, wherein the second cross entropy loss value of the first media sample is smaller than the second cross entropy loss value of the second media sample.

[0114] According to an embodiment of the present disclosure, the media sample determination unit includes a sorting subunit, a product calculation subunit, a difference calculation subunit, a first media sample determination subunit and a second media sample determination subunit.

[0115] The sorting subunit is used to sort the multiple media samples in the media sample set in ascending order according to the multiple second cross entropy loss values ​​to obtain an ascending sorting result.

[0116] The product calculation subunit is used to determine a first value according to a product of a predetermined ratio and a total number of samples of the plurality of media samples.

[0117] The difference calculation subunit is used to determine the second value according to the difference between the total number of samples and the first value.

[0118] The first media sample determination subunit is used to determine the media sample corresponding to the first second numerical element in the ascending sorting result as the first media sample.

[0119] The second media sample determination subunit is used to determine the media sample corresponding to the last first numerical element in the ascending sorting result as the second media sample.

[0120] According to an embodiment of the present disclosure, the media sample set includes interpolated media samples obtained by interpolating missing features for media samples to be interpolated that have missing features, and the interpolation accuracy of the first media sample is greater than the interpolation accuracy of the second media sample.

[0121] According to an embodiment of the present disclosure, the selective learning device of media samples further includes a training sample set acquisition module, a multi-dimensional weight acquisition module, a weighted distance calculation module, a target other media sample determination module and an interpolation module.

[0122] The training sample set acquisition module is used to acquire a training sample set, which includes a media sample to be interpolated, wherein the media sample to be interpolated is represented by a multidimensional feature vector and a multidimensional mask vector corresponding to the multidimensional feature vector, and the media sample to be interpolated also has a true label, and each mask element in the multidimensional mask vector uses a first value or a second value to indicate whether the corresponding feature vector in the multidimensional feature vector is missing or not missing.

[0123] The multi-dimensional weight acquisition module is used to obtain the multi-dimensional weight determined for the multi-dimensional feature vector.

[0124] The weighted distance calculation module is used to calculate the weighted distance between the media sample to be interpolated and other media samples based on multi-dimensional weights, wherein the other media samples represent other media samples in the media sample set that have the same true label as the media sample to be interpolated and are other than the media sample to be interpolated.

[0125] The target other media sample determination module is used to determine a preset number of target other media samples that are closest in weighted distance to the media sample to be interpolated.

[0126] The interpolation module is used to interpolate the feature vector to be interpolated of the media sample to be interpolated according to the target feature vector of the target other media sample to obtain the interpolated media sample, wherein the target feature vector has a corresponding relationship with the feature vector to be interpolated.

[0127] According to an embodiment of the present disclosure, the target feature vector has multiple interpolation modules, including a mean interpolation unit and a mode interpolation unit.

[0128] The mean interpolation unit is used to perform interpolation according to the mean of multiple target feature vectors in response to determining that the feature to be interpolated is a continuous feature.

[0129] The mode interpolation unit is used to perform interpolation according to the mode of multiple target feature vectors in response to determining that the feature to be interpolated is a discrete feature.

[0130] According to an embodiment of the present disclosure, the selective learning device of media samples also includes a third cross entropy loss calculation module and a training module.

[0131] The third cross entropy loss calculation module is used to calculate the third cross entropy loss of the media sample according to the real label and the predicted label of the media sample in the media sample set.

[0132] The training module is used to train the model for a preset number of rounds based on the third cross entropy loss.

[0133] According to the embodiments of the present invention, any one or more of the modules, units, and sub-units, or at least part of the functions of any one of them, can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, units, and sub-units can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, units, and sub-units can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, units, and sub-units can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be performed.

[0134] For example, any of the media sample determination module 510, the first cross entropy loss module 520, the regularization loss module 530, the sample total loss module 540, and the selective learning module 550 may be combined in one module / unit / sub-unit for implementation, or any of the modules / units / sub-units may be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units may be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present disclosure, at least one of the media sample determination module 510, the first cross entropy loss module 520, the regularization loss module 530, the sample total loss module 540, and the selective learning module 550 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or may be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation modes of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the media sample determination module 510, the first cross entropy loss module 520, the regularization loss module 530, the sample total loss module 540, and the selective learning module 550 may be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function may be performed.

[0135] Figure 6 A block diagram of an electronic device suitable for implementing the selective learning method of media samples of the present disclosure is schematically shown according to an embodiment of the present disclosure. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0136] like Figure 6 As shown, the electronic device 600 according to an embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 to a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include an onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0137] In RAM 603, various programs and data required for the operation of electronic device 600 are stored. Processor 601, ROM 602 and RAM 603 are connected to each other via bus 604. Processor 601 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 602 and / or RAM 603. It should be noted that the program can also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0138] According to an embodiment of the present disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to the bus 604. The system 600 may further include one or more of the following components connected to the input / output (I / O) interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 608 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage portion 608 as needed.

[0139] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0140] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the selective learning method of the media sample according to the embodiment of the present disclosure is implemented.

