A data quality detection method and device based on federated learning

Abstract

The application is suitable for the field of computers and provides a data quality detection method and device based on joint learning, wherein the method comprises the following steps: obtaining a total test data set based on test data sets of each participating device; obtaining a plurality of training data sets based on local data sets of each participating device; performing joint learning by using the training data sets and a model issued by a coordinating device to obtain a local model; obtaining a first evaluation parameter and a second evaluation parameter after processing the test data sets and the total test data set by using the local model; and detecting data quality by using the first evaluation parameter and the second evaluation parameter. The application detects the data quality corresponding to each participating device by calculating the evaluation parameters of each participating device, thereby solving the problem of poor overall joint learning effect caused by poor data quality of participating devices in horizontal joint learning.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of computer, and particularly relates to a data quality detection method and device based on federated learning. BACKGROUND

[0002] Federated learning is a distributed learning paradigm based on distributed data sets, aiming to protect the data privacy of data owners in the model training and inference process. Horizontal federated learning is a distributed structure of federated learning, in which the data characteristics of each distributed node are the same, and the sample spaces are different. In the horizontal federated learning algorithm, each participant calculates the model update information according to the local data, and one party or a third party aggregates the model update information and finally calculates the final model. If the data quality of one or more participants is poor, such as large data noise and many missing data points, it will have a negative effect on the overall federated learning. SUMMARY

[0003] Therefore, the present application provides a data quality detection method and device based on federated learning to solve the problem of poor overall federated learning effect caused by poor data quality of one or more participants in horizontal federated learning in the prior art.

[0004] In a first aspect of the embodiments of the present application, a data quality detection method based on federated learning is provided, which is applied to a participating device, the participating device is in communication connection with a coordinating device or other participating devices, and the method comprises the following steps:

[0005] Obtaining a test data set of each participating device and sending it to the coordinating device to obtain a total test data set;

[0006] Obtaining a local data set of each participating device and splitting the local data set into a plurality of mutually disjoint data sets to obtain a plurality of training data sets corresponding to each participating device;

[0007] Performing federated learning using the plurality of training data sets and a model issued by the coordinating device to obtain a local model corresponding to each training data set;

[0008] Processing the test data set and the total test data set using the local model to obtain a first evaluation parameter and a second evaluation parameter of each participating device;

[0009] Detecting the data quality corresponding to each participating device using the first evaluation parameter and the second evaluation parameter.

[0010] In some embodiments, obtaining a test data set of each participating device and sending it to the coordinating device to obtain a total test data set specifically comprises:

[0011] The local data of each participating device is proportionally sampled to obtain a test data set of each participating device;

[0012] The test data set of each participating device is sent to the coordination device to obtain a total test data set.

[0013] In some embodiments, after the local model is used to process the test data set and the total test data set, a first evaluation parameter and a second evaluation parameter of each participating device are obtained, specifically including:

[0014] The total test data set is removed from each test data set to obtain a detection data set corresponding to each participating device;

[0015] The detection data set is input into the local model corresponding to the detection data set to obtain a first prediction value;

[0016] After noise processing of the first prediction value, a first evaluation parameter is obtained.

[0017] In some embodiments, after noise processing of the prediction value, a first evaluation parameter is obtained, and further including:

[0018] Each test data set is input into the local model corresponding to each test data set to obtain a second prediction value;

[0019] After noise processing of the second prediction value, a second evaluation parameter is obtained.

[0020] In some embodiments, the first evaluation parameter and the second evaluation parameter are used to detect the data quality corresponding to each participating device, specifically including:

[0021] The first evaluation parameter and the second evaluation parameter corresponding to each participating device are weighted to obtain a total evaluation parameter corresponding to each participating device;

[0022] The total evaluation parameter is used to detect the data quality corresponding to each participating device.

[0023] In some embodiments, the calculation formula of the total evaluation parameter corresponding to each participating device is:

[0024]

[0025] Wherein, k is the label of the participating device, η k is the total evaluation parameter of participating device k, λ is the weight coefficient, the value range of λ is 0-1, is the first evaluation parameter of participating device k, is the second evaluation parameter of participating device k.

[0026] In some embodiments, after detecting the data quality corresponding to each participating device by using the overall evaluation parameter, the method further comprises:

[0027] determining whether the overall evaluation parameter of each participating device meets a preset condition;

[0028] if yes, screening the local data set of the participating device meeting the data quality requirement.

