Data identification method and device based on artificial intelligence, computer equipment and medium

By filtering and aggregating the convolutional layer and fully connected layer weight parameters of local devices, a global model is built, which solves the problem of degradation and inefficiency of federated learning models in medical image data recognition, and improves model performance and training efficiency.

CN119068293APending Publication Date: 2024-12-03PING AN TECH (SHENZHEN) CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411042695.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

When using federated learning models for medical image data recognition processing in the medical field, there are problems of performance degradation and inefficient training. This is mainly because the medical image data is a non-independent and homogeneous data set, resulting in the problem of domain transfer and the training process being complicated.

Method used

Receive local models uploaded by local devices, filter the weight parameters of the convolutional layer and fully connected layer, perform verification and standardization processing, use preset aggregation strategies to aggregate model parameters, build a target global model for identification processing, and avoid aggregation batches. Normalize the layer to reduce the amount of communication data and reduce the impact of domain transfer.

Benefits of technology

It effectively improves the performance and training efficiency of the target global model, reduces communication costs, and reduces the impact of domain transfer and uneven data distribution on model performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068293A_ABST
    Figure CN119068293A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of artificial intelligence and the field of digital medical treatment, and relates to an artificial intelligence-based data identification method, which comprises the following steps of: receiving a local model uploaded by local equipment; screening model parameters from the local model; performing verification processing on the model parameters based on the verification strategy; if the model parameters pass the verification, performing standardization processing on the model parameters to obtain specified model parameters; performing aggregation operation on the specified model parameters based on an aggregation strategy to obtain target model parameters; constructing a target global model based on the target model parameters; and performing identification processing on the to-be-processed data based on the target global model to generate a corresponding identification result. The invention further provides a data recognition device based on artificial intelligence, computer equipment and a storage medium. In addition, the target global model may be stored in a blockchain. According to the method, the aggregation strategy is used for performing aggregation operation on the model parameters to construct the target global model, so that the performance and the training efficiency of the target global model are effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and the field of digital medicine, and in particular to artificial intelligence-based data recognition methods, devices, computer equipment and storage media. Background Art

[0002] Federated learning is an emerging basic technology for artificial intelligence. Its design goal is to ensure information security during big data exchange, protect the privacy of terminal data and personal data, and ensure efficient machine learning among multiple participants or computing nodes under the premise of legality and compliance. The model training of federated learning is distributed on a main server (server) and multiple terminals (clients). The data of each terminal is not shared with other terminals and the main server. This method allows each terminal to benefit from the data of other terminals while ensuring the security of its own data.

[0003] At present, federated learning models are also widely used in the medical field to perform recognition and processing of medical image data. However, during the training process of the federated learning model, since medical image data belongs to a non-independent and identically distributed dataset, each local model learns different features, which will lead to the problem of domain shift. In federated learning, due to the different data distribution on each local device, there will be deviations when aggregating these local models in the global model. This deviation may cause the performance of the global model to deteriorate. In addition, when training models on non-independent and identically distributed datasets, the training process becomes more complicated and difficult due to problems such as domain shift and unknown labels, which in turn causes low training efficiency of the federated learning model. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose an artificial intelligence-based data recognition method, device, computer equipment and storage medium to solve the technical problems of performance degradation and low training efficiency in the training process of the federated learning model when the federated learning model is used in the medical field to perform recognition and processing of medical image data.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a data recognition method based on artificial intelligence, which adopts the following technical solutions:

[0006] Receiving a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; the number of the local devices includes multiple;

[0007] Filtering model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer;

[0008] Performing verification processing on the model parameters based on a preset verification strategy;

[0009] If the model parameters pass the verification, the model parameters are subjected to a preset standardization process to obtain corresponding designated model parameters;

[0010] Performing an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters;

[0011] Constructing a corresponding target global model based on the target model parameters;

[0012] The received data to be processed is subjected to recognition processing based on the target global model to generate a recognition result corresponding to the data to be processed.

[0013] Furthermore, the step of filtering out model parameters from the local model specifically includes:

[0014] Traversing each layer of the local model to determine the convolutional layer and the fully connected layer of the local model;

[0015] Extracting weight parameters of the convolutional layer;

[0016] Extracting weight parameters of the fully connected layer;

[0017] The model parameters are obtained based on the weight parameters of the convolutional layer and the weight parameters of the fully connected layer.

[0018] Furthermore, the step of verifying the model parameters based on a preset verification strategy specifically includes:

[0019] Obtain the preset integrity verification rules and format verification rules;

[0020] Performing integrity verification on the model parameters based on the integrity verification rule;

[0021] If the model parameters pass the integrity verification, then performing format consistency verification on the model parameters based on the format verification rules;

[0022] If the model parameters pass the format consistency verification, it is determined that the model parameters pass the verification; otherwise, it is determined that the model parameters fail the verification.

