Data value determination method and device in federated learning

Abstract

The invention discloses a data value determination method and device in federated learning. The method comprises the following steps: acquiring data sets of a plurality of target objects participating in federated learning, and determining a first feature set of each data set; performing random replacement extraction on all the features, generating a plurality of second feature sets as training samples, and determining sample tags according to federal learning results; training a data value evaluation model based on a factorization machine by using the training sample and the sample label, determining model parameters including feature weights of different input features and an implicit vector matrix in a multi-dimensional space, the model being used for predicting a federated learning result; and calculating and normalizing the value score of each feature according to the first feature set, the feature weight and the implicit vector matrix of each target object, and summing to obtain the contribution value of the target object in federated learning. The technical problem that the value of data provided for each participant in a federated learning scene is difficult to accurately evaluate is solved.

Description

Technical Field

[0001] The present application relates to the technical field of federated learning, and more specifically, to a method and device for determining data value in federated learning. Background Art

[0002] As a new production factor, data plays a key role in the digital economy era. Federated Learning (FL), as a machine learning framework, aims to ensure data privacy and security, allowing multiple participants to jointly train models without sharing original data to improve model performance. Its design goal is not only to improve model performance, but also to measure the contribution value of each participant's data to the model to support reasonable income distribution. For this reason, data transaction pricing based on federated learning becomes particularly important.

[0003] Existing technologies have obvious limitations when evaluating the contribution value of data from each participant in federated learning. For example, the leave-one-out (LOO) method assumes that each data point has duplicate copies. In actual operation, removing unique instances will change the composition of the data set, resulting in inaccurate evaluation and ignoring the interaction between data. Other technologies, such as using multiple methods to train pricing models in stages, fix the value of data products, fail to consider the usage differences of different participants, and are not suitable for federated learning scenarios. In addition, data pricing methods based on reinforcement learning have deficiencies in dealing with feature superposition effects and non-empty data information, affecting the fairness of the evaluation model. These limitations limit the application effect of existing technologies in federated learning.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a method and device for determining data value in federated learning, so as to at least solve the technical problem that it is difficult to accurately evaluate the value of data provided by each participant in a federated learning scenario.

[0006] According to one aspect of an embodiment of the present application, a method for determining data value in federated learning is provided, including: obtaining multiple data sets corresponding to multiple target objects participating in federated learning, and determining a first feature set corresponding to each data set; performing multiple random feature extractions with replacement on all features in the multiple feature sets to obtain multiple second feature sets, taking each second feature set as a training sample, and obtaining federated learning results for the multiple first feature sets, and determining a sample label for each training sample based on the federated learning results; using multiple training samples and sample labels to train a data value assessment model based on a factorization machine to determine model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in a multidimensional space corresponding to different input features, and the data value assessment model is used to predict the federated learning results based on each input feature and the corresponding feature weights and latent vector matrix; determining the contribution value of the data set corresponding to the target object in federated learning based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix.

[0007] Optionally, obtaining multiple data sets corresponding to multiple target objects participating in federated learning includes: obtaining multiple data sets corresponding to multiple target objects participating in federated learning from a privacy computing platform, wherein the privacy computing platform is used to receive data sets uploaded by each target object, and perform privacy encryption processing on the data sets, wherein the privacy encryption processing method includes at least one of the following: symmetric encryption, asymmetric encryption, homomorphic encryption, and differential privacy processing.

[0008] Optionally, determining the first feature set corresponding to each data set includes: for each data set, preprocessing the data in the data set, wherein the preprocessing method includes: filling missing values ​​in the categorical variable data with unknown, and filling missing values ​​in the numerical variable data with zero values ​​or statistical values, and the statistical values ​​include at least one of the following: mean, median, mode; encoding each preprocessed categorical variable data with a label encoder to obtain a first feature corresponding to each categorical variable data; and / or encoding each preprocessed numerical variable data with a minimum-maximum normalization encoder to obtain a second feature corresponding to each numerical variable data; and combining each first feature and / or each second feature into a first feature set corresponding to the data set.

[0009] Optionally, determining the sample label of each training sample according to the federated learning result includes: using the federated learning result as the sample label of each training sample, and adding noise to the sample labels of a preset proportion of the training samples.

[0010] Optionally, the model formula of the data value assessment model is expressed as follows: In the formula, x={x i}, i = 1, 2, ..., n, represents the second feature set of the input, n represents the number of features in the second feature set, represents the federated learning result based on the prediction of the second feature set, is the penalty factor, y represents the corresponding sample label, w i Represents feature x i The corresponding feature weight, v i ,v j Represents the feature x i and feature x j The corresponding latent vector matrix in multidimensional space is, <v i ,v j > represents the latent vector matrix v i and v j The dot product of .

