Model debiased dataset augmentation method

Abstract

The application discloses a model debiasing dataset enhancement method, comprising the following steps: obtaining a dataset; constructing a bias model and training the bias model by using part of data in the dataset, so as to obtain the bias degree corresponding to each sample in the dataset by using the trained bias model; arranging each sample in the dataset according to the bias degree, and equally dividing each arranged sample into multiple groups, each group corresponding to a bias indicator, so as to obtain a dataset with a bias indicator; constructing a generator and training the generator by using the dataset with the bias indicator, so as to generate pseudo samples with low bias degree by using the trained generator; expanding the original dataset by using the pseudo samples with low bias degree, so as to obtain an expanded dataset; iteratively generating pseudo samples, so as to obtain a final dataset; thus, the purpose of relieving data bias is achieved, and the consumption of artificial resources is avoided.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a dataset augmentation method for model bias removal. Background Technology

[0002] In related technologies, due to limitations in dataset collection methods and the experience of annotation personnel, the constructed training datasets suffer from distribution bias. The model may use the bias in the dataset as a shortcut for prediction, resulting in the model not learning true understanding and reasoning abilities, leading to poor generalization and robustness. To address this, the dataset needs to be manually analyzed beforehand to define possible bias features, and then data containing bias features should be removed from the generated pseudo-samples. This manual analysis limits the model's generalization and increases the consumption of human resources. Summary of the Invention

[0003] This invention aims to at least partially solve one of the technical problems in the aforementioned technologies. To this end, one objective of this invention is to propose a dataset augmentation method for model debiasing, which improves the data distribution of the dataset through iterative data augmentation, thereby alleviating data bias and avoiding the waste of human resources.

[0004] A second objective of this invention is to provide a computer-readable storage medium.

[0005] The third objective of this invention is to provide a computer device.

[0006] To achieve the above objectives, a first aspect of the present invention proposes a dataset augmentation method for model bias removal, comprising the following steps:

[0007] S1, Obtain the dataset;

[0008] S2, construct a bias model and train the bias model using a portion of the data in the dataset to obtain a trained bias model;

[0009] S3, Input the dataset into the trained bias model in order to obtain the bias degree corresponding to each sample in the dataset;

[0010] S4, each sample in the dataset is arranged according to the bias degree, and each sorted sample is divided into multiple groups. Each group is connected with a bias indicator to obtain a dataset with bias indicators.

[0011] S5, Construct a generator and train the generator using the dataset with the bias index so as to generate pseudo samples with low bias through the trained generator.

[0012] S6, the original dataset is augmented using the pseudo-samples with low bias to obtain the augmented dataset;

[0013] Repeat steps S3-S6 using the amplified dataset as a new dataset until the maximum number of iterations is reached to obtain the final dataset.

[0014] According to an embodiment of the present invention, the dataset augmentation method for model debiasing firstly acquires a dataset; then, a bias model is constructed and trained using a portion of the dataset to obtain the bias degree corresponding to each sample in the dataset; next, each sample in the dataset is arranged according to the bias degree, and each sorted sample is divided into multiple groups, with each group corresponding to a bias indicator, to obtain a dataset with bias indicators; then, a generator is constructed and trained using the dataset with bias indicators to generate pseudo-samples with low bias degrees; finally, the original dataset is augmented using pseudo-samples with low bias degrees to obtain an augmented dataset; pseudo-sample generation is iteratively performed to obtain the final dataset; thus, the data distribution of the dataset is improved through iterative data augmentation, achieving the purpose of alleviating data bias and avoiding the consumption of manual resources.

[0015] In addition, the dataset augmentation method for model debiasing proposed in the above embodiments of the present invention may also have the following additional technical features:

[0016] Optionally, each sample in the dataset includes sample information and a corresponding real label. The trained bias model uses the probability of the sample information on the corresponding real label as the basis for evaluating the degree of sample bias. If the probability is closer to 1, the sample is more likely to be a biased sample.

[0017] Optionally, the generator is trained using the dataset with bias indicators to generate pseudo-samples with low bias. This includes: inputting the dataset with bias indicators into the generator, which gradually learns the association between the bias indicator and the samples to obtain a trained generator; and inputting a bias indicator with low bias and a batch of random real labels into the trained generator to obtain corresponding pseudo-samples with low bias.