[0141] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include, but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0142] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 602 and / or the RAM 603 described above and / or one or more memories other than the ROM 602 and the RAM 603 .

[0143] An embodiment of the present disclosure also includes a computer program product, which includes a computer program, which contains program code for executing the method provided by the embodiment of the present disclosure. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the selective learning method of media samples provided by the embodiment of the present disclosure.

[0144] When the computer program is executed by the processor 601, the above functions defined in the system / device of the embodiment of the present disclosure are executed. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0145] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0146] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments and / or claims of the present disclosure can be combined and / or combined in a variety of ways, even if such a combination or combination is not explicitly recorded in the present disclosure. In particular, without departing from the spirit and teaching of the present disclosure, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0148] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A selective learning method for media samples, comprising: Determine a first media sample and a second media sample from the media sample set, wherein a noise content of the first media sample is less than a noise content of the second media sample; Constructing a first cross entropy loss for the first media sample according to the true label and the predicted label of the first media sample; constructing a regularization loss for the second media sample according to the prior label probability distribution and the predicted label probability distribution of the second media sample; Constructing a sample total loss according to the first cross entropy loss and the regularization loss; and A model is trained based on the total sample loss so that the model can selectively learn the first media sample and the second media sample.

2. The method according to claim 1, wherein: The media sample set includes a plurality of media samples; Determining the first media sample and the second media sample from the media sample set includes: For each media sample in the media sample set, calculating a second cross entropy loss value of the media sample according to a true label and a predicted label of the media sample, to obtain a plurality of second cross entropy loss values ​​determined for the plurality of media samples in the media sample set; as well as A first media sample and a second media sample are determined according to the plurality of second cross entropy loss values, wherein the second cross entropy loss value of the first media sample is smaller than the second cross entropy loss value of the second media sample.

3. The method according to claim 2, wherein: The determining the first media sample and the second media sample according to the plurality of second cross entropy loss values ​​comprises: Sorting the multiple media samples in the media sample set in ascending order according to the multiple second cross entropy loss values ​​to obtain an ascending order result; determining a first value according to a product of a predetermined ratio and a total number of samples of the plurality of media samples; Determine a second value according to the difference between the total number of samples and the first value; Determine the media sample corresponding to the second-most numerical element in the ascending sort result as the first media sample; and The media sample corresponding to the last first numerical element in the ascending sorting result is determined as the second media sample.

4. The method according to claim 1, wherein: The media sample set includes interpolated media samples obtained by interpolating missing features for the to-be-interpolated media samples having missing features, and the interpolation accuracy of the first media samples is greater than the interpolation accuracy of the second media samples.

5. The method according to claim 4, further comprising: Before determining the first media sample and the second media sample from the media sample set, Acquire a training sample set, wherein the training sample set includes the media sample to be interpolated, wherein the media sample to be interpolated is represented by a multidimensional feature vector and a multidimensional mask vector corresponding to the multidimensional feature vector, the media sample to be interpolated also has a true label, and each mask element in the multidimensional mask vector uses a first value or a second value to represent whether a corresponding feature vector in the multidimensional feature vector is missing or not missing; Obtaining a multidimensional weight determined for the multidimensional feature vector; Based on the multi-dimensional weights, a weighted distance between the media sample to be interpolated and other media samples is calculated, wherein the other media samples represent other media samples in the media sample set that have the same true label as the media sample to be interpolated and are excluding the media sample to be interpolated; Determine a preset number of target other media samples that are closest in weighted distance to the media sample to be interpolated; The interpolated media sample is obtained by interpolating the feature vector to be interpolated of the media sample to be interpolated according to the target feature vector of the target other media sample, wherein the target feature vector has a corresponding relationship with the feature vector to be interpolated.

6. The method according to claim 5, wherein: There are multiple target feature vectors; interpolating the feature vector to be interpolated of the media sample to be interpolated according to the target feature vector of the target other media sample comprises: In response to determining that the feature to be interpolated is a continuous feature, interpolating according to the mean of a plurality of the target feature vectors; and In response to determining that the feature to be interpolated is a discrete feature, interpolation is performed according to a mode of a plurality of the target feature vectors.

7. The method according to any one of claims 1 to 6, further comprising: Before determining the first media sample and the second media sample from the media sample set, Calculating a third cross entropy loss of the media sample according to the true label and the predicted label of the media sample in the media sample set; as well as Based on the third cross entropy loss, the model is trained for a preset round.

8. A selective learning device for media samples, comprising: A media sample determination module, configured to determine a first media sample and a second media sample from a media sample set, wherein a noise content of the first media sample is less than a noise content of the second media sample; A first cross entropy loss module, configured to construct a first cross entropy loss of the first media sample according to a true label and a predicted label of the first media sample; A regularization loss module, configured to construct a regularization loss of the second media sample according to a priori label probability distribution and a predicted label probability distribution of the second media sample; A sample total loss module, configured to construct a sample total loss according to the first cross entropy loss and the regularization loss; and A selective learning module is used to train a model based on the total loss of the samples so that the model can selectively learn the first media sample and the second media sample.

9. An electronic device, comprising: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method according to any one of claims 1 to 7.

Images