[0023] Furthermore, the step of performing an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters specifically includes:

[0024] Get multiple preset aggregation strategies;

[0025] Determining a target aggregation strategy from all the aggregation strategies;

[0026] Performing an aggregation operation on the specified model parameters based on the target aggregation strategy to obtain corresponding aggregated model parameters;

[0027] The aggregated model parameters are used as the target model parameters.

[0028] Furthermore, the step of constructing a corresponding target global model based on the target model parameters specifically includes:

[0029] Get the preset initial model;

[0030] Perform corresponding replacement processing of model parameters on the initial model based on the target model parameters to obtain a processed initial model;

[0031] Get the preset validation data set;

[0032] Performing performance verification on the processed initial model based on the verification data set;

[0033] If the processed initial model passes the performance verification, the processed initial model is used as the target global model.

[0034] Furthermore, after the step of constructing the corresponding target global model based on the target model parameters, the method further includes:

[0035] Obtaining aggregated statistical data corresponding to the target global model;

[0036] Acquiring performance evaluation data corresponding to the target global model;

[0037] Get the preset communication method;

[0038] Based on the communication method, the target global model, the aggregated statistical data, and the performance evaluation data are sent to the local device.

[0039] Furthermore, after the step of performing preset standardization processing on the model parameters to obtain corresponding designated model parameters, the method further includes:

[0040] Acquire reception time information corresponding to the model parameters;

[0041] Acquire source device information corresponding to the model parameters;

[0042] Determining a target storage medium corresponding to the specified model parameters;

[0043] The designated model parameters, the receiving time information and the source device information are stored based on the target storage medium.

[0044] In order to solve the above technical problems, the embodiment of the present application further provides a data recognition device based on artificial intelligence, which adopts the following technical solution:

[0045] A receiving module, configured to receive a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; and the number of the local devices includes a plurality;

[0046] A screening module, used to screen out model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer;

[0047] A verification module, used for verifying the model parameters based on a preset verification strategy;

[0048] A processing module, configured to perform preset standardization processing on the model parameters to obtain corresponding designated model parameters if the model parameters pass the verification;

[0049] An aggregation module, used to perform an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters;

[0050] A construction module, used for constructing a corresponding target global model based on the target model parameters;

[0051] The recognition module is used to perform recognition processing on the received data to be processed based on the target global model, and generate a recognition result corresponding to the data to be processed.

[0052] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0053] Receiving a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; the number of the local devices includes multiple;

[0054] Filtering model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer;

[0055] Performing verification processing on the model parameters based on a preset verification strategy;

[0056] If the model parameters pass the verification, the model parameters are subjected to a preset standardization process to obtain corresponding designated model parameters;

[0057] Performing an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters;

[0058] Building a corresponding target global model based on the target model parameters;

[0059] The received data to be processed is subjected to recognition processing based on the target global model to generate a recognition result corresponding to the data to be processed.

[0060] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0061] Receiving a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; the number of the local devices includes multiple;

[0062] Filtering model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer;

[0063] Performing verification processing on the model parameters based on a preset verification strategy;

[0064] If the model parameters pass the verification, the model parameters are subjected to a preset standardization process to obtain corresponding designated model parameters;

[0065] Performing an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters;

[0066] Building a corresponding target global model based on the target model parameters;

[0067] The received data to be processed is subjected to recognition processing based on the target global model to generate a recognition result corresponding to the data to be processed.

[0068] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0069] The present application first receives a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; the number of the local devices includes multiple; then the model parameters are screened out from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer; then the model parameters are verified based on a preset verification strategy; if the model parameters pass the verification, the model parameters are subjected to preset standardization to obtain corresponding designated model parameters; subsequently, the designated model parameters are aggregated based on a preset aggregation strategy to obtain aggregated target model parameters; further, a corresponding target global model is constructed based on the target model parameters; finally, the received data to be processed is recognized based on the target global model to generate a recognition result corresponding to the data to be processed. This application filters out model parameters that only contain weight parameters of the convolution layer and weight parameters of the fully connected layer from the local model uploaded by the local device, and after detecting that the model parameters have passed the verification based on the verification strategy, performs preset standardization processing on the model parameters to obtain corresponding designated model parameters, and then performs aggregation operation on the designated model parameters based on the preset aggregation strategy to obtain aggregated target model parameters, and further builds a corresponding target global model based on the target model parameters, and then performs recognition processing on the received data to be processed based on the target global model to generate a recognition result corresponding to the data to be processed. This application only aggregates the weight parameters of the convolution layer and the weight parameters of the fully connected layer of the local model that has passed the verification and standardization processing, but does not aggregate the parameters of the batch normalization layer, which can effectively avoid aggregating the batch normalization layer during the communication round of building the target global model, reducing the amount of data for each communication, reducing the communication cost, and effectively reducing the impact of domain transfer problems and data distribution imbalance problems on the performance of the target global model, thereby effectively improving the performance of the target global model and improving the training efficiency of the target global model. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0071] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0072] Figure 2 A flowchart of an embodiment of an artificial intelligence-based data recognition method according to the present application;

[0073] Figure 3 is a structural schematic diagram of an embodiment of an artificial intelligence-based data recognition device according to the present application;

[0074] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0076] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0077] 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 accompanying drawings.

