Federal learning-based feature exploration method and device, equipment and storage medium
By aggregating and updating model parameters in federated learning, performing feature extraction and loss value calculation, the optimal feature vector of suspicious samples is automatically explored, which solves the problem of missing features of suspicious samples in federated learning, and improves the accuracy and robustness of the model.
Patent Information
- Application Number
- CN202411540865.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-30
- Publication Date
- 2025-03-14
AI Technical Summary
In federated learning, the missing key features of suspicious samples leads to confusion when distinguishing different categories, reducing their accuracy and robustness. How to explore the optimal feature vector of suspicious samples has become a technical problem that needs to be solved urgently.
By obtaining suspicious samples and local model parameters uploaded by each participant, aggregating local model parameters to obtain global model parameters, updating global model parameters, selecting models using global model parameters for feature extraction, calculating the loss value between the prediction results of the feature vector and the real results, and exploring the optimal feature vector among multiple feature vectors based on the preset feature exploration method.
Automatically explore the optimal eigenvector of suspicious samples, which reduces exploration time, improves the exploration efficiency of eigenvectors, and improves the reliability of eigenvectors, avoids the influence of manual intervention.
Smart Images
Figure CN119623666A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology and can be applied to the field of financial technology. In particular, the present invention relates to a feature exploration method, device, equipment and storage medium based on federated learning. Background Art
[0002] As an innovative distributed machine learning framework, federated learning cleverly combines privacy protection and secure encryption technologies. Its core is to allow various decentralized participants to collaborate on model training without exposing their own private data. In this way, federated learning not only promotes cooperation among multiple parties, but also greatly improves the training efficiency and effectiveness of machine learning models.
[0003] In federated learning, the datasets for model training are usually distributed among multiple participants. Since the datasets come from a wide range of sources and are diverse, some suspicious samples are inevitably mixed in. Suspicious samples usually lack key features, which contain information that is essential for identification or classification. When these key features are missing, the federated learning model will be confused when distinguishing different categories, reducing its accuracy and robustness. Therefore, how to explore the optimal feature vectors of suspicious samples has become a technical problem that needs to be solved urgently. Summary of the invention
[0004] The present invention provides a feature exploration method, device, computer equipment and storage medium based on federated learning to solve the technical problem of exploring the optimal feature vector of suspicious samples.
[0005] First, a feature exploration method based on federated learning is provided, including:
[0006] Obtain suspicious samples and local model parameters uploaded by each participant;
[0007] Aggregating the multiple local model parameters to obtain a global model parameter;
[0008] Updating the original parameters of the preset first global model to the global model parameters;
[0009] selecting the first global model using the global model parameters as the second global model;
[0010] Extract features from the suspicious sample to obtain multiple feature vectors, and obtain a prediction result corresponding to each feature vector output by the second global model;
[0011] The loss value between the predicted result and the actual result corresponding to each of the feature vectors is calculated, and based on a preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors.
[0012] Furthermore, the aggregating the multiple local model parameters to obtain the global model parameters includes:
[0013] Select the average aggregation algorithm or the weighted average aggregation algorithm as the target algorithm;
[0014] Through the target algorithm, multiple local model parameters are aggregated to obtain global model parameters.
[0015] Further, updating the preset original parameters of the first global model to the global model parameters includes:
[0016] Obtaining a preset model file of a first global model, and reading original parameters in the model file;
[0017] Obtain an update instruction, execute the update instruction, and update the original parameters to the global model parameters.
[0018] Further, the selecting the first global model using the global model parameters as the second global model includes:
[0019] Determining whether the value of the global model parameter is within a preset range;
[0020] If the value of the global model parameter is within a preset range, the first global model using the global model parameter is selected as the second global model.
[0021] Furthermore, extracting features from the suspicious sample to obtain a plurality of feature vectors, and obtaining a prediction result corresponding to each feature vector output by the second global model, includes:
[0022] Using different feature extraction models to extract features from the suspicious samples respectively to obtain multiple feature vectors;
[0023] Each of the feature vectors is input into the second global model, and a prediction result corresponding to each of the feature vectors output by the second global model is obtained.
