Noise sample processing method, device, equipment, medium and program product

Abstract

The invention discloses a noise sample processing method, device and equipment, a medium and a program product, and relates to the field of financial science and technology, and the method comprises the steps: predicting a current round of input sample set based on a preset classifier, and obtaining a current noise-free sample set and a current potential noise sample set; taking the current noiseless sample set as a training set, and taking the current potential noise sample set and the current noise pool sample as a test set for learning to obtain a current to-be-corrected sample set; determining a target correction sample of the key to-be-corrected sample in the current to-be-corrected sample set, and updating the current noise pool sample; and according to the current noise-free sample set and the target correction sample, updating the current round of input sample set, and returning to execute the operation of predicting the current round of input sample set based on the preset classifier until the noise processing target sample is output. According to the technical scheme of the embodiment of the invention, under the condition that the low noise rate of the sample is ensured, the test cost of the sample is reduced, and the time performance of sample noise processing is improved.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and in particular to a noise sample processing method, device, equipment, medium and program product. Background Art

[0002] Data sets in financial scenarios often contain a lot of noise, redundancy and irrelevant information, which may be caused by errors in data collection, processing or recording, or by factors that cannot be predicted or controlled in actual business. Noise samples will have a great impact on the prediction effect of the model, not only significantly reducing the prediction accuracy and generalization performance of the model, but also affecting feature selection and feature sorting.

[0003] Data noise is mainly divided into feature noise and label noise. Feature noise refers to the noise caused by certain features or attributes of the data itself during data processing and analysis, such as data outliers, missing data, etc. Label noise refers to the deviation between the target label used for training and the true label of the corresponding instance itself. Label noise in financial scenarios may have many reasons. For example, in the default probability prediction model, the customer's loan risk often requires a certain period of performance to be fully exposed, and the label of whether the customer defaults during the performance period is inaccurate. In the credit scoring model, due to biased or inaccurate data, subjective judgment of the scorer, lack of sufficient data to accurately assess the customer's credit status, etc., the user's credit score is different from the actual credit status.

[0004] At present, sample selection methods are mainly divided into data denoising and sampling, which are essentially means of data preprocessing. In practical applications, the method of directly deleting the identified noise samples may cause non-noise samples to be mistakenly deleted, thereby losing effective information, while the method of correcting the identified noise samples can retain the complete sample information, but it is easy to introduce additional noise during the correction process, resulting in poor performance of existing sample denoising time and high inspection costs. Summary of the invention

[0005] The present invention provides a noise sample processing method, device, equipment, medium and program product to solve the problems of poor time performance and high inspection cost in sample data denoising in the financial field.

[0006] According to one aspect of the present invention, a noise sample processing method is provided, comprising:

[0007] Get the current round input sample set;

[0008] Predict the current round of input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set;

[0009] Active learning is performed using the current noise-free sample set as the training set, the current potential noise sample set and the current noise pool sample set as the test set to obtain the current sample set to be corrected;

[0010] Determine the target correction sample of the key sample to be corrected in the current sample set to be corrected, and update the current noise pool sample according to the samples other than the key sample to be corrected in the current sample set to be corrected;

[0011] According to the current noise-free sample set and the target correction sample, the current round input sample set is updated, and the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set is returned until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

[0012] According to another aspect of the present invention, there is provided a noise sample processing device, comprising:

[0013] The sample acquisition module is used to obtain the input sample set of the current round;

[0014] A sample classification module is used to predict the current round of input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set;

[0015] An active learning module is used to perform active learning using the current noise-free sample set as a training set, the current potential noise sample set and the current noise pool sample as a test set to obtain the current sample set to be corrected;

[0016] The sample correction and noise pool update module is used to determine the target correction sample of the key sample to be corrected in the current sample set to be corrected, and update the current noise pool sample according to the samples other than the key sample to be corrected in the current sample set to be corrected;

[0017] The sample processing iteration module is used to update the current round input sample set according to the current noise-free sample set and the target correction sample, and return to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

[0018] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0019] at least one processor; and

[0020] a memory communicatively connected to the at least one processor; wherein,

[0021] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the noise sample processing method described in any embodiment of the present invention.

[0022] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the noise sample processing method described in any embodiment of the present invention when executed.

[0023] According to another aspect of the present invention, a computer program product is provided, characterized in that the computer program product comprises a computer program, and when the computer program is executed by a processor, the noise sample processing method according to any embodiment of the present invention is implemented.

[0024] The technical solution of the embodiment of the present invention obtains the current round input sample set, thereby predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, and then uses the current noise-free sample set as the training set, and the current potential noise sample set and the current noise pool sample as the test set for active learning to obtain the current sample set to be corrected, further determines the target correction sample of the key sample to be corrected in the current sample set to be corrected, and updates the current noise pool sample according to the samples other than the key sample to be corrected in the current sample set to be corrected, so as to update the current round input sample set according to the current noise-free sample set and the target correction sample, and returns to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met, and the noise processing target sample is output. This solution integrates the idea of ​​active learning and iteratively corrects the sample set, which can not only quickly reduce the noise content in the sample set, but also avoid information loss of key samples. On the basis of filtering out unfavorable noise samples, it expands the clean sample set, effectively improves the quality of sample sets in financial scenarios, ensures low noise rate of sample set labels while reducing verification costs, so that the final prediction effect of the model trained on processed samples is close to that of the model trained on noise-free samples, solves the problems of poor time performance and high verification cost of sample data denoising in the financial field, and can reduce the verification cost of samples and improve the time performance of sample noise processing while ensuring low noise rate of samples.