[0029] In a second aspect, the embodiment of the present application provides a data quality detection device based on federated learning, comprising:

[0030] an overall test data set obtaining module configured to obtain test data sets of each participating device and send the test data sets to a coordinating device to obtain an overall test data set;

[0031] a training data set obtaining module configured to obtain local data sets of each participating device and split the local data sets into a plurality of mutually disjoint data sets to obtain a plurality of training data sets corresponding to each participating device;

[0032] a local model obtaining module configured to train a model issued by the coordinating device by using the plurality of training data sets to obtain a local model corresponding to each training data set;

[0033] an evaluation parameter obtaining module configured to obtain first evaluation parameters and second evaluation parameters of each participating device by processing the test data sets and the overall test data set by using the local model;

[0034] a data quality detection module configured to detect the data quality corresponding to each participating device by using the first evaluation parameters and the second evaluation parameters.

[0035] In a third aspect, the embodiment of the present application provides a terminal device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the data quality detection method based on federated learning when executing the computer program.

[0036] In a fourth aspect, the embodiment of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executable on a processor to implement the steps of the data quality detection method based on federated learning.

[0037] The beneficial effects of the data quality detection method based on joint learning provided by this invention are at least as follows: First, this invention obtains the test datasets of each participating device and sends them to the coordinating device to obtain the total test dataset; second, it obtains the local datasets of each participating device and splits the local datasets into multiple non-overlapping datasets to obtain multiple training datasets corresponding to each participating device; third, it performs joint learning using the multiple training datasets and the model issued by the coordinating device to obtain the local model corresponding to each training dataset; then, it processes the test datasets and the total test dataset using the local model to obtain the first evaluation parameter and the second evaluation parameter of each participating device; finally, it uses the first evaluation parameter and the second evaluation parameter to detect the data quality corresponding to each participating device, thereby solving the problem in the prior art where poor data quality of one or more participating devices in horizontal joint learning leads to poor overall joint learning performance. In this invention, by splitting the local dataset into multiple disjoint datasets to obtain the training set, joint learning is performed, resulting in a more accurate model. Before obtaining the evaluation parameters, noise is processed on the data, introducing a certain degree of uncertainty and protecting the privacy of participating devices, thus solving the technical problem of poor data privacy protection in the Internet of Things. After obtaining the test datasets for each participating device by proportionally extracting local data from each participating device, a total test dataset is obtained based on the test datasets. The evaluation parameters are obtained by using both the test datasets and the total test dataset, taking into account the data of multiple participating devices, thus making the evaluation parameters more widely applicable. Finally, based on the evaluation parameters, local datasets of participating devices that meet preset conditions are selected to participate in joint learning, thereby improving the effect of joint learning. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of data quality detection based on joint learning provided in an embodiment of the present invention;

[0040] Figure 2 This is a flowchart illustrating the process of obtaining test datasets from each participating device and sending them to the coordinating device to obtain the total test dataset, as provided in this embodiment of the invention.

[0041] Figure 3 This is a flowchart illustrating the process of obtaining the first evaluation parameters of each participating device according to an embodiment of the present invention.

[0042] Figure 4 is a flow implementation diagram of a process for obtaining a second evaluation parameter of each participating device provided by an embodiment of the present application;

[0043] Figure 5 is a flow implementation diagram of a process for detecting data quality corresponding to each participating device using the first evaluation parameter and the second evaluation parameter provided by an embodiment of the present application;

[0044] Figure 6 is a flowchart of a data quality detection device based on federated learning provided by an embodiment of the present application;

[0045] Figure 7 is a schematic diagram of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0046] In the following description, specific details are set forth in order to provide a thorough understanding of embodiments of the application. However, persons of ordinary skill in the art will readily recognize that embodiments of the application can be practiced without these specific details. In other instances, well-known structures, devices, circuits, and processes have not been described in detail so as not to unnecessarily obscure aspects of the application. Based on the description of the described embodiments of the application, all other embodiments that person of ordinary skill in the art acquires are within the scope of protection of the present application. If not specifically indicated, the technical means used in the embodiments are conventional means known to those skilled in the art.

[0047] In order to illustrate the technical solutions of the present application, the following will be described through specific embodiments.

[0048] First embodiment

[0049] Figure 1 is a flowchart of a data quality detection method based on federated learning provided by an embodiment of the present application.