[0024] Furthermore, the calculation of the loss value between the predicted result and the true result corresponding to each of the feature vectors, and the exploration of the optimal feature vector of the suspicious sample from a plurality of the feature vectors based on a preset feature exploration method and the loss value, includes:
[0025] Calculate the loss value between the predicted result and the actual result corresponding to each of the feature vectors by using a preset loss function, wherein the loss function includes one of a cross entropy loss function and an absolute loss function or a combination thereof;
[0026] Among the multiple feature vectors, a feature vector that minimizes the loss value is found as the optimal feature vector of the suspicious sample.
[0027] Furthermore, in the calculation of the loss value between the predicted result and the true result corresponding to each of the feature vectors, after exploring the optimal feature vector of the suspicious sample from the plurality of feature vectors based on a preset feature exploration method and the loss value, the feature exploration method includes:
[0028] A save instruction is obtained, and the save instruction is executed to save the suspicious sample and the optimal feature vector to a preset storage area.
[0029] In a second aspect, a feature exploration device based on federated learning is provided, including:
[0030] The first acquisition module is used to obtain suspicious samples and local model parameters uploaded by each participant;
[0031] An aggregation module, used for aggregating the plurality of local model parameters to obtain a global model parameter;
[0032] An updating module, used for updating the original parameters of the preset first global model to the global model parameters;
[0033] A selection module, configured to select the first global model using the global model parameters as a second global model;
[0034] A second acquisition module is used to extract features from the suspicious sample to obtain multiple feature vectors, and obtain a prediction result corresponding to each feature vector output by the second global model;
[0035] The exploration module is used to calculate the loss value between the predicted result and the actual result corresponding to each of the feature vectors, and based on a preset feature exploration method and the loss value, explore the optimal feature vector of the suspicious sample among the multiple feature vectors.
[0036] In a third aspect, a computer device is provided, including 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 above-mentioned feature exploration method when executing the computer program.
[0037] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned feature exploration method are implemented.
[0038] The present application provides a feature exploration method, device, computer equipment and storage medium based on federated learning, which obtain suspicious samples and local model parameters uploaded by each participant; aggregate multiple local model parameters to obtain global model parameters; update the original parameters of a preset first global model to the global model parameters; select the first global model using the global model parameters as the second global model; extract features of the suspicious samples to obtain multiple feature vectors, and obtain the prediction results corresponding to each feature vector output by the second global model; calculate the loss value between the prediction result corresponding to each feature vector and the true result, and based on the preset feature exploration method and the loss value, Loss value, among the multiple feature vectors, the optimal feature vector of the suspicious sample is explored, and the beneficial effects are in two aspects. On the one hand, the loss value between the predicted result and the actual result corresponding to each feature vector is calculated, and based on the preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors, which solves the problem of how to explore the optimal feature vector of the suspicious sample. Due to automatic exploration, the exploration time of the optimal feature vector of the suspicious sample is reduced, which is beneficial to improve the exploration efficiency of the optimal feature vector of the suspicious sample; on the other hand, it will not be affected by human intervention, so it is beneficial to improve the reliability of the optimal feature vector of the suspicious sample. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0040] Figure 1 is a schematic diagram of an application environment of a feature exploration method in one embodiment of the present invention;
[0041] Figure 2 A schematic diagram of a flow chart of a feature exploration method provided by an embodiment of the present invention;
[0042] Figure 3 yes Figure 1 A schematic flow chart of a specific implementation of step S23;
[0043] Figure 4 yes Figure 1 A schematic flow chart of a specific implementation of step S25;
[0044] Figure 5 yes Figure 1 A schematic flow chart of a specific implementation of step S26;
[0045] Figure 6 is a structural schematic diagram of a feature exploration device in one embodiment of the present invention;
[0046] Figure 7 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;
[0047] Figure 8 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0049] See also Figure 1 , Figure 1 FIG. 1 is a schematic diagram of an application environment of a feature exploration method in an embodiment of the present invention. The feature exploration method provided in an embodiment of the present invention can be applied in the following embodiments: Figure 1 In the application environment, the client is a participant and communicates with the server through the network.