[0078] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0079] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0080] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, and the like.

[0081] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0082] It should be noted that the artificial intelligence-based data recognition method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based data recognition device is generally set in the server / terminal device.

[0083] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0084] Continue to refer Figure 2 , showing a flow chart of an embodiment of an artificial intelligence-based data recognition method according to the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The artificial intelligence-based data recognition method provided in the embodiment of the present application can be applied to any scenario where data recognition processing is required, and the artificial intelligence-based data recognition method can be applied to products in these scenarios, for example, financial data recognition processing in the financial insurance field. The artificial intelligence-based data recognition method comprises the following steps:

[0085] Step S201, receiving a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; and the number of the local devices includes multiple.

[0086] In this embodiment, the electronic device (e.g. Figure 1The server / terminal device shown in the figure) can receive the local model uploaded by the local device through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, Wi-Fi connection, Bluetooth connection, Wi MAX connection, Zigbee connection, UWB (Ultra Wi DeBand) connection, and other wireless connection methods currently known or to be developed in the future. Among them, on each local device participating in the training, a model related to the current task and pre-trained on a large-scale data set is selected. Exemplarily, in image recognition tasks, pre-trained models such as ResNet (deep residual network) and VGG (Visual Geometry Group) convolutional neural network can be selected; in natural language processing tasks, pre-trained models such as BERT (Bidi Rectification Encoder Representations from Transformers) language representation model and GPT (Generate Pre-Trained) deep learning model can be selected.

[0087] Specifically, the loading process of the pre-trained model includes: collecting non-independent and identically distributed data sets on the local device. These data sets have certain similarities with the data sets trained by the pre-trained model and contain local data distribution characteristics. Then the data set is pre-processed, including data cleaning, labeling, normalization and other operations to ensure data quality. After that, the pre-trained model is loaded using an appropriate framework by reading the model weights and architecture from the model library or local file. Follow-up: Modify the output layer of the pre-trained model according to the task requirements of the local data set. Exemplarily, in a classification task, if the output layer of the pre-trained model is for a specific category, and the local data set has a different number of categories, it is necessary to adjust the number of neurons in the output layer. The fine-tuning process of the pre-trained model includes: in order to maintain the stability of the pre-trained model in low-level feature extraction, first freeze part of the layers of the pre-trained model (such as the first few layers of the convolutional layer). These layers have learned general feature representations during the pre-training process and are still useful for new tasks. Then select a suitable optimizer (such as Adam, SGD, etc.) and learning rate. Since fine-tuning is performed on the basis of a pre-trained model, the learning rate is usually smaller than when training from scratch to prevent the destruction of the feature representation that has been learned. The pre-trained model is then trained using a local data set. During the training process, only the weights of the unfrozen layers are updated, while the weights of the frozen layers remain unchanged. Specifically, the weights can be updated using a mini-batch gradient descent or other optimization algorithm. In addition, during the training process, the performance of the model is monitored through a validation set, and the learning rate, optimizer parameters, etc. are adjusted as needed to obtain a corresponding local model. Subsequently, the model weights and model architecture of the fine-tuned local model are saved locally or in the cloud for subsequent use or deployment. This embodiment effectively enhances the generalization ability and performance of the generated local model based on the combination strategy of the pre-trained model and local fine-tuning, so that it can better adapt to data with different distributions.

[0088] Step S202, filtering out model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer.

[0089] In this embodiment, the specific implementation process of filtering out model parameters from the local model will be described in further detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0090] Step S203: verifying the model parameters based on a preset verification strategy.

[0091] In this embodiment, the specific implementation process of the verification process of the model parameters based on the preset verification strategy will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0092] Step S204: If the model parameters pass the verification, the model parameters are subjected to a preset standardization process to obtain corresponding designated model parameters.

[0093] In this embodiment, the model parameters are standardized according to the actual aggregation requirements to ensure that the model parameters can be uniformly processed by subsequent aggregation operations. The standardization process may include data type conversion, dimension adjustment, etc. Specifically, a corresponding standardization script may be written according to specific tasks and data set requirements, and then the received model parameters are converted and adjusted accordingly using the standardization script to obtain the corresponding specified model parameters.

[0094] Step S205 , performing an aggregation operation on the designated model parameters based on a preset aggregation strategy to obtain aggregated target model parameters.