[0025] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A flowchart of a noise sample processing method provided in Embodiment 1 of the present invention;

[0028] Figure 2 A flowchart of a noise sample processing method provided in Embodiment 2 of the present invention;

[0029] Figure 3 A noise sample recognition logic diagram provided in the second embodiment of the present invention;

[0030] Figure 4 A schematic diagram of a noise sample processing method framework provided in Embodiment 2 of the present invention;

[0031] Figure 5 A schematic diagram of a sample iterative processing process provided in Embodiment 2 of the present invention;

[0032] Figure 6 A schematic diagram of the structure of a noise sample processing device provided in Embodiment 3 of the present invention;

[0033] Figure 7 A schematic diagram of the structure of an electronic device that can be used to implement an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0035] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] Embodiment 1

[0037] Figure 1 This is a flowchart of a noise sample processing method provided in the first embodiment of the present invention. This embodiment can be applied to the case of efficient noise filtering of samples in the financial field. The method can be executed by a noise sample processing device. The noise sample processing device can be implemented in the form of hardware and / or software. The noise sample processing device can be configured in an electronic device. The electronic device can be a computer or a server, etc. This solution performs multiple rounds of iterative denoising on the sample set of the original financial scene. The processing logic of each iterative processing round is the same, and one iterative processing process is used as an example for explanation. Figure 1 As shown, the method includes:

[0038] Step 110: Obtain the current round input sample set.

[0039] The current round input sample set may be a sample set to be processed in the current iterative processing round.

[0040] In an embodiment of the present invention, if the current round is the first time, the input sample set of the current round is the sample set of the original financial scenario (specifically, a sample data set under a certain type of financial scenario can be collected according to identification needs); if the current round is not the first time, the input sample set of the current round is the entire noise-free sample set determined in the previous round of the current round.

[0041] Step 120: predict the current round input sample set based on multiple preset classifiers to obtain a current noise-free sample set and a current potential noise sample set.

[0042] The preset classifier may be a pre-set base classifier. The base classifier may include one or more classifiers in basic statistical methods such as decision tree, logistic regression, K nearest neighbor, SVM (Support Vector Machine) and the like. The current noise-free sample set may be a sample set that is first identified as noise-free in the current round of input sample sets. The current potential noise sample set may be a sample set that is first identified as having noise in the current round of input sample sets.

[0043] In an embodiment of the present invention, the current round input sample set can be input into multiple preset classifiers respectively, so as to identify the type of samples in the current round input sample set through multiple preset classifiers, that is, determine whether each sample is a noisy sample, and then obtain the current noise-free sample set composed of the identified noise-free samples, and the current potential noise sample set composed of the identified noise samples.

[0044] Step 130: Active learning is performed using the current noise-free sample set as a training set, the current potential noise sample set and the current noise pool sample set as a test set to obtain a current sample set to be corrected.

[0045] Among them, active learning can be understood as a way to reduce the labeling cost and train a model with better performance through technical means or mathematical methods. Active learning generally includes but is not limited to several steps such as sample selection, model training, model prediction, and model update. The current sample set to be corrected can be a candidate sample set that needs to be corrected for labels, which is identified from the potential noise sample set and the current noise pool sample.

[0046] It should be noted that if the current round is the first time, the current noise pool samples are empty. If the current round is the second time, the current noise pool samples are the samples in the sample set of the original financial scenario except the noise-free sample set and the potential noise sample set predicted in the first round. That is, when the current round is not the first time, the current noise pool samples are the samples in the input sample set of the previous round except the predicted noise-free sample set and the potential noise sample set.

[0047] In an embodiment of the present invention, a required number of classifiers can be established, and then the current noise-free sample set can be trained as a training set, so that the trained classifier can be used to actively learn the current potential noise sample set and the current noise pool sample as a test set to determine the current sample set to be corrected that needs label correction.

[0048] Step 140: determine the target correction sample of the key sample to be corrected in the current set of samples to be corrected, and update the current noise pool sample according to the samples in the current set of samples to be corrected except the key sample to be corrected.

[0049] The key samples to be corrected may be the sample set whose labels are corrected in the current sample set to be corrected, and the target corrected samples may be the key samples to be corrected after the labels are corrected.

[0050] In an embodiment of the present invention, key samples to be corrected can be screened out from the current sample set to be corrected based on preset screening rules, and then labels of the key samples to be corrected in the current sample set to be corrected are corrected to obtain target corrected samples, and further samples in the current sample set to be corrected except the target corrected samples are added to the noise pool to update the current noise pool samples.