[0011] Optionally, a data value assessment model based on a factor decomposition machine is trained using multiple training samples and sample labels to determine model parameters, including: inputting each second feature set into the data value assessment model in turn for iterative training until a preset iteration termination condition is met, wherein, during the training process, a cross-entropy loss function is constructed based on the federated learning prediction results and sample labels corresponding to each second feature set, and the model parameters are adjusted based on the cross-entropy loss function combined with the gradient descent method, and the iteration termination condition includes at least one of the following: the number of iterations is greater than a first preset threshold, the change in the model parameters is less than a second preset threshold, and the decrease in the loss function is less than a third preset threshold.

[0012] Optionally, determining the contribution value of the data set corresponding to the target object in federated learning according to the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix includes: for each feature in each first feature set, calculating the value score corresponding to the feature according to the following formula: value(x i )=w i x i +(x i ·v i ) 2 -x i 2 ·v i ·v i , where value(x i ) represents the feature x i Corresponding value scores; normalize the value scores of all features; for each target object, sum the normalized value scores of each feature in the first feature set corresponding to the target object to obtain the contribution value of the data set corresponding to the target object in federated learning.

[0013] According to another aspect of an embodiment of the present application, a device for determining data value in federated learning is also provided, including: an acquisition module, used to acquire multiple data sets corresponding to multiple target objects participating in federated learning, and determine a first feature set corresponding to each data set; a sample processing module, used to perform multiple random feature extractions with replacement on all features in the multiple feature sets to obtain multiple second feature sets, use each second feature set as a training sample, and obtain federated learning results for the multiple first feature sets, and determine the sample label of each training sample based on the federated learning results; a training module, used to train a data value assessment model based on a factor decomposition machine using multiple training samples and sample labels to determine model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in multidimensional space corresponding to different input features, and the data value assessment model is used to predict the federated learning results based on each input feature and the corresponding feature weights and latent vector matrix; an assessment module, used to determine the contribution value of the data set corresponding to the target object in federated learning based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix.

[0014] According to another aspect of an embodiment of the present application, a computer program product is also provided, the computer program product comprising: a computer program, wherein when the computer program is executed by a processor, the above-mentioned method for determining data value in federated learning is implemented.

[0015] According to another aspect of an embodiment of the present application, an electronic device is also provided, which includes: a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above-mentioned method for determining data value in federated learning through the computer program.

[0016] In the embodiment of the present application, by obtaining data sets corresponding to multiple target objects, but not directly sharing the original data, a second feature set is extracted as a training sample, which ensures the security of the data in federated learning. The sample label is determined by the federated learning result, that is, the improvement of the model performance by the feature combination is quantified, which makes the data value assessment more accurate. By comprehensively considering the feature weights and the latent vector matrix through the data value assessment model based on the factor decomposition machine, the contribution of the data set to federated learning can be more comprehensively evaluated, thereby providing a fair basis for data value distribution for the data element trading platform and encouraging the participation of data providers. The introduction of the latent vector matrix allows the model to capture the potential interactions between features, which is particularly important for evaluating the value of data in the federated transfer learning scenario. Traditional methods may ignore the mutual influence between features, and this solution can more accurately reflect the comprehensive value of features by calculating the contribution of feature combinations. Considering the high diversity of data sets of different participants, this technical solution can be applied to a wide range of federated learning scenarios, which makes the data value assessment model more universal and applicable, and can adapt to different industries and data sets of different sizes. And then solve the technical problem that the value of the data provided by each participant in the federated learning scenario is difficult to accurately evaluate. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 It is a flowchart of an optional method for determining data value in federated learning according to an embodiment of the present application;

[0019] Figure 2 is a schematic structural diagram of an optional device for determining data value in federated learning according to an embodiment of the present application;

[0020] Figure 3 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification, claims and drawings of the present application 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 interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] In order to better understand the embodiments of the present application, some nouns or terms that appear in the description of the embodiments of the present application are first translated and explained as follows:

[0024] Federated learning: is a distributed machine learning method whose core concept is to allow multiple devices or organizations to collaboratively train models without sharing original data. Federated learning combines machine learning technology and privacy protection mechanisms to solve the problem of how to effectively use data scattered across different devices or organizations for model training under the constraints of data privacy and security laws. In the federated learning framework, data is retained on the local devices or servers of each participant and is not concentrated in a central location. Each participant trains the model using local data, and then sends updates to the model (such as gradients or model parameters) to the central server, which aggregates these updates, optimizes the global model, and then distributes the optimized model parameters to the participants, who further train the model with local data. This process is repeated until the preset training goal is reached or the model converges.