[0050] The server obtains suspicious samples and local model parameters through each client;
[0051] Aggregating the multiple local model parameters to obtain a global model parameter;
[0052] Updating the original parameters of the preset first global model to the global model parameters;
[0053] selecting the first global model using the global model parameters as the second global model;
[0054] Extract features from the suspicious sample to obtain multiple feature vectors, and obtain a prediction result corresponding to each feature vector output by the second global model;
[0055] The loss value between the predicted result and the actual result corresponding to each of the feature vectors is calculated, and based on a preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors.
[0056] In the scheme implemented by the above-mentioned feature exploration method, device, equipment and medium, the beneficial effects are in two aspects. On the one hand, the loss value between the predicted result and the actual result corresponding to each feature vector is calculated, and based on the preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors, which solves the problem of how to explore the optimal feature vector of the suspicious sample. Due to automatic exploration, the exploration time of the optimal feature vector of the suspicious sample is reduced, which is beneficial to improving the exploration efficiency of the optimal feature vector of the suspicious sample; on the other hand, it will not be affected by human intervention, which is beneficial to improving the reliability of the optimal feature vector of the suspicious sample.
[0057] The device running the client is referred to as a client device.
[0058] The device running the server is referred to as the server device.
[0059] The client devices may include but are not limited to various personal computers, laptops, smart phones, tablet computers, and portable wearable devices.
[0060] The server device can be implemented with an independent task database or a task database cluster composed of multiple task databases. The present invention is described in detail below through specific embodiments.
[0061] See also Figure 2 , Figure 2 A flow chart of a feature exploration method provided by an embodiment of the present invention includes the following steps:
[0062] S21, obtaining suspicious samples and local model parameters uploaded by each participant;
[0063] In the framework of federated learning, each participant is usually a device with computing power and data storage capabilities. These participants can be smartphones, IoT sensors, personal computers or other smart terminals. Each of these participants has an independent data set, which may contain rich information such as user's personal behavior, device status, environmental information, etc. Each participant is responsible for training the local model locally using its own data set to generate local model parameters.
[0064] Among them, the participants obtain the confidence of the training samples in the data set through the local model. When the confidence is lower than the preset value, the training samples are marked as suspicious samples.
[0065] Among them, the local model actively excludes input data that does not meet specific conditions or categories through rejection inference technology, thereby improving the accuracy, generalization ability and privacy protection performance of the model.
[0066] Among them, rejection inference technology can improve the accuracy of local models. By rejecting data that obviously does not belong to the target category, the local model can avoid misclassifying these data, thereby improving the accuracy of overall classification. For example, when the local model is a face recognition model, if an image that does not match any known face is input, the local model will reject the image and not recognize it, thus avoiding the occurrence of misclassification.
[0067] S22, aggregating the plurality of local model parameters to obtain a global model parameter;
[0068] The step of aggregating the plurality of local model parameters to obtain a global model parameter includes:
[0069] Select the average aggregation algorithm or the weighted average aggregation algorithm as the target algorithm;
[0070] Through the target algorithm, multiple local model parameters are aggregated to obtain global model parameters.
[0071] Among them, the average aggregation algorithm, as one of the basic methods of data processing, usually refers to adding all the values in a set of data and dividing it by the number of data to obtain an average value. This average value reflects the overall trend or central position of the data, and is especially suitable for situations where the data is relatively evenly distributed. Among them, the weighted average aggregation algorithm is an extension of the average aggregation algorithm, which takes into account the importance or weight of different data points. In this method, each data point is assigned a weight value, which reflects the degree of influence of the data point on the final result. The weighted average is calculated as the sum of the products of all values and their corresponding weights divided by the sum of all weights. This method is particularly useful when the data points have different importance or credibility, because it can more accurately reflect the comprehensive situation of the data.