[0051] Step 150: Update the current round input sample set according to the current noise-free sample set and the target correction sample, and return to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

[0052] The sample optimization iteration cutoff condition may be a preset sample processing iteration stop condition. The noise processing target sample may be a sample that satisfies the sample optimization iteration cutoff condition after performing multiple noise filtering iterations on the first round of sample sets.

[0053] Optionally, the sample optimization iteration cutoff condition may include but is not limited to the number of samples in the new sample set output from two rounds tending to be stable, the performance of the preset classifier tending to be stable, or the number of noise samples identified in a round is less than a threshold.

[0054] In an embodiment of the present invention, the current noise-free sample set and the target corrected sample can be used as the new current round input sample set, and the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set is returned, and it is judged whether the sample set output in each round meets the sample optimization iteration cutoff condition. If the sample optimization iteration cutoff condition is not met, the iteration is continued. If the sample optimization iteration cutoff condition is met, the samples in the final output sample set are used as noise processing target samples, so as to train the model through the noise processing target samples, and then perform high-accuracy financial risk identification through the trained model.

[0055] The technical solution of the embodiment of the present invention obtains the current round input sample set, thereby predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, and then uses the current noise-free sample set as the training set, and the current potential noise sample set and the current noise pool sample as the test set for active learning to obtain the current sample set to be corrected, further determines the target correction sample of the key sample to be corrected in the current sample set to be corrected, and updates the current noise pool sample according to the samples other than the key sample to be corrected in the current sample set to be corrected, so as to update the current round input sample set according to the current noise-free sample set and the target correction sample, and returns to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met, and the noise processing target sample is output. This solution integrates the idea of ​​active learning and iteratively corrects the sample set, which can not only quickly reduce the noise content in the sample set, but also avoid information loss of key samples. On the basis of filtering out unfavorable noise samples, it expands the clean sample set, effectively improves the quality of sample sets in financial scenarios, ensures low noise rate of sample set labels while reducing verification costs, so that the final prediction effect of the model trained on processed samples is close to that of the model trained on noise-free samples, solves the problems of poor time performance and high verification cost of sample data denoising in the financial field, and can reduce the verification cost of samples and improve the time performance of sample noise processing while ensuring low noise rate of samples.

[0056] Embodiment 2

[0057] Figure 2 This is a flowchart of a noise sample processing method provided in the second embodiment of the present invention. This embodiment is specific based on the above embodiment and provides a specific optional implementation method for determining the target correction sample of the key to-be-corrected sample in the current to-be-corrected sample set. Figure 2 As shown, the method includes:

[0058] Step 210: Obtain the input sample set of the current round.

[0059] Step 220: Predict the current round input sample set based on multiple preset classifiers to obtain a current noise-free sample set and a current potential noise sample set.

[0060] In an optional embodiment of the present invention, after predicting the current round of input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, it can also include: if the current potential noise sample set has sample overlap with the previous round of target correction samples, then the previous round of target correction samples in the current potential noise sample set are eliminated; and the previous round of target correction samples are added to the current noise-free sample set.

[0061] The target correction sample of the previous round may be a target correction sample determined in the previous round corresponding to the round of the input sample set of the current round.

[0062] In an embodiment of the present invention, the last round of target correction samples can be obtained, and the current potential noise sample set is compared with the last round of target correction samples. When it is identified that the current potential noise sample set and the last round of target correction samples have sample overlap, the last round of target correction samples in the current potential noise sample set are removed, and the last round of target correction samples are added to the current noise-free sample set. Since the last round of target correction samples are samples with correct labels, by removing the last round of target correction samples in the current potential noise sample set, the waste of computing resources and verification costs caused by repeated predictions and repeated label corrections can be avoided.

[0063] Step 230: Active learning is performed using the current noise-free sample set as a training set, the current potential noise sample set and the current noise pool sample set as a test set to obtain a current sample set to be corrected.

[0064] Step 240: Sort the samples in the current sample set to be corrected according to the sample importance index to obtain sorting data of the samples to be corrected.

[0065] The sample importance index may be a pre-set condition for screening the current set of samples to be corrected. The sample ranking data to be corrected may be used to describe the ranking of samples in the current set of samples to be corrected according to their importance.

[0066] In the embodiment of the present invention, the importance of each sample in the current sample set to be corrected can be evaluated according to the sample importance index, so as to sort the samples in the current sample set to be corrected based on their importance, and obtain sorting data of the samples to be corrected.

[0067] In an optional embodiment of the present invention, before sorting the samples in the current sample set to be corrected according to the sample importance index to obtain the sample sorting data to be corrected, it can also include: determining each predicted label of each sample in the current sample set to be corrected by each preset classifier, and the integrated sample label of each sample in the current sample set to be corrected; counting the number of times that the predicted label of each sample in the current sample set to be corrected is inconsistent with the corresponding integrated sample label, and obtaining the number of label prediction differences of each sample in the current sample set to be corrected; determining the sample importance index of each sample in the current sample set to be corrected according to the number of label prediction differences and the information entropy of each sample in the current sample set to be corrected.