[0095] In this embodiment, the specific implementation process of performing the aggregation operation on the specified model parameters based on the preset aggregation strategy to obtain the aggregated target model parameters will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0096] Step S206: construct a corresponding target global model based on the target model parameters.

[0097] In this embodiment, the specific implementation process of constructing the corresponding target global model based on the target model parameters will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated on here.

[0098] Step S207: performing recognition processing on the received data to be processed based on the target global model, and generating a recognition result corresponding to the data to be processed.

[0099] In this embodiment, in the medical field, due to the special properties of medical image data, the impact of non-independent and identically distributed data sets on model training is more significant. Medical image data usually comes from different institutions and equipment, and there are differences between these institutions and equipment, such as imaging technology, scanning parameters, acquisition conditions, etc. These differences will lead to different data distributions on different institutions and equipment, making it difficult to generalize the trained model to other institutions and equipment. Specifically, the present application can be applied in the field of digital medicine. The above-mentioned data to be processed is specifically medical image data. The medical image data can be input into the target global model, and the medical image data can be recognized and processed by the target global model, and then the recognition result matching the medical image data is output,

[0100] This application filters out model parameters that only contain weight parameters of the convolution layer and weight parameters of the fully connected layer from the local model uploaded by the local device, and after detecting that the model parameters have passed the verification based on the verification strategy, performs preset standardization processing on the model parameters to obtain corresponding designated model parameters, and then performs aggregation operation on the designated model parameters based on the preset aggregation strategy to obtain aggregated target model parameters, and further builds a corresponding target global model based on the target model parameters, and then performs recognition processing on the received data to be processed based on the target global model to generate a recognition result corresponding to the data to be processed. This application only aggregates the weight parameters of the convolution layer and the weight parameters of the fully connected layer of the local model that has passed the verification and standardization processing, but does not aggregate the parameters of the batch normalization layer, which can effectively avoid aggregating the batch normalization layer during the communication round of building the target global model, reducing the amount of data for each communication, reducing the communication cost, and effectively reducing the impact of domain transfer problems and data distribution imbalance problems on the performance of the target global model, thereby effectively improving the performance of the target global model and improving the training efficiency of the target global model.

[0101] In some optional implementations, step S202 includes the following steps:

[0102] Each layer of the local model is traversed to determine the convolutional layer and the fully connected layer of the local model.

[0103] In this embodiment, after the local model is obtained, each layer of the local model is traversed and the convolutional layer, the fully connected layer and the batch normalization layer in the local model are distinguished.

[0104] Extract the weight parameters of the convolutional layer.

[0105] In this embodiment, the weight parameters of the convolutional layer specifically refer to the weight and bias parameters of the convolutional layer.

[0106] Extract the weight parameters of the fully connected layer.

[0107] In this embodiment, the weight parameters of the fully connected layer refer to the weight and bias parameters of the fully connected layer.

[0108] The model parameters are obtained based on the weight parameters of the convolutional layer and the weight parameters of the fully connected layer.

[0109] In this embodiment, the obtained weight parameters of the convolutional layer and the weight parameters of the fully connected layer can be used together as model parameters of the above-mentioned local model.

[0110] This application determines the convolution layer and the fully connected layer of the local model by traversing each layer of the local model; then extracts the weight parameters of the convolution layer and the weight parameters of the fully connected layer; and then obtains the model parameters based on the weight parameters of the convolution layer and the weight parameters of the fully connected layer, so as to quickly and intelligently filter out the required model parameters from the local model, thereby improving the efficiency of obtaining model parameters. In addition, by subsequently aggregating only the weight parameters of the convolution layer and the weight parameters of the fully connected layer of the obtained local model without aggregating the parameters of the batch normalization layer, it is possible to avoid aggregating the batch normalization layer during the communication round of constructing the target global model, thereby reducing the amount of data for each communication, reducing the communication cost, and effectively reducing the impact of domain transfer problems and data distribution imbalance problems on the performance of the target global model, thereby improving the performance and stability of the target global model.

[0111] In some optional implementations of this embodiment, step S203 includes the following steps:

[0112] Get the preset integrity verification rules and format verification rules.

[0113] In this embodiment, the above integrity verification rules are pre-built according to actual integrity verification requirements to check whether the received model parameters are complete and have no missing or damaged parts. The above format verification rules are pre-built according to actual format verification requirements to verify whether the received model parameters are consistent with the expected format, and the expected format includes the structure, data type, dimension, etc. of the standardized parameters.

[0114] The model parameters are integrity verified based on the integrity verification rules.