[0050] As shown in Figure 1 , the data quality detection method based on federated learning comprises steps S110-S150:

[0051] S110, obtaining a test data set of each participating device and sending it to a coordination device to obtain a total test data set;

[0052] S120, obtaining a local data set of each participating device and splitting the local data set into a plurality of mutually disjoint data sets to obtain a plurality of training data sets corresponding to each participating device;

[0053] S130, joint learning is performed by using the multiple training data sets and the model issued by the coordination device to obtain a local model corresponding to each training data set;

[0054] S140, after processing the test data sets and the total test data set by using the local models, a first evaluation parameter and a second evaluation parameter of each participating device are obtained;

[0055] S150, the data quality corresponding to each participating device is detected by using the first evaluation parameter and the second evaluation parameter.

[0056] The embodiment of the present application provides a data quality detection method based on joint learning. The method first obtains test data sets of participating devices and sends the test data sets to a coordination device to obtain a total test data set. Then, local data sets of the participating devices are obtained, and the local data sets are split into multiple mutually disjoint data sets to obtain multiple training data sets corresponding to each participating device. Joint learning is performed by using the multiple training data sets and the model issued by the coordination device to obtain a local model corresponding to each training data set. After processing the test data sets and the total test data set by using the local models, a first evaluation parameter and a second evaluation parameter of each participating device are obtained. Finally, the data quality corresponding to each participating device is detected by using the first evaluation parameter and the second evaluation parameter. Thus, the problem of poor overall joint learning effect caused by poor data quality of one or more participating devices in horizontal joint learning in the prior art is solved. The method splits the local data set into multiple mutually disjoint data sets to obtain training sets, and then performs joint learning to obtain a model with higher accuracy. Before obtaining the evaluation parameters, the data is processed for noise, a certain degree of uncertainty is introduced, and the privacy of the participating devices is protected, thereby solving the technical problem of poor data privacy protection in the Internet of Things. After obtaining the test data sets of the participating devices by proportionally extracting the local data of each participating device, the total test data set is obtained based on the test data sets. In the process of obtaining the evaluation parameters, the test data sets and the total test data set are used. The evaluation parameters obtained in this way take into account the data of multiple participating devices, thereby making the application scope of the evaluation parameters more extensive. Finally, the local data set of the participating device meeting the preset condition is selected based on the evaluation parameters to participate in joint learning, thereby improving the effect of joint learning.

[0057] Specifically, the data quality detection method based on federated learning is applied to participating devices, which are communicatively connected with a coordinating device or other participating devices. The local data set can be obtained according to the sensor corresponding to each participating device, or can be obtained according to the data generated by the coordinating device, such as browsing records, purchase records, consumption data, etc. The test data set can be obtained based on the local data set of each participating device, and the total test data set can be obtained based on the test data set of each participating device. For specific implementation methods of obtaining the test data set of each participating device and sending it to the coordinating device to obtain the total test data set, please refer to Figure 2 , Figure 2 is a flow implementation diagram provided by the present application in an embodiment for obtaining the test data set of each participating device and sending it to the coordinating device to obtain the total test data set.

[0058] As shown in Figure 2 , obtaining the test data set of each participating device and sending it to the coordinating device to obtain the total test data set can include the following steps S210-S230:

[0059] S210, proportionally extracting the local data of each participating device to obtain the test data set of each participating device;

[0060] S220, sending the test data set of each participating device to the coordinating device to obtain the total test data set.

[0061] Specifically, in step S210, the proportion can be 0.01-0.05. When there are many participating devices or the data amount in the local data set of each participating device is large, a smaller proportion can avoid the server from being unable to meet the computing capacity of the large data amount; when extracting data, a random data selection method can be used.

[0062] Specifically, in step S120, the multiple training data sets corresponding to each participating device can be obtained by splitting the local data set. The local data set of each participating device is split to obtain multiple training data sets that are mutually exclusive (no repeated data between each training data set). The number of training data sets obtained after splitting by different participating devices can be different, and in general, the number of splits is greater than or equal to 100; by splitting the local data set to obtain multiple mutually exclusive training data sets, the generalization ability of the model can be improved. After splitting the local data set into multiple mutually exclusive data sets to obtain the training set, federated learning is performed, and thus the model accuracy obtained after training the training data set is higher.