[0068] The predicted label can be the sample type (noise sample or noise-free sample) of the sample in the current sample set to be corrected predicted by the preset classifier. The integrated sample label can be the sample label determined by combining the predicted labels of the sample by multiple preset classifiers. The number of label prediction differences can be the total number of predicted labels for the same sample that is different from the integrated sample label.

[0069] In an embodiment of the present invention, each preset classifier can be used to predict the sample type of each sample in the current sample set to be corrected, and the predicted label of each sample in the current sample set to be corrected can be obtained, and then the multiple predicted labels of each sample can be combined to determine the corresponding integrated sample label, and further the total number of times the predicted label of each sample in the current sample set to be corrected is inconsistent with the corresponding integrated sample label is used as the number of label prediction differences of the corresponding sample in the current sample set to be corrected, and the information entropy of each sample in the current sample set to be corrected is calculated, so as to comprehensively combine the number of label prediction differences and the information entropy of each sample in the current sample set to be corrected, and determine the sample importance index of each sample in the current sample set to be corrected, that is, comprehensively evaluate the importance of each sample in the current sample set to be corrected from multiple dimensions to screen out samples with greater correction value.

[0070] In an optional embodiment of the present invention, determining the predicted labels of each preset classifier for each sample in the current sample set to be corrected, as well as the integrated sample labels of each sample in the current sample set to be corrected, may include: predicting the sample type of each sample in the current sample set to be corrected based on each preset classifier, to obtain the predicted label of each sample in the current sample set to be corrected; and determining the corresponding integrated sample label according to the predicted label of each sample in the current sample set to be corrected using the majority voting rule.

[0071] In an embodiment of the present invention, each sample in the current sample set to be corrected can be synchronously input into multiple preset classifiers, and the sample type of each sample in the current sample set to be corrected can be predicted by the multiple preset classifiers to obtain the predicted labels of each sample in the current sample set to be corrected, and then the predicted labels of the samples with the most votes in the current sample set to be corrected can be determined by using the majority voting rule, and the predicted labels of the samples with the most votes can be used as the integrated sample labels of the samples. The integrated sample labels of the samples can be determined by the majority voting rule, and the comprehensive evaluation of the prediction results of multiple classifiers can be achieved under the premise of ensuring low computing resource consumption.

[0072] For example, the identification of noise samples refers to the voting filtering method. M sub-classifiers are established on the training set to perform K-fold cross-validation. In each cross-validation, K-1 folds are used as training data, and the Kth fold is used as test data to predict the sample label of the Kth fold. Each sample generates M predicted labels. According to the majority voting rule, more than If the predicted label is inconsistent with the original label of the sample, the sample is considered to be a potential noise sample, otherwise it is a noise-free sample.

[0073] Figure 3 A recognition logic diagram of a noise sample provided in the second embodiment of the present invention, the input sample set is Data, 3 classifiers are selected, and the data set is randomly divided into 10 parts, 9 of which are used as training sets and the remaining 1 is used as a validation set (i.e., a test set). Assume that there are 5 samples in the validation set, 2 of which have original labels of 0 and 3 of which have labels of 1. The training set is trained on each preset classifier, and then the 5 samples are predicted respectively. Each sample will get 3 predicted labels. Samples 1 and 3 are both correctly classified and are therefore considered to be noise-free samples; sample 2 is completely misclassified and is considered to be a noise sample, indicated in red; sample 4 is misclassified twice and correctly classified once. According to the majority voting rule, it is also considered to be a noise sample, also indicated in red. Sample 5 is correctly classified twice and misclassified once and is judged to be a noise-free sample. After the first step is completed, the input data set Data is divided into two parts: a noise-free sample set XU and a potential noise sample set XN.

[0074] In an optional embodiment of the present invention, determining the sample importance index of each sample in the current sample set to be corrected based on the number of label prediction differences and the information entropy of each sample in the current sample set to be corrected can include: normalizing the number of label prediction differences of each sample in the current sample set to be corrected and the information entropy of each sample in the current sample set to be corrected to obtain the first data to be weighted and the second data to be weighted of each sample in the current sample set to be corrected; calculating the weighted sum of the first data to be weighted and the second data to be weighted of each sample in the current sample set to be corrected to obtain the sample importance index of each sample in the current sample set to be corrected.

[0075] The first data to be weighted may be a normalized result of the number of label prediction differences, and the second data to be weighted may be a normalized result of the information entropy of the samples in the current sample set to be corrected.

[0076] In an embodiment of the present invention, the number of label prediction differences of each sample in the current sample set to be corrected can be normalized based on the total number of samples in the current sample set to be corrected to obtain the first weighted data, and the information entropy of each sample in the current sample set to be corrected can be normalized based on the maximum information entropy and the minimum information entropy of the samples in the current sample set to be corrected to obtain the second weighted data of each sample in the current sample set to be corrected, and then for each sample in the current sample set to be corrected, the first weighted data and the second weighted data are weighted and summed to obtain the sample importance index of each sample in the current sample set to be corrected. In the above scheme, the sample dimensions are unified through normalization processing, and the sample importance index is determined based on the weighted method, which can conveniently adjust the weight of the part that affects the sample importance index, giving users stronger operability.