[0115] In this embodiment, the above-mentioned integrity verification rules may specifically include file integrity verification rules or data integrity verification rules. File integrity verification rules include: if the model parameters are sent in the form of a file, use a hash function (such as SHA-256) to calculate the hash value of the file, and compare it with the locally stored hash value. If the two are the same, it is determined that the model parameters have passed the integrity verification, otherwise it is determined that the model parameters have not passed the integrity verification. Data integrity verification rules: If the model parameters are sent as serialized data, check whether their length and structure meet expectations, or whether they contain necessary fields and types. If the length and structure of the above-mentioned model parameters meet expectations, or contain necessary fields and types, it is determined that the model parameters have passed the integrity verification, otherwise it is determined that the model parameters have not passed the integrity verification.

[0116] If the model parameters pass the integrity verification, format consistency verification is performed on the model parameters based on the format verification rules.

[0117] In this embodiment, the process of format consistency verification may include: pre-defining a template or specification of model parameters, including the name, type, shape and other information of each parameter. Then matching the received model parameters with the template to check whether the name, type, shape and other information of each parameter are consistent with the template. If they are consistent, it is determined that the model parameters have passed the format consistency verification, otherwise it is determined that the model parameters have not passed the format consistency verification. If inconsistent parameter formats are detected, error information may be recorded and the corresponding local device may be notified to make corrections.

[0118] If the model parameters pass the format consistency verification, it is determined that the model parameters pass the verification; otherwise, it is determined that the model parameters fail the verification.

[0119] In this embodiment, the model parameters are determined to have passed the verification only if it is detected that the model parameters have passed both the integrity verification and the format consistency verification. If at least one of the model parameters (such as the integrity verification or the format consistency verification) fails, the model parameters are determined to have failed the verification.

[0120] After filtering out the model parameters from the local model, the present application uses integrity verification rules and format verification rules to perform integrity verification and format consistency verification on the model parameters respectively. Only when it is detected that the model parameters have passed both integrity verification and format consistency verification, will the model parameters be determined to have passed the verification. By performing verification processing on the model parameters, the present application can ensure that the received model parameters are complete, consistent, and meet expectations, which is conducive to ensuring the accuracy of subsequent aggregation operations.

[0121] In some optional implementations, step S205 includes the following steps:

[0122] Get multiple preset aggregation strategies.

[0123] In this embodiment, the above-mentioned multiple aggregation strategies specifically include simple average, weighted average and federal average aggregation strategies. The simple average aggregation strategy includes: directly calculating the arithmetic mean of the model parameters of all local models as the parameters of the global model. This method is simple and direct, but may be affected by uneven data distribution. The weighted average aggregation strategy includes: before averaging, weighting the model parameters according to the quality of each local model, such as the accuracy on the validation set or the amount of training data. This method can more fairly reflect the contribution of each local device. The federal average aggregation strategy is a special weighted average method that takes into account the differences in local data updates and use, which helps to improve the convergence and performance of the global model.

[0124] A target aggregation strategy is determined from all the aggregation strategies.

[0125] In this embodiment, a strategy can be randomly selected from all the aggregation strategies and used as the target aggregation strategy. Preferably, the federated average aggregation strategy is used as the target aggregation strategy.

[0126] An aggregation operation is performed on the specified model parameters based on the target aggregation strategy to obtain corresponding aggregated model parameters.

[0127] In this embodiment, according to the aggregation processing steps in the selected target aggregation strategy, an aggregation operation may be performed on the specified model parameters to obtain corresponding aggregated model parameters.

[0128] The aggregated model parameters are used as the target model parameters.

[0129] In this embodiment, the target model parameters include weight parameters of the aggregated convolutional layer obtained after the aggregation operation, and weight parameters of the aggregated fully connected layer.

[0130] This application determines the target aggregation strategy from a plurality of preset aggregation strategies, and then uses the target aggregation strategy to perform aggregation operations on the specified model parameters to obtain the corresponding target model parameters, thereby achieving rapid and accurate completion of the aggregation operation on the specified model parameters, and ensuring the data accuracy of the generated target model parameters. In addition, by only aggregating the weight parameters of the convolutional layer and the weight parameters of the fully connected layer of the obtained local model, without aggregating the parameters of the batch normalization layer, it is possible to avoid aggregating the batch normalization layer during the communication round of building the target global model, thereby reducing the communication overhead, and effectively reducing the impact of the domain transfer problem and the uneven data distribution problem on the performance of the target global model subsequently constructed based on the target model parameters.

[0131] In some optional implementations, step S206 includes the following steps:

[0132] Get the preset initial model.

[0133] In this embodiment, the initial model may be a local model uploaded by any local device, or a pre-trained model may be used.

[0134] Based on the target model parameters, corresponding replacement processing of model parameters is performed on the initial model to obtain a processed initial model.