[0072] S23, updating the preset original parameters of the first global model to the global model parameters;
[0073] Among them, the continuous updating and optimization of global model parameters is the key to improving model performance in federated learning. Through continuous iteration of this process, federated learning can use data scattered across multiple participants to jointly train the first global model while protecting privacy, thereby improving the model's generalization ability and prediction accuracy.
[0074] S24, selecting the first global model using the global model parameters as the second global model;
[0075] The selecting the first global model using the global model parameters as the second global model comprises:
[0076] Determining whether the value of the global model parameter is within a preset range;
[0077] If the value of the global model parameter is within a preset range, the first global model using the global model parameter is selected as the second global model.
[0078] Among them, the global model parameters are obtained by integrating the local model parameter updates of multiple participants, which means that the second global model can make full use of the large amount of data distributed on different participants, so as to learn richer features and more accurate patterns, which helps to improve the generalization ability of the model.
[0079] S25, extracting features from the suspicious sample to obtain a plurality of feature vectors, and obtaining a prediction result corresponding to each of the feature vectors output by the second global model;
[0080] Among them, suspicious samples include but are not limited to data samples in the financial field, data samples of scientific and technological data, and data samples of insurance data.
[0081] S26, calculating the loss value between the predicted result and the actual result corresponding to each of the feature vectors, and exploring the optimal feature vector of the suspicious sample from the multiple feature vectors based on a preset feature exploration method and the loss value.
[0082] In the feature exploration process, different feature vectors are tried, and the loss value of the second global model under each feature vector is calculated. By comparing the loss values, the feature vector that minimizes the loss value, that is, the optimal feature vector, can be explored.
[0083] Wherein, after calculating the loss value between the predicted result and the true result corresponding to each of the feature vectors, based on a preset feature exploration method and the loss value, exploring the optimal feature vector of the suspicious sample among the multiple feature vectors, the feature exploration method includes:
[0084] A save instruction is obtained, and the save instruction is executed to save the suspicious sample and the optimal feature vector to a preset storage area.
[0085] In the embodiment of the present invention, the beneficial effects lie in two aspects. On the one hand, the loss value between the predicted result and the actual result corresponding to each feature vector is calculated, and based on the preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors, thereby solving the problem of how to explore the optimal feature vector of the suspicious sample. Due to the automatic exploration, the exploration time of the optimal feature vector of the suspicious sample is reduced, which is beneficial to improving the exploration efficiency of the optimal feature vector of the suspicious sample. On the other hand, it will not be affected by human intervention, which is beneficial to improving the reliability of the optimal feature vector of the suspicious sample.
[0086] See also Figure 3 , Figure 3 yes Figure 1 A specific implementation flow diagram of step S23 is described in detail as follows:
[0087] S31, obtaining a preset model file of a first global model, and reading original parameters in the model file;
[0088] S32, obtaining an update instruction, executing the update instruction, and updating the original parameters to the global model parameters.
[0089] In an embodiment of the present invention, federated learning avoids directly transmitting sensitive data to the server device and protects the data privacy of the participants. Updating the original parameters to global parameters is a key step in this process because it ensures that the first global model can learn the global data distribution and characteristics without exposing the original data.
[0090] See also Figure 4 , Figure 4 yes Figure 1 A specific implementation flow diagram of step S25 is described in detail as follows:
[0091] S41, using different feature extraction models to extract features from the suspicious samples respectively to obtain multiple feature vectors;
[0092] S42: Input each of the feature vectors into the second global model, and obtain a prediction result corresponding to each of the feature vectors output by the second global model.
[0093] In an embodiment of the present invention, a prediction result corresponding to each of the feature vectors output by the second global model is obtained, and the understanding and analysis capabilities of the second global model for the suspicious sample can be determined through the prediction results.