[0063] Specifically, in step S130, the following steps are included: (1) each participating device needs to accept the same model issued by the coordination device, which can be any general supervised learning model, such as linear regression, logistic regression, gradient boosting tree, neural network, etc.; (2) multiple training data sets are used to train the same model issued by the coordination device; (3) the trained model parameters are sent to the coordination device, the model parameters are aggregated and used to update the model, and then the updated model corresponding to the coordination device is obtained; (4) the updated model parameters are sent to each participating device, and each training data set is used to continue training the model; (5) steps (3) and (4) are repeated multiple times until the model trained by each training data set meets the preset condition, and the local model corresponding to each training set is obtained. The local model parameters corresponding to different training sets in different participating devices may be different, and the local model parameters corresponding to different training sets in the same participating device may also be different. The preset condition includes that the model accuracy meets the requirements and the number of training times is sufficient.

[0064] Specifically, after obtaining the local model, the first evaluation parameter and the second evaluation parameter of each participating device can be obtained according to the local data set; for the specific implementation method of obtaining the first evaluation parameter of each participating device, please refer to Figure 3 , Figure 3 is a flow implementation diagram provided by the present application in an embodiment for obtaining the first evaluation parameter of each participating device.

[0065] As shown in Figure 3 , obtaining the first evaluation parameter of each participating device can include the following steps S310-S330:

[0066] S310, the total test data set is removed from each test data set to obtain the detection data set corresponding to each participating device;

[0067] S320, inputting the detection data set into the local model corresponding to the detection data set to obtain the first predicted value;

[0068] S330, after noise processing of the first predicted value, the first evaluation parameter is obtained.

[0069] Specifically, the label corresponding to the participating device is k (k is a positive integer greater than or equal to 1), and the detection data set of the participating device k is a data set obtained by removing the test data set corresponding to the participating device k from the total test data set. The detection data set of the participating device k is input into the local model corresponding to each training data set of the participating device k, and n prediction values are obtained, n is the number of splits of the local data set of the participating device k into training data sets, so n is also equal to the number of training data sets and local models corresponding to the participating device k. In particular, the detection data set of the participating device k is input into a local model to obtain a prediction value, the prediction value is the loss value after the detection data set is input into the local model, which is directly obtained after the detection data set is input into the local model, and the sum of all prediction values corresponding to the participating device k is the first prediction value corresponding to the participating device k. The noise in the noise processing of the first prediction value can follow Gaussian, Laplace distribution, etc. For Gaussian distribution, μ ~ N(0, σ 2 ), μ and σ represent the mean and standard deviation, respectively. σ can be determined according to the privacy protection degree of the local data set corresponding to the participating device k, and then the first evaluation parameter corresponding to the participating device k is the sum of the first prediction value corresponding to the participating device k and μ. After the data is processed by noise, the first evaluation parameter is obtained, a certain degree of uncertainty is introduced, the privacy of the participating device is protected, and thus the technical problem of poor data privacy protection of the Internet of Things is solved.

[0070] Specifically, after obtaining the first evaluation parameter, the second evaluation parameter needs to be obtained; the specific implementation method of obtaining the second evaluation parameter of each participating device is described in Figure 4 , Figure 4 is a flow implementation diagram provided by the present application in an embodiment for obtaining the second evaluation parameter of each participating device.

[0071] As shown in Figure 4 , obtaining the second evaluation parameter of each participating device can include the following steps S410-S420:

[0072] S410, input each test data set into the local model corresponding to each test data set to obtain a second prediction value;

[0073] S420, after noise processing of the second prediction value, a second evaluation parameter is obtained.

[0074] Specifically, the test data set corresponding to the participating device k is input into a local model to obtain a prediction value, the prediction value is the loss value after the test data set is input into the local model, which is directly obtained after the test data set is input into the local model, and the sum of all prediction values corresponding to the test data set of the participating device k is the second prediction value corresponding to the participating device k. The noise in the noise processing of the second prediction value can follow Gaussian, Laplace distribution, etc. For Gaussian distribution, μ ~ N(0, σ2 ), and sigma can be determined according to the privacy protection degree of the local data set corresponding to the participating device k, and the second evaluation parameter corresponding to the participating device k is the sum of the second prediction value corresponding to the participating device k and mu. The second evaluation parameter is obtained by processing the data with noise, which introduces a certain degree of uncertainty and protects the privacy of the participating device, thereby solving the technical problem of poor data privacy protection in the Internet of Things.