[0135] In this embodiment, the weight parameters of the convolution layer and the weight parameters of the fully connected layer in the initial model can be replaced by using the target model parameters, that is, the weight parameters of the aggregated convolution layer obtained through the aggregation operation, and the weight parameters of the aggregated fully connected layer, and the weight parameters of the unaggregated batch normalization layer in the initial model remain unchanged. Among them, although the parameters of the batch normalization layer do not participate in the aggregation, they still play an important role in the model training and reasoning process.

[0136] Get the preset validation dataset.

[0137] In this embodiment, data with a target ratio can be randomly selected from the non-IID data set and used as the verification data set. There is no specific limitation on the value of the target ratio, for example, 0.35 can be selected.

[0138] The performance of the processed initial model is verified based on the verification data set.

[0139] In this embodiment, after the processed initial model is generated, the performance of the processed initial model can be further verified by evaluating it in the above-mentioned verification data set to determine whether the performance of the processed initial model meets the preset performance requirements. If the performance of the processed initial model meets the above-mentioned preset performance requirements, it is determined that the processed initial model has passed the performance verification, and if the performance of the processed initial model does not meet the above-mentioned preset performance requirements, it is determined that the processed initial model has not passed the performance verification. Among them, the above-mentioned preset performance requirements can be set according to the actual model verification requirements, for example, they can be set to accuracy thresholds, recall thresholds, and so on.

[0140] If the processed initial model passes the performance verification, the processed initial model is used as the target global model.

[0141] In this embodiment, if the processed initial model passes the performance verification, it means that the performance of the processed initial model meets the requirements, and the processed initial model will be used as the target global model. If the processed initial model fails the performance verification, it means that the performance of the processed initial model does not meet the requirements, and the aggregation strategy or fine-tuning process can be adjusted according to the evaluation results until a global model with performance that meets the requirements is obtained.

[0142] The present application obtains a preset initial model, and performs corresponding replacement processing on the model parameters of the initial model based on the target model parameters. After obtaining the processed initial model, the application further intelligently performs performance verification on the processed initial model based on the obtained preset verification data set. Only after it is detected that the processed initial model passes the performance verification will the processed initial model be used as the target global model, thereby effectively ensuring the quality and reliability of the target global model.

[0143] In some optional implementations of this embodiment, after step S206, the electronic device may further perform the following steps:

[0144] Obtain aggregate statistics corresponding to the target global model.

[0145] In this embodiment, the above-mentioned aggregated statistical data refers to statistical data recorded in the process of performing an aggregation operation on the specified model parameters using an aggregation strategy to obtain aggregated target model parameters.

[0146] Acquire performance evaluation data corresponding to the target global model.

[0147] In this embodiment, the above-mentioned performance evaluation data may refer to the performance evaluation data obtained after using the above-mentioned verification data set to perform performance evaluation processing on the target global model.

[0148] Get the preset communication method.

[0149] In this embodiment, the above communication method specifically adopts the same method as the designated communication method for the local device to upload the local model to the server.

[0150] Based on the communication method, the target global model, the aggregated statistical data, and the performance evaluation data are sent to the local device.

[0151] In this embodiment, the target global model, the aggregated statistical data, and the performance evaluation data are packaged to obtain packaged data, and then the packaged data is sent to each local device using the above communication method.

[0152] This application obtains the aggregated statistical data and performance evaluation data corresponding to the target global model, and then based on a preset communication method, sends the target global model, the aggregated statistical data and the performance evaluation data to the local device, so as to quickly and smoothly complete the distribution processing of the target global model. In addition, when sending the target global model, by attaching the aggregated statistical data and performance evaluation data corresponding to the target global model, the sending intelligence of the target global model is improved, and it is beneficial for relevant personnel to be able to consult the aggregated process data of the target global model, thereby improving the work experience of relevant personnel.

[0153] In some optional implementations of this embodiment, after step S204, the electronic device may further perform the following steps:

[0154] Acquire reception time information corresponding to the model parameters.

[0155] In this embodiment, the above-mentioned receiving time information refers to the time information synchronously recorded when the server receives the local model including the model parameters uploaded locally.

[0156] Obtain source device information corresponding to the model parameters.

[0157] In this embodiment, the source device information refers to the device information of the local device corresponding to the model parameters recorded by the server when receiving the local model containing the model parameters uploaded locally. The source device information may include the device name or device number of the local device.

[0158] A target storage medium corresponding to the specified model parameters is determined.

[0159] In this embodiment, there is no specific limitation on the selection of the above-mentioned target storage medium. For example, any one of a blockchain, a file system, a database, and a network disk may be used.

[0160] The designated model parameters, the receiving time information and the source device information are stored based on the target storage medium.

[0161] In this embodiment, the designated model parameters, the receiving time information and the source device information are converted into a target storage format, and then the converted designated model parameters, the receiving time information and the source device information are stored in the target storage medium. The target storage format may include a file format, a data table format, etc.