[0077] Step 250: Filter out key samples to be corrected from the current set of samples to be corrected according to the sorting data of the samples to be corrected and the preset correction ratio.

[0078] The preset correction ratio may be a preset sample selection ratio, that is, the ratio of key samples to be corrected to the total number of samples in the current sample set to be corrected.

[0079] In the embodiment of the present invention, based on the sorting data of samples to be corrected, the samples in the current sample set to be corrected can be sorted from high to low according to importance, and based on a preset correction ratio, the samples with the highest sorting can be screened out as key samples to be corrected.

[0080] Step 260: Based on the expert-corrected sample labels, the sample labels of the key samples to be corrected are corrected to obtain target corrected samples, and the current noise pool samples are updated according to the samples in the current set of samples to be corrected except the key samples to be corrected.

[0081] The expert-corrected sample labels may be sample labels of key samples to be corrected that are re-labeled by experts.

[0082] In an embodiment of the present invention, the sample labels of the key samples to be corrected can be handed over to experts for re-labeling to obtain expert-corrected sample labels, so as to complete the correction of the sample labels of the key samples to be corrected based on the expert-corrected sample labels to obtain the target corrected samples, and then update the current noise pool samples according to the samples in the current sample set to be corrected except the key samples to be corrected.

[0083] Step 270: Update the current round input sample set according to the current noise-free sample set and the target correction sample, and return to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

[0084] In a specific example, the method framework for noise sample processing is as follows Figure 4 As shown in the figure, assuming the original sample set of the original financial scenario (abbreviated as X), firstly, X is divided into two parts, namely, a noise-free sample set (abbreviated as XU) and a potential noise sample set (abbreviated as XN) through a filter (i.e., a preset classifier). Then, based on the noise-free sample set, important samples are selected from the potential noise sample set and handed over to experts for re-labeling to obtain the correct label of the sample. The noise-free sample set predicted in each round is combined with the important samples after label correction, i.e., the key samples to be corrected (abbreviated as XR) as the output new sample set of the new round (abbreviated as X_new). The next round of iteration is performed on the new sample set. Compared with the sample set of the original financial scenario, the part of samples not included in the new sample set is added to the noise pool (abbreviated as X-X_new), and together with the potential noise sample set of the next round, it is used as the test set of the active learning step, from which the key samples to be corrected in the next round are selected. As the number of iterations increases, the noise recognition ability of the preset classifier continues to improve, that is, the actual number of noises included in the noise-free sample set is getting smaller and smaller, and the samples that were not selected in the previous round of active learning are not absolutely unimportant. These samples should participate in the next round of active learning.

[0085] Figure 5 A schematic diagram of a sample iteration processing process provided in the second embodiment of the present invention, assuming that a total of f iterations are performed, the original sample set of the original financial scenario is recorded as X_org, and the output data set is X_new. In the first iteration, M preset classifiers are established in X_org to perform K-fold cross validation to predict each sample, and noise filtering is performed based on the predicted results. After filtering, X_org is divided into two parts: a noise-free sample set XU(1) and a potential noise sample set XN(1) (indicated in red); the learning step of active learning uses XU(1) as the training set, establishes M preset classifiers to predict the samples in XN(1), and calculates the sample importance index r of each sample in XN(1). (i) And according to r (i) Sorting is performed, and the labels of the first q proportion of samples in XN(1) are queried and corrected. At this time, XN(1) is divided into two parts: the corrected important sample set, i.e., the key samples to be corrected, is XR(1) (green part) and the remaining unselected samples. The unselected sample set enters the noise pool noise_pool(1) (orange part). XR(1) and XU(1) become the output new sample set X_new(1) (in yellow) after the first round of iteration.

[0086] The second iteration of filtering is performed on the new sample set X_new(1) output from the first iteration. Compared with X_org, the new data set lacks samples from the noise pool. After filtering, X_new(1) is divided into two parts, XU(2) and XN(2). It should be noted that if there are overlapping samples in XN(2) and XR(1), the sample should be deleted from XN(2) and added to XU(2), because the XR(1) set is a sample that has been verified by experts to confirm that the label is correct. In fact, at this time, the sum of XU(2), XN(2) and noise_pool(1) is X_org. The learning step of active learning is based on XU(2) as the training set, but the test set at this time is the samples in XN(2) and noise_pool(1). The same calculation is used for r (n) After sorting, the first q proportion of samples are queried and corrected. The corrected sample set is XR(2). XR(2) and XU(2) are the output sample set X_new(2) after the second round of iteration, and all other samples (the part where X_org is larger than X_new(2)) are added to the noise pool noise_pool(2). Each subsequent iteration is the same as the second iteration, which will not be described here. After f iterations, the final output sample set is X_new(f).