[0075] Specifically, in step S150, the specific implementation method of detecting the data quality corresponding to each participating device by using the first evaluation parameter and the second evaluation parameter is as follows: Figure 5 , Figure 5 is a flowchart of detecting the data quality corresponding to each participating device by using the first evaluation parameter and the second evaluation parameter provided in an embodiment of the present application.

[0076] As shown in Figure 5 , obtaining the second evaluation parameter of each participating device can include the following steps S510-S520:

[0077] S510, weighted calculation of the first evaluation parameter and the second evaluation parameter corresponding to each participating device to obtain the total evaluation parameter corresponding to each participating device;

[0078] S520, detecting the data quality corresponding to each participating device by using the total evaluation parameter.

[0079] Specifically, the calculation formula of the total evaluation parameter corresponding to each participating device is:

[0080]

[0081] wherein k is the index of the participating device, η k is the total evaluation parameter of the participating device k, lambda is the weight coefficient, the value range of lambda is 0-1, is the first evaluation parameter of the participating device k, is the second evaluation parameter of the participating device k. Lambda can be determined according to actual experience. After proportionally extracting the local data of each participating device to obtain the test data set of each participating device, the total test data set is obtained based on the test data set; at the same time, by splitting the local data set of each participating device into multiple mutually disjoint data sets, multiple training data sets corresponding to each participating device are obtained; the test data set and the total test data set and the local model trained by the training data set are used in the process of obtaining the evaluation parameter, and the evaluation parameter obtained in this way considers the data of multiple participating devices, thereby making the application range of this evaluation parameter more extensive.

[0082] Specifically, after step S150, it further includes:

[0083] determining whether the total evaluation parameter of each participating device meets a preset condition;

[0084] If yes, the local data set of the participating device meeting the data quality requirement is screened out.

[0085] Specifically, the larger the total evaluation parameter is, the larger the loss value corresponding to the test data set or the detection data set input into the local model is; similarly, the smaller the total evaluation parameter is, the smaller the loss value is. The preset condition sets a total evaluation parameter threshold according to actual experience. If the total evaluation parameter corresponding to the participating device is greater than the total evaluation parameter threshold, it indicates that the data quality of the local data set corresponding to the participating device does not meet the requirement, and the local data set of the participating device should not be used in joint learning. If the total evaluation parameter corresponding to the participating device is less than the total evaluation parameter threshold, it indicates that the data quality of the local data set corresponding to the participating device is high, and the local data set of the participating device can be used in joint learning. The local data set with high data quality is screened out through the total evaluation parameter of the participating device, and the local data set is used to participate in joint learning, thereby improving the effect of joint learning.

[0086] Second embodiment

[0087] Based on the same inventive concept as the method in the first embodiment, the present embodiment also provides a data quality detection device based on joint learning.

[0088] Figure 6 A flowchart of the joint learning device based on the risk state of the energy equipment provided by the present application.

[0089] As shown in Figure 6 The device 6 shown includes a total test data set acquisition module 61, a training data set acquisition module 62, a local model acquisition module 63, an evaluation parameter acquisition module 64, and a data quality detection module 65.

[0090] The total test data set acquisition module is configured to acquire the test data set of each participating device and send it to the coordinating device to obtain a total test data set.

[0091] The training data set acquisition module is configured to acquire the local data set of each participating device and split the local data set into a plurality of mutually disjoint data sets to obtain a plurality of training data sets corresponding to each participating device.

[0092] The local model acquisition module is configured to train the model issued by the coordinating device using the plurality of training data sets to obtain a local model corresponding to each training data set.

[0093] The evaluation parameter acquisition module is configured to process the test datasets and the total test dataset using the local model to obtain the first evaluation parameter and the second evaluation parameter for each participating device.

[0094] The data quality detection module is configured to detect the data quality of each participating device using the first evaluation parameter and the second evaluation parameter.

[0095] In some exemplary embodiments, the total test dataset acquisition module specifically includes:

[0096] The test dataset acquisition unit is configured to extract local data from each participating device proportionally to obtain the test dataset for each participating device.

[0097] The total test dataset acquisition unit is configured to send the test datasets of each participating device to the coordinating device to obtain the total test dataset.