[0094] See also Figure 5 , Figure 5 yes Figure 1A specific implementation flow diagram of step S26 is described in detail as follows:
[0095] S51, calculating the loss value between the predicted result and the true result corresponding to each of the feature vectors by using a preset loss function, wherein the loss function includes one of a cross entropy loss function and an absolute loss function or a combination thereof;
[0096] Among them, the cross entropy loss function and the absolute loss function are two important loss functions in the field of machine learning. They are each suitable for different types of tasks and have different characteristics.
[0097] Among them, the cross entropy loss function is mainly used in classification problems, especially multi-classification tasks. It guides the model's learning process by measuring the difference between the predicted probability distribution and the true label distribution, so that the probability distribution predicted by the model is as close to the true distribution as possible. The cross entropy loss function has a large penalty for misclassified samples, which helps improve the model's sensitivity to misclassification and classification accuracy.
[0098] Among them, the absolute loss function is widely used in regression problems. It calculates the absolute value of the difference between the predicted value and the true value. It is more robust to outliers or abnormal values because the absolute value loss function will not produce excessively large gradients due to individual abnormal values, which helps to maintain the stability of the model.
[0099] S52, searching among the plurality of feature vectors for a feature vector that minimizes the loss value as the optimal feature vector of the suspicious sample.
[0100] Among the multiple feature vectors, the feature vector that minimizes the loss value is found as the optimal feature vector of the suspicious sample.
[0101] The step of selecting the feature vector with the smallest loss value as the optimal feature vector of the suspicious sample includes:
[0102] A random search algorithm is used to find a feature vector that minimizes the loss value among the multiple feature vectors, and the feature vector that minimizes the loss value is selected as the optimal feature vector of the suspicious sample.
[0103] Among them, random search algorithms are usually easy to understand and implement, and do not require in-depth mathematical analysis of the problem or complex programming skills.
[0104] In an embodiment of the present invention, the feature vector with the smallest loss value is selected as the optimal feature vector of the suspicious sample, which solves the problem of how to explore the optimal feature vector of the suspicious sample. Due to the automatic exploration, the exploration time of the optimal feature vector of the suspicious sample is reduced, which is beneficial to improving the exploration efficiency of the optimal feature vector of the suspicious sample.
[0105] See also Figure 6 , Figure 6 is a schematic diagram of a structure of a feature exploration device in one embodiment of the present invention, Figure 6 As shown, the feature exploration device includes a first acquisition module 101, an aggregation module 102, an update module 103, a selection module 104, a second acquisition module 105, and an exploration module 106. The functional modules are described in detail as follows:
[0106] The first acquisition module 101 is used to acquire suspicious samples and local model parameters uploaded by each participant;
[0107] Aggregation module 102, used for aggregating the plurality of local model parameters to obtain global model parameters;
[0108] An updating module 103, used for updating the original parameters of the preset first global model to the global model parameters;
[0109] A selection module 104, configured to select the first global model using the global model parameters as a second global model;
[0110] The second acquisition module 105 is used to extract features from the suspicious sample to obtain multiple feature vectors, and obtain a prediction result corresponding to each feature vector output by the second global model;
[0111] The exploration module 106 is used to calculate the loss value between the predicted result and the actual result corresponding to each of the feature vectors, and explore the optimal feature vector of the suspicious sample from the multiple feature vectors based on a preset feature exploration method and the loss value.
[0112] In one embodiment, the aggregation module 102 includes:
[0113] A first selection subunit is used to select an average aggregation algorithm or a weighted average aggregation algorithm as a target algorithm;
[0114] The aggregation subunit is used to aggregate the multiple local model parameters through the target algorithm to obtain the global model parameters.
[0115] In one embodiment, the updating module 103 includes:
[0116] A first acquisition subunit is used to acquire a model file of a preset first global model and read original parameters in the model file;
[0117] The updating subunit is used to obtain an updating instruction, execute the updating instruction, and update the original parameters to the global model parameters.