[0087] Active learning is a cyclic process. The model is trained on a small number of labeled samples. The most important samples are selected from the unlabeled sample set according to certain criteria and handed over to experts for manual labeling. The model is retrained with the new knowledge obtained for the next round of queries. The potential noise sample set identified in the first step is regarded as an unlabeled sample set, and the sample set containing label noise is regarded as a labeled sample set. The active learning and noise sample identification processes can be naturally combined. M preset classifiers are established on the predicted noise-free sample set, and each sample n in the noise sample set is predicted. The original label of each sample is y. n , generate predicted labels With the predicted probability p mnc , n=1,2,…N represents the sample number of the noise sample set, m=1,2,…,M represents the preset classifier number, c represents the number of categories, c=2 in the binary classification problem, and the query function r (n) is based on With p mnc Calculated. According to r (n) After sorting the noise samples, the samples in the top q proportion are selected as important samples for manual inspection. The sample set after manual inspection is recorded as XR.

[0088]

[0089] Predict labels for an ensemble of M preset classifiers;

[0090]

[0091] α is a hyperparameter; Bm (n) Indicates the number of times the predicted labels of each base classifier are inconsistent with the ensemble predicted labels, Bm (n) The larger the value, the more controversial the sample is, the more difficult it is to classify correctly, and the higher its value. (n) Information entropy represents the degree of uncertainty, Be (n) The larger the value is, the greater the probability that the sample belongs to the noise sample; (n) It's Bm (n) With Be (n) The normalized weighted sum, r (n) The larger the value, the more important the sample is.

[0092] For example, assume that the samples labeled as noise in the first step are the second and fourth samples, the current noise pool contains a sample numbered 6, and the noise-free sample set XU is used as a training set to train on three classifiers. The noise sample set XN and each sample in the current noise pool are predicted. Similarly, each sample will get 3 predicted labels and 3 sets of prediction probabilities p mn For the second sample, the original label category is 0, and the predicted labels of the three sub-classifiers are (For the sake of convenience, sample number 2 has n=1, sample number 4 has n=2, and sample number 6 has n=3), and the corresponding predicted probabilities are p 11 =(0.1, 0.9), p 21 =(0.2, 0.8), p 31 =(0.3, 0.7), where p 111 =0.1, p 112 =0.9, indicating that the probability that the first classifier predicts that the first sample belongs to the first category (0) is 0.1, and the probability that the first classifier predicts that the first sample belongs to the second category (1) is 0.9; for the fourth sample, the original label category is 1, and the predicted labels of the three sub-classifiers are The corresponding predicted probabilities are p 12 =(0.7, 0.3), p 22 =(0.55, 0.45), p 32 =(0.4, 0.6); For the sixth sample, the original label category is 1, and the predicted labels of the three sub-classifiers are The corresponding predicted probabilities are p 13 =(0.6, 0.4), p23 =(0.2, 0.8), p 33 =(0.32, 0.68). According to the majority voting classification rule, the integrated prediction label of sample 2 is Ensemble prediction label for sample 4 Ensemble predicted label for sample 6 Based on this, the Bm of each sample is calculated based on the above formula. (n) With Be (n) , we get r (2) >r (3) >r (1) .

[0093] When (2) >r (3) >r (1) When , the sample importance ranking is: sample 4> sample 6> sample 2. Sample 4 is selected and handed over to the manual expert for inspection. After the inspection, the label of sample 4 is corrected from 1 to 0. At this time, the sample set XR includes the sample 4 after manual inspection. The sample set XR and the noise-free sample set XU constitute the new data set after this learning, while samples 2 and 6 enter the noise pool to participate in the next round of active learning process.

[0094] It should be noted that the number of cross-validation times K, the query ratio q, the weight coefficient α, and the threshold ε set in the sample optimization iteration cutoff condition are all hyperparameters involved in this algorithm and can be adjusted according to the data set situation.

[0095] This scheme combines active learning to propose a new iterative framework. As the number of iterations increases, the classifier's ability to identify potential noise samples gradually increases, improving the accuracy of noise sample identification. The identified potential noise samples are not deleted directly, but the important samples are selected through active learning methods and handed over to experts for inspection and then added to the predicted noise-free sample set to generate a new sample set. On the one hand, the noise content in the sample set can be quickly reduced, and on the other hand, important samples are retained to avoid information loss.

[0096] In the active learning process, the definition of uncertainty samples is expanded. The focus is not on finding noise samples, but on screening out samples with higher values ​​for testing to prevent the loss of important information.