[0162] This application obtains the receiving time information and source device information corresponding to the model parameters, and then uses the target storage medium corresponding to the specified model parameters to store and process the specified model parameters, the receiving time information, and the source device information. This application ensures the data security of the specified model parameters by storing the specified model parameters that have been verified and standardized. In addition, the receiving time information and source device information corresponding to the specified model parameters will also be intelligently recorded, which is conducive to subsequent tracking and auditing, and improves the intelligence of information storage.

[0163] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0164] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned target global model, the above-mentioned target global model can also be stored in a node of a blockchain.

[0165] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0166] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0167] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0168] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0169] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0170] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a data recognition device based on artificial intelligence, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0171] like Figure 3 As shown, the artificial intelligence-based data identification device 300 described in this embodiment includes: a receiving module 301, a screening module 302, a verification module 303, a processing module 304, an aggregation module 305, a construction module 306 and an identification module 307. Among them:

[0172] The receiving module 301 is used to receive a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; the number of the local devices includes multiple;

[0173] A screening module 302, used to screen out model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer;

[0174] Verification module 303, used for verifying the model parameters based on a preset verification strategy;

[0175] The processing module 304 is used to perform a preset standardization process on the model parameters to obtain corresponding designated model parameters if the model parameters pass the verification;

[0176] Aggregation module 305, used to perform aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters;

[0177] A construction module 306 is used to construct a corresponding target global model based on the target model parameters;

[0178] The recognition module 307 is used to perform recognition processing on the received data to be processed based on the target global model, and generate a recognition result corresponding to the data to be processed.

[0179] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0180] In some optional implementations of this embodiment, the screening module 302 includes:

[0181] A traversal submodule, used to traverse each layer of the local model to determine the convolutional layer and the fully connected layer of the local model;

[0182] A first extraction submodule, used to extract weight parameters of the convolutional layer; and

[0183] A second extraction submodule, used to extract the weight parameters of the fully connected layer;

[0184] A generation submodule is used to obtain the model parameters based on the weight parameters of the convolutional layer and the weight parameters of the fully connected layer.

[0185] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0186] In some optional implementations of this embodiment, the verification module 303 includes:

[0187] A first acquisition submodule is used to acquire preset integrity verification rules and format verification rules;

[0188] A first verification submodule, configured to perform integrity verification on the model parameters based on the integrity verification rule;

[0189] A second verification submodule, configured to perform format consistency verification on the model parameters based on the format verification rules if the model parameters pass the integrity verification;

[0190] The determination submodule is used to determine that the model parameters have passed the verification if the model parameters have passed the format consistency verification, and otherwise determine that the model parameters have not passed the verification.

[0191] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0192] In some optional implementations of this embodiment, the aggregation module 305 includes:

[0193] The second acquisition submodule is used to acquire multiple preset aggregation strategies;

[0194] A first determination submodule, configured to determine a target aggregation strategy from all the aggregation strategies;

[0195] An aggregation submodule, used to perform an aggregation operation on the specified model parameters based on the target aggregation strategy to obtain corresponding aggregated model parameters;

[0196] The second determining submodule is used to use the aggregated model parameters as the target model parameters.

[0197] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0198] In some optional implementations of this embodiment, the construction module 306 includes:

[0199] The third acquisition submodule is used to acquire a preset initial model;

[0200] A replacement submodule, used for performing corresponding replacement processing of model parameters on the initial model based on the target model parameters to obtain a processed initial model;

[0201] The fourth acquisition submodule is used to acquire a preset verification data set;

[0202] A third verification submodule, used to perform performance verification on the processed initial model based on the verification data set;

[0203] The third determination submodule is used to use the processed initial model as the target global model if the processed initial model passes the performance verification.

[0204] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0205] In some optional implementations of this embodiment, the artificial intelligence-based data recognition device further includes:

[0206] A first acquisition module, configured to acquire aggregated statistical data corresponding to the target global model;

[0207] A second acquisition module is used to acquire performance evaluation data corresponding to the target global model;

[0208] A third acquisition module is used to acquire a preset communication method;

[0209] A sending module is used to send the target global model, the aggregated statistical data and the performance evaluation data to the local device based on the communication method.

[0210] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0211] In some optional implementations of this embodiment, the artificial intelligence-based data recognition device further includes:

[0212] A fourth acquisition module, used to acquire reception time information corresponding to the model parameters;

[0213] A fifth acquisition module, used to acquire source device information corresponding to the model parameters;

[0214] A determination module, used to determine a target storage medium corresponding to the specified model parameters;

[0215] A storage module is used to store and process the specified model parameters, the receiving time information and the source device information based on the target storage medium.

[0216] In this embodiment, the operations performed by the above modules or units correspond one-to-one to the steps of the artificial intelligence-based data recognition method in the aforementioned implementation mode, and will not be repeated here.