[0118] In one embodiment, the selection module 104 includes:
[0119] A judging subunit, used to judge whether the value of the global model parameter is within a preset range;
[0120] The second selection subunit is used to select the first global model using the global model parameter as the second global model if the value of the global model parameter is within a preset range.
[0121] In one embodiment, the second acquisition module 105 includes:
[0122] The subunit is used to use different feature extraction models to extract features from the suspicious samples respectively to obtain multiple feature vectors;
[0123] The second acquisition subunit is used to input each of the feature vectors into the second global model to obtain the prediction result corresponding to each of the feature vectors output by the second global model.
[0124] In one embodiment, the exploration module 106 includes:
[0125] A calculation subunit, used to calculate the loss value between the predicted result and the true result corresponding to each of the feature vectors by using a preset loss function, wherein the loss function includes one of a cross entropy loss function and an absolute loss function or a combination thereof;
[0126] The exploration subunit is used to find a feature vector that minimizes the loss value among the multiple feature vectors as the optimal feature vector of the suspicious sample.
[0127] In one embodiment, the feature exploration device further includes:
[0128] The saving module is used to obtain a saving instruction, execute the saving instruction, and save the suspicious sample and the optimal feature vector to a preset storage area.
[0129] In the embodiment of the present invention, the beneficial effects lie in two aspects. On the one hand, the loss value between the predicted result and the actual result corresponding to each feature vector is calculated, and based on the preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors, thereby solving the problem of how to explore the optimal feature vector of the suspicious sample. Due to the automatic exploration, the exploration time of the optimal feature vector of the suspicious sample is reduced, which is beneficial to improving the exploration efficiency of the optimal feature vector of the suspicious sample. On the other hand, it will not be affected by human intervention, which is beneficial to improving the reliability of the optimal feature vector of the suspicious sample.
[0130] For the specific limitations of the feature exploration device, please refer to the limitations of the feature exploration method above, which will not be repeated here.
[0131] Each module in the above-mentioned feature exploration device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.
[0132] See also Figure 7 , Figure 7 is a schematic diagram of a computer device in one embodiment of the present invention. In one embodiment, a computer device is provided. The computer device may be a type of server device. The internal structure diagram thereof may be as shown in FIG. Figure 7 The computer device includes a processor, a memory, a network interface and a database connected through a system bus.
[0133] The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, a function or step of a feature exploration method based on federated learning can be implemented on a server-side device.
[0134] See also Figure 8 , Figure 8 is another structural diagram of a computer device in one embodiment of the present invention. In one embodiment, a computer device is provided. The computer device may be another type of server device. Its internal structure diagram may be as shown in FIG. Figure 8As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external task database via a network connection. When the computer program is executed by the processor, a function or step of a feature exploration method based on federated learning can be implemented on a client device.
[0135] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor.
[0136] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant description in the aforementioned method embodiment. In order to avoid repetition, they will not be described one by one here.
[0137] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
[0138] The above description and accompanying drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process and other changes. The embodiments represent possible changes only. Unless explicitly required, separate components and functions are optional, and the order of operation may vary. Parts and subsamples of some embodiments may be included in or replace parts and subsamples of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and are not used to limit the claims. As used in the description of the embodiments and claims, unless the context clearly indicates, the singular forms of "a", "an" and "the" are intended to include plural forms as well. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of listings containing one or more associated ones. In addition, when used in the present application, the term "comprise" and its variants "comprises" and / or comprising refer to the presence of a stated sub-sample, whole, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other sub-samples, wholes, steps, operations, elements, components and / or groups of these. In the absence of further restrictions, the elements defined by the sentence "comprising a ..." do not exclude the presence of other identical elements in the process, method or device comprising the elements. In this article, each embodiment may focus on the differences from other embodiments, and the same and similar parts between the various embodiments may refer to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, then the relevant parts can refer to the description of the method part.