[0025] The information collected in the embodiments of the present application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0026] Example 1

[0027] According to an embodiment of the present application, a method for determining data value in federated learning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] Figure 1 is a flow chart of a method for determining data value in federated learning provided in an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0029] Step S102, obtaining multiple data sets corresponding to multiple target objects participating in federated learning, and determining a first feature set corresponding to each data set;

[0030] Step S104, performing multiple random feature extractions with replacement on all features in the multiple feature sets to obtain multiple second feature sets, taking each second feature set as a training sample, and obtaining federated learning results for the multiple first feature sets, and determining a sample label for each training sample based on the federated learning results;

[0031] Step S106, using multiple training samples and sample labels to train a data value assessment model based on a factorization machine, and determining model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in a multidimensional space corresponding to different input features, and the data value assessment model is used to predict federated learning results based on each input feature and the corresponding feature weights and latent vector matrix;

[0032] Step S108, determining the contribution value of the data set corresponding to the target object in federated learning based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix.

[0033] The following describes the various steps of the method for determining data value in federated learning in combination with the specific implementation process.

[0034] Acquire multiple data sets corresponding to multiple target objects participating in federated learning, and determine a first feature set corresponding to each data set.

[0035] As an optional implementation, the process of obtaining multiple data sets corresponding to multiple target objects participating in federated learning can be carried out by taking the following steps: obtaining multiple data sets corresponding to multiple target objects participating in federated learning from a privacy computing platform, wherein the privacy computing platform is used to receive the data sets uploaded by each target object, and perform privacy encryption processing on the data sets, wherein the privacy encryption processing method includes at least one of the following: symmetric encryption, asymmetric encryption, homomorphic encryption, and differential privacy processing.

[0036] As an optional implementation, the process of determining the first feature set corresponding to each data set can refer to the following steps: for each data set, preprocess the data in the data set, wherein the preprocessing method includes: for missing values ​​in categorical variable data, fill them with unknown, and for missing values ​​in numerical variable data, fill them with zero values ​​or statistical values, and the statistical values ​​include at least one of the following: mean, median, mode; use a label encoder to encode each preprocessed categorical variable data to obtain the first feature corresponding to each categorical variable data; and / or use a minimum-maximum normalization encoder to encode each preprocessed numerical variable data to obtain the second feature corresponding to each numerical variable data; each first feature and / or each second feature is combined into the first feature set corresponding to the data set.

[0037] After obtaining the above-mentioned first feature set, multiple random feature extractions with replacement are performed on all features in multiple feature sets to obtain multiple second feature sets. Each second feature set is used as a training sample, and federated learning results for multiple first feature sets are obtained. The sample label of each training sample is determined based on the federated learning results.

[0038] As an optional implementation, the process of determining the sample label of each training sample based on the federated learning result may be as follows: using the federated learning result as the sample label of each training sample, and adding noise to the sample labels of a preset proportion of the training samples.

[0039] For example, multiple hospitals cooperate to use federated learning to predict the probability of a certain disease (such as cancer) while protecting patient privacy. The data set of each hospital, i.e., the first feature set, contains the patient's basic information (age, gender), health status (weight, blood pressure), genetic genes, disease history and other features. Randomly select features from the features of all hospitals. For example, the first extraction may include age, genetic genes and blood pressure to form a second feature set as part of the training sample. Use this extracted second feature set for federated learning to train the disease prediction model. The results may be expressed as the accuracy of the model, AUC value, etc., such as the model accuracy is increased from 80% to 85%. The increase in model accuracy (5%) is used as the sample label of the training sample to reflect the contribution of this feature combination to the model. According to the preset ratio (such as 10%), some training samples are randomly selected and noise is added to their sample labels. For example, the label of the first group of training samples (age and genetic genes) was originally 5%, but adding noise to the labels of 10% of the samples may change the label to 4% or 6%. The purpose of adding noise is to increase the uncertainty of the labels, protect the privacy of the participating hospital data, and prevent the specific data of each hospital from being reversely inferred through the federated learning results.