[0098] In some exemplary embodiments, the evaluation parameter acquisition module specifically includes:

[0099] The detection dataset acquisition unit is configured to remove each test dataset from the total test dataset to obtain the detection dataset corresponding to each participating device.

[0100] The first prediction value acquisition unit is configured to input the detection dataset into the local model corresponding to the detection dataset to obtain the first prediction value;

[0101] The first evaluation parameter acquisition unit is configured to process the first predicted value for noise to obtain the first evaluation parameter.

[0102] The second prediction value acquisition unit is configured to input each test dataset into the local model corresponding to each test dataset to obtain the second prediction value;

[0103] The second evaluation parameter acquisition unit is configured to process the second predicted value for noise to obtain the second evaluation parameter.

[0104] In some exemplary embodiments, the data quality detection module specifically includes:

[0105] The overall evaluation parameter acquisition unit is configured to calculate the first evaluation parameter and the second evaluation parameter corresponding to each participating device in a weighted manner, so as to obtain the overall evaluation parameter corresponding to each participating device.

[0106] The data quality detection unit is configured to detect the data quality of each participating device using the overall evaluation parameters;

[0107] The formula for calculating the total evaluation parameters for each participating device is as follows:

[0108]

[0109] wherein k is the index of the participating device, η k is the total evaluation parameter of the participating device k, λ is a weight coefficient, and the value range of λ is 0-1, is the first evaluation parameter of the participating device k, is the second evaluation parameter of the participating device k.

[0110] In some example embodiments, the apparatus further comprises:

[0111] a judging module configured to judge whether the total evaluation parameter of each participating device meets a preset condition;

[0112] an executing module configured to, if yes, screen out the local data set of the participating device meeting the data quality requirement.

[0113] Third embodiment

[0114] The above method and apparatus can be applied to terminal devices such as desktop computers, notebooks, palm computers and cloud servers.

[0115] Figure 7 A schematic diagram of a terminal device to which the method and apparatus provided in an embodiment of the present application can be applied is shown in the figure. The device 7 comprises a memory 71, a processor 70, and a computer program 72 stored in the memory 71 and executable on the processor 70, and the processor 70 implements the steps of the method for detecting data quality based on joint learning when executing the computer program 72. For example Figure 6 the functions of the modules 61-65 shown.

[0116] The device 7 can be a computing device such as a cloud server. The terminal device can include, but is not limited to, the processor 70 and the memory 71. Those skilled in the art can understand that Figure 7 The device 7 is only an example and does not constitute a limitation on the terminal device 7, and can include more or fewer components than shown, or combine certain components, or different components, for example, the terminal device can also include input / output devices, network access devices, buses, etc.

[0117] The processor 70 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0118] The memory 71 can be an internal storage unit of the device 7, such as a hard disk or a memory of the device 7. The memory 71 can also be an external storage device of the device 7, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 71 can include both an internal storage unit and an external storage device of the device 7. The memory 71 is used to store the computer program and other programs and data required by the terminal device. The memory 71 can also be used to temporarily store data that has been output or is to be output.

[0119] The integrated module / unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be implemented by a computer program instructing related hardware to complete, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0120] Specifically, the embodiments of the present application further provide a computer readable storage medium, which can be a computer readable storage medium included in the memory in the above-mentioned embodiments, or a computer readable storage medium that exists separately and is not assembled into a terminal device. The computer readable storage medium stores one or more computer programs:

[0121] The computer readable storage medium includes the computer readable storage medium storing the computer program, and the computer program is executed by the processor to implement the steps of the data quality detection method based on joint learning.

[0122] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be realized in the form of hardware or software. In addition, the specific name of each functional unit and module is only for the convenience of mutual distinction, and does not limit the protection scope of the present application. The specific working process of the unit and module in the above system can be referred to the corresponding process in the foregoing method embodiment, which will not be described here.

[0123] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments.