[0139] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software may depend on the specific application and design constraints of the technical solution. Technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present disclosure. Technicians can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0140] In the embodiments disclosed herein, the disclosed methods and products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units can be only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some sub-samples can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to implement this embodiment. In addition, each functional unit in the embodiment of the present disclosure may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit.
[0141] The flowchart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and the block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.
Claims
1. A feature exploration method based on federated learning, characterized in that: include: Obtain suspicious samples and local model parameters uploaded by each participant; Aggregating the multiple local model parameters to obtain a global model parameter; Updating the original parameters of the preset first global model to the global model parameters; selecting the first global model using the global model parameters as the second global model; Extract features from the suspicious sample to obtain multiple feature vectors, and obtain a prediction result corresponding to each feature vector output by the second global model; The loss value between the predicted result and the actual result corresponding to each of the feature vectors is calculated, and based on a preset feature exploration method and the loss value, the optimal feature vector of the suspicious sample is explored among the multiple feature vectors.
2. The feature exploration method according to claim 1, characterized in that: The aggregating the multiple local model parameters to obtain the global model parameters includes: Select the average aggregation algorithm or the weighted average aggregation algorithm as the target algorithm; Through the target algorithm, multiple local model parameters are aggregated to obtain global model parameters.
3. The feature exploration method according to claim 1, characterized in that: The updating of the preset original parameters of the first global model to the global model parameters comprises: Obtaining a preset model file of a first global model, and reading original parameters in the model file; Obtain an update instruction, execute the update instruction, and update the original parameters to the global model parameters.
4. The feature exploration method according to claim 1, characterized in that: The selecting the first global model using the global model parameters as the second global model comprises: Determining whether the value of the global model parameter is within a preset range; If the value of the global model parameter is within a preset range, the first global model using the global model parameter is selected as the second global model.
5. The feature exploration method according to claim 1, characterized in that: The extracting features of the suspicious sample to obtain a plurality of feature vectors, and obtaining a prediction result corresponding to each feature vector output by the second global model, includes: Using different feature extraction models to extract features from the suspicious samples respectively, to obtain multiple feature vectors; Each of the feature vectors is input into the second global model, and a prediction result corresponding to each of the feature vectors output by the second global model is obtained.
6. The feature exploration method according to claim 1, characterized in that: The calculating the loss value between the predicted result and the true result corresponding to each of the feature vectors, and exploring the optimal feature vector of the suspicious sample from the plurality of feature vectors based on a preset feature exploration method and the loss value, includes: Calculate the loss value between the predicted result and the actual result corresponding to each of the feature vectors by using a preset loss function, wherein the loss function includes a cross entropy loss function, an absolute loss function, or a combination thereof; Among the multiple feature vectors, a feature vector that minimizes the loss value is found as the optimal feature vector of the suspicious sample.
7. The feature exploration method according to any one of claims 1 to 6, characterized in that: After calculating the loss value between the predicted result and the true result corresponding to each of the feature vectors, and exploring the optimal feature vector of the suspicious sample from the plurality of feature vectors based on a preset feature exploration method and the loss value, the feature exploration method includes: A save instruction is obtained, and the save instruction is executed to save the suspicious sample and the optimal feature vector to a preset storage area.
8. A feature exploration device based on federated learning, characterized in that: include: The first acquisition module is used to obtain suspicious samples and local model parameters uploaded by each participant; An aggregation module, used for aggregating the plurality of local model parameters to obtain a global model parameter; An updating module, used for updating the original parameters of the preset first global model to the global model parameters; A selection module, configured to select the first global model using the global model parameters as a second global model; A second acquisition module is used to extract features from the suspicious sample to obtain multiple feature vectors, and obtain a prediction result corresponding to each feature vector output by the second global model; The exploration module is used to calculate the loss value between the predicted result and the actual result corresponding to each of the feature vectors, and based on a preset feature exploration method and the loss value, explore the optimal feature vector of the suspicious sample among the multiple feature vectors.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the feature exploration method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the feature exploration method according to any one of claims 1 to 7 are implemented.