[0040] Traffic management departments in multiple cities jointly use federated learning to predict the probability of traffic accidents, improve road safety, and protect the privacy of traffic data. The data set of each city, namely the first feature set, contains features such as vehicle type, time, weather conditions, road conditions, and vehicle speed. Features are randomly extracted from the traffic data of all cities. For example, the first extraction may include vehicle type and time to form a second feature set. Federated learning is used to train the traffic accident prediction model using the extracted second feature set. The result may be an improvement in the performance of the model in predicting the accident rate, such as the recall rate of the model from 60% to 65%. The increase in recall rate (5%) is used as a sample label for subsequent data value assessment model training to quantify the contribution of feature combinations to model improvement. According to a preset ratio (such as 20%), some training samples are randomly selected and noise is added to their sample labels. For example, the recall rate improvement label of the first extracted training sample (vehicle type and weather conditions) is changed from 5% to 4.8% or 5.2%, ensuring that the privacy of the original data is still protected even if the model parameters are shared in federated learning.

[0041] In each of the above scenarios, the result of federated learning is an improvement in model performance, which can be an increase in accuracy, recall or AUC (Area Under the Curve) value. Through multiple random extractions with replacement, multiple second feature sets are obtained, each of which is a training sample. The model is trained using the federated learning framework, and the changes in model performance are observed. These changes are quantified as sample labels, reflecting the contribution of specific feature combinations to the model. Sample labels not only reflect the direct value of the features, but also imply the interaction value between features, which is crucial for fairly measuring the value of the data of each participant in federated learning. In order to further protect data privacy, noise is added to a preset proportion of sample labels to ensure that the privacy and sensitive information of the original data will not be leaked even in the federated learning environment. In this way, the data value of each participant can be fairly evaluated, while promoting data collaboration and privacy protection under the federated learning framework.

[0042] After obtaining the above-mentioned training samples and sample labels, multiple training samples and sample labels are used to train a data value assessment model based on a factor decomposition machine to determine model parameters, where the model parameters include: feature weights corresponding to different input features and latent vector matrices in multidimensional space corresponding to different input features. The data value assessment model is used to predict the federated learning results based on each input feature and the corresponding feature weights and latent vector matrix.

[0043] Among them, the model formula of the data value assessment model can be expressed as follows:

[0044]

[0045] In the formula, x={x i}, i = 1, 2, ..., n, represents the second feature set of the input, n represents the number of features in the second feature set, represents the federated learning result based on the prediction of the second feature set, is the penalty factor, y represents the corresponding sample label, w i Represents feature x i The corresponding feature weight, v i ,v j Represents the feature x i and feature x j The corresponding latent vector matrix in multidimensional space is, <v i ,v j > represents the latent vector matrix v i and v j The dot product of .

[0046] As an optional implementation, using multiple training samples and sample labels to train a data value assessment model based on a factor decomposition machine to determine model parameters can include the following process: inputting each second feature set into the data value assessment model in turn for iterative training until a preset iteration termination condition is met, wherein, during the training process, a cross-entropy loss function is constructed based on the federated learning prediction results and sample labels corresponding to each second feature set, and the model parameters are adjusted based on the cross-entropy loss function combined with the gradient descent method, and the iteration termination condition includes at least one of the following: the number of iterations is greater than a first preset threshold, the change in the model parameters is less than a second preset threshold, and the decrease in the loss function is less than a third preset threshold.

[0047] After multiple iterations of training, the model determines that for each input feature x i The corresponding feature weight w i The model also determines for each input feature x i and x j The latent vector matrix v in multidimensional space i and v j After training, the model parameters (including feature weights and latent vector matrices) will be fixed. The feature weights reflect the importance of the features in prediction, while the latent vector matrix captures the potential interactions between features. These parameters will be used to predict the potential value of any given feature set in federated learning.

[0048] After obtaining the above-mentioned model parameters such as latent vector matrix and feature weights, the contribution value of the data set corresponding to the target object in federated learning is determined according to the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix. This process can be performed as follows:

[0049] For each feature in each first feature set, the value score corresponding to the feature is calculated according to the following formula:

[0050] value(x i )=w i x i +(x i ·v i ) 2 -x i 2 ·v i ·v i

[0051] In the formula, value(x i ) represents the feature x i The corresponding value score;

[0052] The value scores of all features are normalized to ensure that the value scores of all features are on the same scale and to avoid feature scores with large values ​​from having too great an impact on the overall value assessment. Commonly used normalization methods include maximum and minimum value normalization, etc. The specific selection should be based on the characteristics and needs of the data.