[0097] The technical solution of the embodiment of the present invention obtains the current round input sample set, thereby predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, and then uses the current noise-free sample set as the training set, the current potential noise sample set and the current noise pool sample as the test set for active learning to obtain the current sample set to be corrected, and sorts the samples in the current sample set to be corrected according to the sample importance index to obtain the sorted data of the samples to be corrected, and further selects the key samples to be corrected from the current sample set to be corrected according to the sorted data of the samples to be corrected and the preset correction ratio, thereby correcting the sample labels of the key samples to be corrected based on the expert correction sample labels to obtain the target correction samples, and updates the current noise pool samples according to the samples other than the key samples to be corrected in the current sample set to be corrected, and then updates the current round input sample set according to the current noise-free sample set and the target correction samples, and returns to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, until the sample optimization iteration cutoff condition is met, and the noise processing target sample is output. This solution integrates artificial experts with active learning ideas, selects key samples to be corrected based on sample importance indicators and preset correction ratios, and uses the labels of key samples to be corrected given by domain experts to effectively improve the accuracy of sample labels. It also implements iterative correction of sample sets based on autonomous learning, which can not only quickly reduce the noise content in the sample set, but also avoid information loss of key samples. On the basis of filtering out unfavorable noise samples, it expands the clean sample set, effectively improves the quality of sample sets in financial scenarios, ensures low noise content in sample set labels while reducing verification costs, and makes the final prediction effect of the model trained on processed samples close to that of the model trained on noise-free samples, solving the problems of poor time performance and high verification cost in sample data denoising in the financial field. It can reduce the verification cost of samples and improve the time performance of sample noise processing while ensuring low noise rate of samples.

[0098] Embodiment 3

[0099] Figure 6 This is a structural schematic diagram of a noise sample processing device provided in Embodiment 3 of the present invention.

[0100] like Figure 6 As shown, the device comprises:

[0101] The sample acquisition module 310 is used to obtain the input sample set of the current round;

[0102] The sample classification module 320 is used to predict the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set;

[0103] An active learning module 330 is used to perform active learning using the current noise-free sample set as a training set, the current potential noise sample set and the current noise pool sample as a test set to obtain a current sample set to be corrected;

[0104] The sample correction and noise pool updating module 340 is used to determine the target correction sample of the key sample to be corrected in the current sample set to be corrected, and update the current noise pool sample according to the samples other than the key sample to be corrected in the current sample set to be corrected;

[0105] The sample processing iteration module 350 is used to update the current round input sample set according to the current noise-free sample set and the target correction sample, and return to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

[0106] The technical solution of the embodiment of the present invention obtains the current round input sample set, thereby predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, and then uses the current noise-free sample set as the training set, and the current potential noise sample set and the current noise pool sample as the test set for active learning to obtain the current sample set to be corrected, further determines the target correction sample of the key sample to be corrected in the current sample set to be corrected, and updates the current noise pool sample according to the samples other than the key sample to be corrected in the current sample set to be corrected, so as to update the current round input sample set according to the current noise-free sample set and the target correction sample, and returns to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met, and the noise processing target sample is output. This solution integrates the idea of ​​active learning and iteratively corrects the sample set, which can not only quickly reduce the noise content in the sample set, but also avoid information loss of key samples. On the basis of filtering out unfavorable noise samples, it expands the clean sample set, effectively improves the quality of sample sets in financial scenarios, ensures low noise rate of sample set labels while reducing verification costs, so that the final prediction effect of the model trained on processed samples is close to that of the model trained on noise-free samples, solves the problems of poor time performance and high verification cost of sample data denoising in the financial field, and can reduce the verification cost of samples and improve the time performance of sample noise processing while ensuring low noise rate of samples.

[0107] Optionally, the sample correction and noise pool update module 340 is specifically used to sort the samples in the current sample set to be corrected according to the sample importance index to obtain the sorting data of the samples to be corrected; screen out the key samples to be corrected from the current sample set to be corrected according to the sorting data of the samples to be corrected and a preset correction ratio; and correct the sample labels of the key samples to be corrected based on the expert correction sample labels to obtain the target corrected samples.

[0108] Optionally, the noise sample processing device also includes a sample importance index determination module, which is used to determine the predicted labels of each sample in the current sample set to be corrected by each preset classifier, as well as the integrated sample label of each sample in the current sample set to be corrected; count the number of times the predicted label of each sample in the current sample set to be corrected is inconsistent with the corresponding integrated sample label, and obtain the number of label prediction differences of each sample in the current sample set to be corrected; determine the sample importance index of each sample in the current sample set to be corrected based on the number of label prediction differences and the information entropy of each sample in the current sample set to be corrected.

[0109] Optionally, the noise sample processing device also includes a sample deduplication module, which is used to eliminate the previous round of target correction samples in the current potential noise sample set if there is sample overlap between the current potential noise sample set and the previous round of target correction samples; and add the previous round of target correction samples to the current noise-free sample set.

[0110] Optionally, a sample importance index determination module is specifically used to predict the sample type of each sample in the current sample set to be corrected based on each of the preset classifiers, and obtain the predicted label of each sample in the current sample set to be corrected; according to the predicted label of each sample in the current sample set to be corrected, the corresponding integrated sample label is determined using the majority voting rule.