[0217] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0218] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0219] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0220] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of a data recognition method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0221] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run the computer-readable instructions stored in the memory 41 or process data, such as running the computer-readable instructions of the artificial intelligence-based data recognition method.

[0222] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0223] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the artificial intelligence-based data recognition method as described above.

[0224] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0225] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A data recognition method based on artificial intelligence, characterized in that: The steps include: Receiving a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; the number of the local devices includes multiple; Filtering model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer; Performing verification processing on the model parameters based on a preset verification strategy; If the model parameters pass the verification, the model parameters are subjected to a preset standardization process to obtain corresponding designated model parameters; Performing an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters; Constructing a corresponding target global model based on the target model parameters; The received data to be processed is subjected to recognition processing based on the target global model to generate a recognition result corresponding to the data to be processed.

2. The artificial intelligence-based data recognition method according to claim 1, characterized in that: The step of filtering out model parameters from the local model specifically includes: Traversing each layer of the local model to determine the convolutional layer and the fully connected layer of the local model; Extracting weight parameters of the convolutional layer; Extracting weight parameters of the fully connected layer; The model parameters are obtained based on the weight parameters of the convolutional layer and the weight parameters of the fully connected layer.

3. The artificial intelligence-based data recognition method according to claim 1, characterized in that: The step of verifying the model parameters based on a preset verification strategy specifically includes: Obtain preset integrity verification rules and format verification rules; Performing integrity verification on the model parameters based on the integrity verification rule; If the model parameters pass the integrity verification, then performing format consistency verification on the model parameters based on the format verification rules; If the model parameters pass the format consistency verification, it is determined that the model parameters pass the verification; otherwise, it is determined that the model parameters fail the verification.

4. The artificial intelligence-based data recognition method according to claim 1, characterized in that: The step of performing an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters specifically includes: Get multiple preset aggregation strategies; Determining a target aggregation strategy from all the aggregation strategies; Performing an aggregation operation on the specified model parameters based on the target aggregation strategy to obtain corresponding aggregated model parameters; The aggregated model parameters are used as the target model parameters.

5. The artificial intelligence-based data recognition method according to claim 1, characterized in that: The step of constructing a corresponding target global model based on the target model parameters specifically includes: Get the preset initial model; Perform corresponding replacement processing of model parameters on the initial model based on the target model parameters to obtain a processed initial model; Get the preset validation data set; Performing performance verification on the processed initial model based on the verification data set; If the processed initial model passes the performance verification, the processed initial model is used as the target global model.

6. The artificial intelligence-based data recognition method according to claim 1, characterized in that: After the step of constructing the corresponding target global model based on the target model parameters, the method further includes: Obtaining aggregated statistical data corresponding to the target global model; Acquiring performance evaluation data corresponding to the target global model; Get the preset communication method; Based on the communication method, the target global model, the aggregated statistical data, and the performance evaluation data are sent to the local device.

7. The artificial intelligence-based data recognition method according to claim 1, characterized in that: After the step of performing preset standardization processing on the model parameters to obtain corresponding designated model parameters, the method further includes: Acquire reception time information corresponding to the model parameters; Acquire source device information corresponding to the model parameters; Determining a target storage medium corresponding to the specified model parameters; The designated model parameters, the receiving time information and the source device information are stored based on the target storage medium.

8. A data recognition device based on artificial intelligence, characterized in that: include: A receiving module, configured to receive a local model uploaded by a local device; wherein the local model is a model obtained by fine-tuning a pre-trained model on the local device using respective non-independent and identically distributed data sets; and the number of the local devices includes a plurality; A screening module, used to screen out model parameters from the local model; wherein the model parameters only include weight parameters of the convolutional layer and weight parameters of the fully connected layer; A verification module, used for verifying the model parameters based on a preset verification strategy; A processing module, configured to perform preset standardization processing on the model parameters to obtain corresponding designated model parameters if the model parameters pass the verification; An aggregation module, used to perform an aggregation operation on the specified model parameters based on a preset aggregation strategy to obtain aggregated target model parameters; A construction module, used for constructing a corresponding target global model based on the target model parameters; The recognition module is used to perform recognition processing on the received data to be processed based on the target global model, and generate a recognition result corresponding to the data to be processed.

9. A computer device comprising a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the artificial intelligence-based data recognition method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the artificial intelligence-based data recognition method according to any one of claims 1 to 7.

Citation Information

Patent Citations

  • Verifiable federated learning method based on BLS signature and gradient rarefaction

    CN117035063A

  • Anonymization processing method and system for federated learning

    CN117614688A

  • Federal learning system and method for medical data, storage medium and program product

    CN117634650A

  • Big federal model-based clinical auxiliary decision-making method and system for co-disease among complex diseases

    CN117954082A