[0053] For each target object, the normalized value scores of each feature in the first feature set corresponding to the target object are summed to obtain the contribution value of the data set corresponding to the target object in federated learning. This summation process is actually a quantification of the overall data value of the target object. It combines the direct value of the features provided by the target object and the interactive value between the features, thereby more comprehensively evaluating the contribution of the data to the federated learning model.

[0054] For example, for disease prediction in the medical and health field, multiple hospitals jointly predict the development trend of diabetes through federated learning. The data set (first feature set) of each hospital contains information such as the patient's age, gender, BMI, blood pressure, and blood sugar level. Create a second feature set. For example, the first random extraction obtains blood sugar level and BMI as the second feature set; the second extraction obtains age and blood pressure. Using federated learning to train the disease prediction model, the first extracted feature set increases the model accuracy from 85% to 92% (7% improvement), and the second extracted feature increases the accuracy to 88% (3% improvement). These improvement values ​​are used as sample labels for training samples. Based on the above sample labels, the value evaluation model is trained to obtain model parameters such as latent vector matrix and feature weights, thereby determining the overall data value. For example, if the blood sugar level and BMI of hospital A contribute particularly well to the model, then the data value of hospital A will be higher.

[0055] For user behavior prediction in the field of smart advertising, multiple e-commerce platforms predict user purchase behavior through federated learning. The data set (first feature set) of each platform contains information such as user browsing history, search keywords, purchase records, user preferences, etc. Create a second feature set: For example, randomly extract browsing history and search keywords from all platforms, as well as purchase records and user preferences to form different second feature sets. After federated learning, the accuracy changes of models trained with different feature sets in predicting purchase behavior are used as sample labels. For example, the former combination increases the accuracy from 70% to 80% (10% increase), while the latter combination increases it to 75% (5% increase). The training samples with sample labels are used to train the data value evaluation model to quantify the value of different feature sets, obtain model parameters such as latent vector matrix and feature weights, and thus determine the overall data value. For example, if the purchase records and user preferences of platform B are very critical to prediction, then the data value of platform B will be significantly improved.

[0056] For device failure prediction of IoT technology, multiple IoT device manufacturers predict device failures through federated learning. Each manufacturer's data set (first feature set) includes device type, operating time, temperature, humidity, vibration data, etc. The device type and operating time, as well as temperature and humidity features are randomly extracted to form the second feature set. The device failure prediction model is trained using the federated learning framework, and the improvement of each feature set on the model performance (such as reducing the false alarm rate) is recorded. These improvements are used as sample labels. The sample labels and the generated second feature set are used to train the value evaluation model to obtain model parameters such as latent vector matrix and feature weights, thereby determining the overall data value. For example, if the device type and operating time data of manufacturer C are particularly helpful in improving model performance, the data value of manufacturer C will be higher.

[0057] In all these scenarios, each participant (such as hospitals, e-commerce platforms, and manufacturers) provides a unique data set, and a model (such as disease prediction, user behavior prediction, and equipment failure prediction) is jointly optimized through federated learning. And through training, model parameters such as latent vector matrices and feature weights are obtained to determine the value of each data feature. This method ensures a fair assessment of the data value, while protecting the data privacy of the participants and promoting data cooperation across organizations.

[0058] In the embodiment of the present application, by obtaining data sets corresponding to multiple target objects, but not directly sharing the original data, a second feature set is extracted as a training sample, which ensures the security of the data in federated learning. The sample label is determined by the federated learning result, that is, the improvement of the model performance by the feature combination is quantified, which makes the data value assessment more accurate. By comprehensively considering the feature weights and the latent vector matrix through the data value assessment model based on the factor decomposition machine, the contribution of the data set to federated learning can be more comprehensively evaluated, thereby providing a fair basis for data value distribution for the data element trading platform and encouraging the participation of data providers. The introduction of the latent vector matrix allows the model to capture the potential interactions between features, which is particularly important for evaluating the value of data in the federated transfer learning scenario. Traditional methods may ignore the mutual influence between features, and this solution can more accurately reflect the comprehensive value of features by calculating the contribution of feature combinations. Considering the high diversity of data sets of different participants, this technical solution can be applied to a wide range of federated learning scenarios, which makes the data value assessment model more universal and applicable, and can adapt to different industries and data sets of different sizes. And then solve the technical problem that the value of the data provided by each participant in the federated learning scenario is difficult to accurately evaluate.