[0124] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0125] In the embodiments provided by the present application, it should be understood that the disclosed apparatus / terminal device and method can be implemented by other ways. For example, the above-mentioned apparatus / terminal device embodiments are only schematic, and the division of the modules or units is only a logical function division, and there can be another division way in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0126] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0127] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0128] The integrated module / unit, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer-readable storage medium. When the processor executes the computer program, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0129] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A data quality detection method based on joint learning, characterized in that, The data quality detection method is applied to participating devices, which are communicatively connected to coordinating devices or other participating devices, including: Obtain the test datasets from each participating device and send them to the coordinating device to obtain the total test dataset; Obtain the local dataset of each participating device, and split the local dataset into multiple non-overlapping datasets to obtain multiple training datasets corresponding to each participating device; Joint learning is performed using the multiple training datasets and the model distributed by the coordination device to obtain the local model corresponding to each training dataset; After processing the test datasets and the total test dataset using the local model, the first evaluation parameters and the second evaluation parameters of each participating device are obtained. The first and second evaluation parameters are used to detect the data quality of each participating device, and it is determined whether the total evaluation parameters of each participating device meet the preset conditions; if so, the local datasets of the participating devices that meet the data quality requirements are selected for joint learning. The data quality of each participating device is detected using the first evaluation parameter and the second evaluation parameter, specifically including: weighted calculation of the first evaluation parameter and the second evaluation parameter corresponding to each participating device to obtain the total evaluation parameter corresponding to each participating device; and the data quality of each participating device is detected using the total evaluation parameter. After processing the test datasets and the total test dataset using the local model, the first evaluation parameter and the second evaluation parameter of each participating device are obtained. Specifically, this includes: removing each test dataset from the total test dataset to obtain the detection dataset corresponding to each participating device; inputting the detection dataset into the local model corresponding to the detection dataset to obtain the first predicted value; and processing the first predicted value for noise to obtain the first evaluation parameter. After processing the noise in the predicted values ​​to obtain the first evaluation parameter, the method further includes: inputting each test dataset into the local model corresponding to each test dataset to obtain the second predicted value; and processing the noise in the second predicted value to obtain the second evaluation parameter.

2. The method according to claim 1, characterized in that, Obtain the test datasets from each participating device and send them to the coordinating device to obtain the total test dataset, which specifically includes: Local data from each participating device is extracted proportionally to obtain the test dataset for each participating device. The test datasets of each participating device are sent to the coordinating device to obtain the total test dataset.

3. The method according to claim 1, characterized in that, The formula for calculating the total evaluation parameters for each participating device is as follows: ； Where k is the label of the participating equipment. Let λ be the overall evaluation parameter for equipment k, and let λ be the weighting coefficient, with a value ranging from 0 to 1. The first evaluation parameter for participating equipment k. This is the second evaluation parameter for participating device k.

4. A data quality detection device based on joint learning, characterized in that, The device includes: The total test dataset acquisition module is configured to acquire the test datasets of each participating device and send them to the coordinating device to obtain the total test dataset; The training dataset acquisition module is configured to acquire the local dataset of each participating device and split the local dataset into multiple non-overlapping datasets to obtain multiple training datasets corresponding to each participating device. The local model acquisition module is configured to train the model distributed by the coordination device using the multiple training datasets to obtain the local model corresponding to each training dataset. The evaluation parameter acquisition module is configured to process the test datasets and the total test dataset using the local model to obtain the first evaluation parameter and the second evaluation parameter for each participating device. The data quality detection module is configured to detect the data quality corresponding to each participating device using the first evaluation parameter and the second evaluation parameter. Specifically, the detection of the data quality corresponding to each participating device using the first evaluation parameter and the second evaluation parameter includes: weighted calculation of the first evaluation parameter and the second evaluation parameter corresponding to each participating device to obtain the total evaluation parameter corresponding to each participating device; and detection of the data quality corresponding to each participating device using the total evaluation parameter. The judgment module is configured to determine whether the total evaluation parameters of each participating device meet preset conditions. If the execution module is configured to do so, it will filter out the local datasets of participating devices that meet the data quality requirements. After processing the test datasets and the total test dataset using the local model, the first evaluation parameter and the second evaluation parameter of each participating device are obtained. Specifically, this includes: removing each test dataset from the total test dataset to obtain the detection dataset corresponding to each participating device; inputting the detection dataset into the local model corresponding to the detection dataset to obtain the first predicted value; and processing the first predicted value for noise to obtain the first evaluation parameter. After processing the noise in the predicted values ​​to obtain the first evaluation parameter, the method further includes: inputting each test dataset into the local model corresponding to each test dataset to obtain the second predicted value; and processing the noise in the second predicted value to obtain the second evaluation parameter.

5. A terminal 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 steps of the method as described in any one of claims 1 to 3.

6. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 3.

Images