[0111] Optionally, a sample importance index determination module is specifically used to normalize the number of label prediction differences of each sample in the current sample set to be corrected and the information entropy of each sample in the current sample set to be corrected to obtain the first data to be weighted and the second data to be weighted of each sample in the current sample set to be corrected; calculate the weighted sum of the first data to be weighted and the second data to be weighted of each sample in the current sample set to be corrected to obtain the sample importance index of each sample in the current sample set to be corrected.

[0112] The noise sample processing device provided in the embodiment of the present invention can execute the noise sample processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0113] Embodiment 4

[0114] Figure 7 A schematic diagram of an electronic device that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0115] like Figure 7 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0116] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0117] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as a noise sample processing method.

[0118] In some embodiments, the noise sample processing method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the noise sample processing method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the noise sample processing method in any other appropriate manner (e.g., by means of firmware).

[0119] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0120] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0121] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, 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), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0122] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0123] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0124] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of traditional physical hosts and VPS servers, which are difficult to manage and have weak business scalability.

[0125] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0126] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A noise sample processing method, characterized in that: include: Get the current round input sample set; Predicting the current round of input sample set based on multiple preset classifiers to obtain a current noise-free sample set and a current potential noise sample set; Active learning is performed using the current noise-free sample set as a training set, the current potential noise sample set and the current noise pool sample as a test set to obtain a current sample set to be corrected; Determine a target correction sample for a key sample to be corrected in the current set of samples to be corrected, and update the current noise pool sample according to samples other than the key sample to be corrected in the current set of samples to be corrected; According to the current noise-free sample set and the target correction sample, the current round input sample set is updated, and the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set is returned until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

2. The method according to claim 1, characterized in that: The step of determining a target correction sample of a key sample to be corrected in the current set of samples to be corrected comprises: According to the sample importance index, the samples in the current sample set to be corrected are sorted to obtain the sorting data of the samples to be corrected; According to the sorting data of the samples to be corrected and the preset correction ratio, the key samples to be corrected are screened out from the current set of samples to be corrected; Based on the expert-corrected sample labels, the sample labels of the key samples to be corrected are corrected to obtain the target corrected samples.

3. The method according to claim 2, characterized in that Before sorting the samples in the current sample set to be corrected according to the sample importance index to obtain the sorting data of the samples to be corrected, the method further includes: Determine each predicted label of each sample in the current sample set to be corrected by each of the preset classifiers, and an integrated sample label of each sample in the current sample set to be corrected; Counting the number of times that the predicted label of each sample in the current sample set to be corrected is inconsistent with the corresponding integrated sample label, to obtain the number of label prediction differences of each sample in the current sample set to be corrected; According to the number of label prediction differences and the information entropy of each sample in the current sample set to be corrected, a sample importance index of each sample in the current sample set to be corrected is determined.

4. The method according to claim 1, characterized in that: After predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set, the method further includes: If the current potential noise sample set has sample overlap with the previous round of target correction samples, then the previous round of target correction samples in the current potential noise sample set are removed; The target correction samples of the previous round are added to the current noise-free sample set.

5. The method according to claim 3, characterized in that: The determining of each predicted label of each sample in the current to-be-corrected sample set by each preset classifier and the integrated sample label of each sample in the current to-be-corrected sample set includes: Based on each of the preset classifiers, respectively predict the sample type of each sample in the current sample set to be corrected, to obtain a predicted label for each sample in the current sample set to be corrected; According to the predicted label of each sample in the current sample set to be corrected, the corresponding integrated sample label is determined by using the majority voting rule.

6. The method according to claim 3, characterized in that Determining the sample importance index of each sample in the current sample set to be corrected according to the number of label prediction differences and the information entropy of each sample in the current sample set to be corrected includes: Normalizing the number of label prediction differences of each sample in the current sample set to be corrected and the information entropy of each sample in the current sample set to be corrected to obtain first weighted data and second weighted data of each sample in the current sample set to be corrected; A weighted sum of the first data to be weighted and the second data to be weighted of each sample in the current sample set to be corrected is calculated to obtain a sample importance index of each sample in the current sample set to be corrected.

7. A noise sample processing device, characterized in that: include: The sample acquisition module is used to obtain the input sample set of the current round; A sample classification module, used for predicting the current round input sample set based on multiple preset classifiers to obtain a current noise-free sample set and a current potential noise sample set; An active learning module, used to perform active learning using the current noise-free sample set as a training set, the current potential noise sample set and the current noise pool sample as a test set, to obtain a current sample set to be corrected; A sample correction and noise pool updating module, used to determine a target correction sample of a key sample to be corrected in the current sample set to be corrected, and to update the current noise pool sample according to samples other than the key sample to be corrected in the current sample set to be corrected; The sample processing iteration module is used to update the current round input sample set according to the current noise-free sample set and the target correction sample, and return to execute the operation of predicting the current round input sample set based on multiple preset classifiers to obtain the current noise-free sample set and the current potential noise sample set until the sample optimization iteration cutoff condition is met and the noise processing target sample is output.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the noise sample processing method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the noise sample processing method according to any one of claims 1 to 6 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the noise sample processing method according to any one of claims 1 to 6.

Images