[0059] Example 2

[0060] According to an embodiment of the present application, a device for determining data value in federated learning for implementing the method for determining data value in federated learning in embodiment 1 is also provided, such as Figure 2 As shown, the data value determination device in the federated learning at least includes: an acquisition module 21, a sample processing module 22, a training module 23 and an evaluation module 24, wherein:

[0061] An acquisition module 21 is used to acquire multiple data sets corresponding to multiple target objects participating in federated learning, and determine a first feature set corresponding to each data set;

[0062] A sample processing module 22 is used to perform multiple random feature extractions with replacement on all features in the multiple feature sets to obtain multiple second feature sets, use each second feature set as a training sample, obtain federated learning results for the multiple first feature sets, and determine a sample label for each training sample based on the federated learning results;

[0063] A training module 23 is used to train a data value assessment model based on a factorization machine using multiple training samples and sample labels to determine model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in a multidimensional space corresponding to different input features, and the data value assessment model is used to predict federated learning results based on each input feature and the corresponding feature weights and latent vector matrix;

[0064] The evaluation module 24 is used to determine the contribution value of the data set corresponding to the target object in the federated learning based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix.

[0065] The functions of each module of the data value determination device in federated learning are explained below in conjunction with the specific implementation process.

[0066] The acquisition module acquires multiple data sets corresponding to multiple target objects participating in federated learning, and determines a first feature set corresponding to each data set.

[0067] As an optional implementation, the process of obtaining multiple data sets corresponding to multiple target objects participating in federated learning can be carried out in the following steps: obtaining multiple data sets corresponding to multiple target objects participating in federated learning from a privacy computing platform, wherein the privacy computing platform is used to receive the data sets uploaded by each target object, and perform privacy encryption processing on the data sets, wherein the privacy encryption processing method includes at least one of the following: symmetric encryption, asymmetric encryption, homomorphic encryption, and differential privacy processing.

[0068] As an optional implementation, the process of determining the first feature set corresponding to each data set can refer to the following steps: for each data set, preprocess the data in the data set, wherein the preprocessing method includes: for missing values ​​in categorical variable data, fill them with unknown, and for missing values ​​in numerical variable data, fill them with zero values ​​or statistical values, and the statistical values ​​include at least one of the following: mean, median, mode; use a label encoder to encode each preprocessed categorical variable data to obtain the first feature corresponding to each categorical variable data; and / or use a minimum-maximum normalization encoder to encode each preprocessed numerical variable data to obtain the second feature corresponding to each numerical variable data; each first feature and / or each second feature is combined into the first feature set corresponding to the data set.

[0069] After obtaining the above-mentioned first feature set, the sample processing module performs multiple random feature extractions with replacement on all features in multiple feature sets to obtain multiple second feature sets, takes each second feature set as a training sample, and obtains federated learning results for multiple first feature sets, and determines the sample label of each training sample based on the federated learning results.

[0070] As an optional implementation, the process of determining the sample label of each training sample based on the federated learning result may be as follows: using the federated learning result as the sample label of each training sample, and adding noise to the sample labels of a preset proportion of the training samples.

[0071] After obtaining the above-mentioned training samples and sample labels, the training module uses multiple training samples and sample labels to train a data value assessment model based on a factor decomposition machine to determine the model parameters, where the model parameters include: feature weights corresponding to different input features and latent vector matrices in multidimensional space corresponding to different input features. The data value assessment model is used to predict the federated learning results based on each input feature and the corresponding feature weights and latent vector matrix.

[0072] Among them, the model formula of the data value assessment model can be expressed as follows:

[0073]

[0074] In the formula, x={x i}, i = 1, 2, ..., n, represents the second feature set of the input, n represents the number of features in the second feature set, represents the federated learning result based on the prediction of the second feature set, is the penalty factor, y represents the corresponding sample label, w i Represents feature x iThe corresponding feature weight, v i ,v j Represents the feature x i and feature x j The corresponding latent vector matrix in multidimensional space is, <v i ,v j > represents the latent vector matrix v i and v j The dot product of .

[0075] As an optional implementation, using multiple training samples and sample labels to train a data value assessment model based on a factor decomposition machine to determine model parameters can include the following process: inputting each second feature set into the data value assessment model in turn for iterative training until a preset iteration termination condition is met, wherein, during the training process, a cross-entropy loss function is constructed based on the federated learning prediction results and sample labels corresponding to each second feature set, and the model parameters are adjusted based on the cross-entropy loss function combined with the gradient descent method, and the iteration termination condition includes at least one of the following: the number of iterations is greater than a first preset threshold, the change in the model parameters is less than a second preset threshold, and the decrease in the loss function is less than a third preset threshold.