[0018] Alternatively, the generator can be optimized using the following loss function:

[0019]

[0020] Where S represents the dataset; x (i) b represents the information of the i-th sample;(i) Indicates the i-th bias indicator; y (i) θ represents the i-th real label; g This indicates a generator.

[0021] To achieve the above objectives, a second aspect of the present invention provides a computer-readable storage medium storing a model-biased dataset augmentation program, which, when executed by a processor, implements the model-biased dataset augmentation method as described above.

[0022] According to the computer-readable storage medium of the present invention, a dataset augmentation program for model debiasing is stored. When the dataset augmentation program for model debiasing is executed by the processor, it implements the dataset augmentation method for model debiasing as described above. Thus, the data distribution of the dataset is improved through iterative data augmentation, thereby alleviating data bias and avoiding the waste of human resources.

[0023] To achieve the above objectives, a third aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the dataset augmentation method for model debiasing as described above.

[0024] According to the computer device of the present invention, the processor executes the above-described model debiasing dataset augmentation program, thereby improving the data distribution of the dataset through iterative data augmentation, achieving the purpose of alleviating data bias, and avoiding the consumption of human resources. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a dataset augmentation method for model debiasing according to an embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of an iterative data augmentation framework according to an embodiment of the present invention. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0028] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0029] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0030] Figure 1 This is a flowchart illustrating a dataset augmentation method for model bias removal according to an embodiment of the present invention, as shown below. Figure 1 As shown, the dataset augmentation method for bias removal in this model includes the following steps:

[0031] S1, Obtain the dataset.

[0032] It should be noted that the dataset S includes multiple samples, and the i-th sample is (x (i) ,y (i) ), where y (i) It is sample x (i) The true label.

[0033] S2. Construct a bias model and train it using a portion of the dataset to obtain a trained bias model.

[0034] It should be noted that since bias models obtained using a small portion of the dataset tend to overfit the bias features, it is easy to assign a high confidence level to data containing bias features.

[0035] S3, input the dataset into the trained bias model in order to obtain the bias degree corresponding to each sample in the dataset.

[0036] In other words, a bias model θ is trained using a small portion of the dataset. s This is so that it can be used to measure the degree of bias in the sample.

[0037] It should be noted that the trained bias model is based on the sample information x. (i) In the corresponding real label y (i) The probability p is used as a basis for assessing the degree of sample bias. The closer the probability p is to 1, the more biased the sample (x) is. (i) ,y (i) The more likely it is to be a biased sample.

[0038] S4. Arrange each sample in the dataset according to the degree of bias, and divide each sorted sample into multiple groups. Each group corresponds to a bias indicator to obtain a dataset with bias indicators.

[0039] As an example, such as Figure 2 As shown in the first step, each sample in the dataset is sorted in ascending order according to the degree of bias; and the dataset S is divided into N equal groups, each group having a corresponding bias indicator b. i Where b1 represents the indicator with the lowest bias, b N The indicator representing the highest degree of bias.

[0040] S5. Construct a generator and train it using a dataset with a bias metric so that the trained generator can generate pseudo-samples with low bias.

[0041] As an example, a dataset with bias indicators is used to train the generator so as to generate pseudo samples with low bias. This includes: inputting the dataset with bias indicators into the generator, which gradually learns the association between the bias indicator and the samples to obtain a trained generator; and inputting a bias indicator with low bias and a batch of random real labels into the trained generator to obtain corresponding pseudo samples with low bias.

[0042] In other words, such as Figure 2 As shown in the second step, the sample fine-tuning generator θ in the dataset S is used. g Where, the given is the target label y (i) and bias indicator b i The requirement is to generate the corresponding text x. (i) The loss function is shown below and is used to optimize the generator θ. g Generator θ g In this process, the correlation between the bias indicator and the sample can be gradually learned:

[0043]

[0044] Where S represents the dataset; x (i) b represents the information of the i-th sample; (i) Indicates the i-th bias indicator; y (i) θ represents the i-th real label; g This indicates a generator.

[0045] In addition, such as Figure 2As shown in the third step, given a bias indicator b with a low bias degree, and a batch of labels y are randomly given, the generator is required to generate a batch of pseudo samples, so as to obtain a batch of pseudo samples with a low bias degree.