[0076] After obtaining the above-mentioned model parameters such as latent vector matrix and feature weights, the evaluation module determines the contribution value of the data set corresponding to the target object in federated learning based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix. This process can be performed as follows:

[0077] For each feature in each first feature set, the value score corresponding to the feature is calculated according to the following formula:

[0078] value(x i )=w i x i +(x i ·v i ) 2 -x i 2 ·v i ·v i

[0079] In the formula, value(x i ) represents the feature x i The corresponding value score;

[0080] Normalize the value scores of all features;

[0081] For each target object, the normalized value scores of each feature in the first feature set corresponding to the target object are summed to obtain the contribution value of the data set corresponding to the target object in federated learning.

[0082] It should be noted that the modules in the device for determining the data value in federated learning in the embodiment of the present application correspond one-to-one to the implementation steps of the method for determining the data value in federated learning in Example 1. Since a detailed description has been given in Example 1, some details not reflected in this embodiment can be referred to Example 1 and will not be elaborated here.

[0083] Example 3

[0084] According to an embodiment of the present application, a computer program product is also provided, which includes a computer program, wherein when the computer program is executed by a processor, the method for determining data value in federated learning in Example 1 is implemented.

[0085] According to an embodiment of the present application, a non-volatile storage medium is also provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the method for determining data value in federated learning in Example 1 by running the computer program.

[0086] According to an embodiment of the present application, a processor is also provided, which is used to run a computer program, wherein the data value determination method in federated learning in Example 1 is executed when the computer program is running.

[0087] According to an embodiment of the present application, an electronic device is also provided, which includes: a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the data value determination method in federated learning in Example 1 through the computer program.

[0088] Specifically, the following steps are executed when the computer program is running: multiple data sets corresponding to multiple target objects participating in federated learning are obtained, and a first feature set corresponding to each data set is determined; multiple random feature extractions with replacement are performed on all features in the multiple feature sets to obtain multiple second feature sets, each second feature set is used as a training sample, and federated learning results for the multiple first feature sets are obtained, and a sample label for each training sample is determined based on the federated learning results; a data value assessment model based on a factorization machine is trained using multiple training samples and sample labels to determine model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in multidimensional space corresponding to different input features, and the data value assessment model is used to predict the federated learning results based on each input feature and the corresponding feature weights and latent vector matrix; and the contribution value of the data set corresponding to the target object in federated learning is determined based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix.

[0089] As an optional implementation, the electronic device may be in the form of a mobile terminal, a computer terminal or a similar computing device. Figure 3 FIG. 1 shows a hardware structure block diagram of an electronic device for implementing a method for determining data value in federated learning. Figure 3 As shown, the electronic device 30 may include one or more (302a, 302b, ..., 302n are used to illustrate) processors 302 (the processor 302 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission device 306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 3 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 3 More or fewer components as shown, or with Figure 3 Different configurations are shown.

[0090] It should be noted that the one or more processors 302 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the electronic device 30. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0091] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the data value determination method in the federated learning in the embodiment of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, realizing the vulnerability detection method of the above-mentioned application. The memory 304 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 304 may further include a memory remotely arranged relative to the processor 302, and these remote memories may be connected to the electronic device 30 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0092] The transmission device 306 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the electronic device 30. In one example, the transmission device 306 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 306 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0093] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the electronic device 30 .

[0094] The serial numbers of the above embodiments are only for description and do not represent the advantages or disadvantages of the embodiments.

[0095] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0096] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods 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. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0097] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0098] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0099] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0100] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining data value in federated learning, characterized in that: include: Acquire multiple data sets corresponding to multiple target objects participating in federated learning, and determine a first feature set corresponding to each of the data sets; Performing multiple random feature extractions with replacement on all features in the multiple feature sets to obtain multiple second feature sets, taking each of the second feature sets as a training sample, and obtaining federated learning results for the multiple first feature sets, and determining a sample label for each training sample based on the federated learning results; Using a plurality of the training samples and sample labels to train a data value assessment model based on a factorization machine, and determining model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in a multidimensional space corresponding to different input features, and the data value assessment model is used to predict federated learning results based on each input feature and the corresponding feature weights and latent vector matrices; The contribution value of the data set corresponding to the target object in federated learning is determined according to the first feature set corresponding to each target object, the feature weight of each feature in the first feature set, and the latent vector matrix.

2. The method according to claim 1, characterized in that Get multiple data sets corresponding to multiple target objects participating in federated learning, including: Acquire multiple data sets corresponding to multiple target objects participating in federated learning from a privacy computing platform, wherein the privacy computing platform is used to receive the data sets uploaded by each of the target objects and perform privacy encryption processing on the data sets, wherein the privacy encryption processing method includes at least one of the following: symmetric encryption, asymmetric encryption, homomorphic encryption, and differential privacy processing.

3. The method according to claim 1, characterized in that Determining a first feature set corresponding to each of the data sets includes: For each data set, preprocessing the data in the data set, wherein the preprocessing method includes: filling missing values ​​in categorical variable data with unknown, and filling missing values ​​in numerical variable data with zero value or statistical value, wherein the statistical value includes at least one of the following: mean, median, mode; Using a label encoder to encode each of the preprocessed categorical variable data to obtain a first feature corresponding to each of the categorical variable data; and / or using a minimum-maximum normalization encoder to encode each of the preprocessed numerical variable data to obtain a second feature corresponding to each of the numerical variable data; The first features and / or the second features are combined into the first feature set corresponding to the data set.

4. The method according to claim 1, characterized in that Determining a sample label for each training sample based on the federated learning result includes: The federated learning result is used as the sample label of each training sample, and noise is added to the sample labels of a preset proportion of the training samples.

5. The method according to claim 1, characterized in that The model formula of the data value assessment model is expressed as follows: In the formula, x={x i }, i = 1, 2, ..., n, represents the second feature set of the input, n represents the number of features in the second feature set, represents the federated learning result predicted based on the second feature set, is the penalty factor, y represents the corresponding sample label, w i Represents feature x i The corresponding feature weight, v i ,v j Represents the feature x i and feature x j The corresponding latent vector matrix in multidimensional space is, <v i ,v j > represents the latent vector matrix v i and v j The dot product of .

6. The method according to claim 5, characterized in that Using the plurality of training samples and sample labels to train a data value assessment model based on a factorization machine, and determining model parameters, includes: Each of the second feature sets is input into the data value assessment model in turn for iterative training until a preset iteration termination condition is met. During the training process, a cross-entropy loss function is constructed based on the federated learning prediction results corresponding to each of the second feature sets and the sample labels, and the model parameters are adjusted based on the cross-entropy loss function combined with the gradient descent method. The iteration termination condition includes at least one of the following: the number of iterations is greater than a first preset threshold, the change in the model parameters is less than a second preset threshold, and the decrease in the loss function is less than a third preset threshold.

7. The method according to claim 5, characterized in that Determining the contribution value of the data set corresponding to each target object in federated learning according to the first feature set corresponding to each target object, the feature weight of each feature in the first feature set, and a latent vector matrix, includes: For each feature in each of the first feature sets, the value score corresponding to the feature is calculated according to the following formula: value(x i )=w i x i +(x i ·v i ) 2 -x i 2 ·v i ·v i In the formula, value(x i ) represents the feature x i The corresponding value score; Normalizing the value scores of all features; For each of the target objects, the normalized value scores of the features in the first feature set corresponding to the target object are summed to obtain the contribution value of the data set corresponding to the target object in federated learning.

8. A device for determining data value in federated learning, characterized in that: include: An acquisition module, used to acquire multiple data sets corresponding to multiple target objects participating in federated learning, and determine a first feature set corresponding to each of the data sets; A sample processing module, configured to perform multiple random feature extractions with replacement on all features in the plurality of feature sets to obtain multiple second feature sets, use each of the second feature sets as a training sample, obtain federated learning results for the plurality of first feature sets, and determine a sample label for each training sample based on the federated learning results; A training module, used to train a data value assessment model based on a factorization machine using a plurality of the training samples and sample labels, and determine model parameters, wherein the model parameters include: feature weights corresponding to different input features and latent vector matrices in a multidimensional space corresponding to different input features, and the data value assessment model is used to predict federated learning results based on each input feature and the corresponding feature weights and latent vector matrix; An evaluation module is used to determine the contribution value of the data set corresponding to the target object in federated learning based on the first feature set corresponding to each target object, the feature weights of each feature in the first feature set, and the latent vector matrix.

9. A computer program product, characterized in that include: A computer program, wherein when the computer program is executed by a processor, the method for determining data value in federated learning as described in any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the method for determining data value in federated learning as described in any one of claims 1 to 7 through the computer program.

Images