[0046] S6. The original dataset is augmented using pseudo-samples with low bias to obtain the augmented dataset.

[0047] It should be noted that, as Figure 2 As shown in step four, after obtaining the pseudo-samples, they are added back to the original dataset S, and the above steps are repeated until the maximum number of iterations is reached; then, the natural language understanding model is trained using the augmented dataset, and the impact of data bias is mitigated by changing the data distribution.

[0048] In other words, the augmented dataset is used as a new dataset, and steps S3-S6 are repeated until the maximum number of iterations is reached in order to obtain the final dataset.

[0049] In summary, the dataset augmentation method for model debiasing according to embodiments of the present invention, in order to reduce the cost of manual data analysis, automatically measures the bias degree of each sample by training a bias model without manual intervention; to enable the generator to learn the relationship between the bias degree and sample features, samples are grouped according to the bias degree, each sample is concatenated with a corresponding bias indicator, and the generator is trained; given a bias indicator and a specific label, the generator generates a batch of pseudo-samples to augment the original dataset, and then regroups them according to the bias degree; the above process is repeated multiple times, and the final dataset is used to train the model for downstream tasks. Thus, through multiple iterations, the generating model can be made more biased towards the desired features, resulting in higher data quality.

[0050] To implement the above embodiments, this invention also proposes a computer-readable storage medium storing a model-biased dataset augmentation program, which, when executed by a processor, implements the model-biased dataset augmentation method as described above.

[0051] According to the computer-readable storage medium of the present invention, a dataset augmentation program for model debiasing is stored. When the dataset augmentation program for model debiasing is executed by the processor, it implements the dataset augmentation method for model debiasing as described above. Thus, the data distribution of the dataset is improved through iterative data augmentation, thereby alleviating data bias and avoiding the waste of human resources.

[0052] To implement the above embodiments, this invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the dataset augmentation method for model debiasing as described above.

[0053] According to the computer device of the present invention, the processor executes the above-described model debiasing dataset augmentation program, thereby improving the data distribution of the dataset through iterative data augmentation, achieving the purpose of alleviating data bias, and avoiding the consumption of human resources.

[0054] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0055] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0056] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0057] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0058] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0059] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0060] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0061] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0062] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0063] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0064] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0065] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A dataset augmentation method for model bias removal, characterized in that, Includes the following steps: S1, Obtain the dataset; S2, construct a bias model and train the bias model using a portion of the data in the dataset to obtain a trained bias model; S3, Input the dataset into the trained bias model in order to obtain the bias degree corresponding to each sample in the dataset; S4, each sample in the dataset is arranged according to the bias degree, and each sorted sample is divided into multiple groups. Each group is connected with a bias indicator to obtain a dataset with bias indicators. S5, Construct a generator and train the generator using the dataset with the bias index so as to generate pseudo samples with low bias through the trained generator. S6, the dataset is augmented using the pseudo-samples with low bias to obtain the augmented dataset; Repeat steps S3-S6 using the amplified dataset as a new dataset until the maximum number of iterations is reached to obtain the final dataset.

2. The dataset augmentation method for model bias correction as described in claim 1, characterized in that, Each sample in the dataset includes sample information and a corresponding real label. The trained bias model uses the probability of the sample information on the corresponding real label as the basis for evaluating the degree of sample bias. If the probability is closer to 1, the sample is more likely to be a biased sample.

3. The dataset augmentation method for model bias removal as described in claim 2, characterized in that, The generator is trained using the dataset with the bias index to generate pseudo-samples with low bias, including: A dataset with bias indicators is input into a generator, which gradually learns the correlation between the bias indicators and the samples in order to obtain a trained generator. A bias indicator with a low bias degree and a batch of random real labels are input into the trained generator to obtain corresponding pseudo samples with a low bias degree.

4. The dataset augmentation method for model bias removal as described in claim 1, characterized in that, The generator is optimized using the following loss function: Where S represents the dataset; x (i) b represents the information of the i-th sample; (i) Indicates the i-th bias indicator; y (i) θ represents the i-th real label; g This indicates a generator.

5. A computer-readable storage medium, characterized in that, It stores a model-biased dataset augmentation program, which, when executed by a processor, implements the model-biased dataset augmentation method as described in any one of claims 1-4.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the dataset augmentation method for model debiasing as described in any one of claims 1-4.

Images