An AI model generation method based on artificial intelligence

Abstract

The present invention relates to an artificial intelligence-based AI model generation method, which involves the field of model construction, including: interacting with a first user terminal to obtain model input data attributes, model output data attributes, and a data acquisition source link to obtain a first training data set; when the first data volume of the first training data set is less than a convergence data volume threshold, generating a data acquisition task instruction; publishing the data acquisition task instruction to a data acquisition blockchain, obtaining a first response node up to the Nth response node, and obtaining a second training data set up to the N+1th training data set; training a target AI model based on the first training data set, the second training data set, up to the N+1th training data set; and encrypting and transmitting the topological structure feature information and node parameter feature information of the target AI model to the first user terminal. This method solves the technical problem of the difficulty in sample data collection in the prior art, which leads to the problem of insufficient adaptability of AI models to different scenarios.

Description

Technical Field

[0001] The present invention relates to the field of model building technology, and in particular to an AI model generation method based on artificial intelligence. Background Art

[0002] Training large AI models requires extensive sample data sets, and the accuracy of these data is crucial for ensuring their accuracy. However, in actual application model training, the challenge of limited training samples often arises. Furthermore, the information silo effect further complicates sample data collection, making AI model training and application difficult.

[0003] To solve this problem, existing technologies have adopted methods such as federated learning and compiler compilation. However, the resulting sample training models are highly versatile, but they are still insufficiently practical when facing actual application nodes. Due to the current difficulties in collecting sample data, the AI ​​model has insufficient adaptability to scenarios. Summary of the Invention

[0004] The present invention aims to solve the technical problem that sample data collection is difficult in the prior art, which leads to insufficient adaptability of AI models to different scenarios, and provides an AI model generation method based on artificial intelligence to solve the problem.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] In a first aspect, the present invention provides an AI model generation method based on artificial intelligence, which is applied to a server and includes:

[0007] Interacting with the first user terminal to obtain AI model building requirements, wherein the AI ​​model building requirements include model input data attributes, model output data attributes, and a data acquisition source link;

[0008] Accessing the data collection source link to perform data encryption collection according to the model input data attributes and the model output data attributes to obtain a first training data set;

[0009] When a first data volume of the first training data set is less than a convergence data volume threshold, generating a data acquisition task instruction according to the model input data attribute and the model output data attribute;

[0010] Publishing the data collection task instruction to the data collection blockchain, obtaining the first response node up to the Nth response node;

[0011] Traversing the first response node until the Nth response node performs data encryption collection to obtain a second training data set until the N+1th training data set;

[0012] Constructing a pre-trained AI model based on the second training data set up to the N+1th training data set, and training the pre-trained AI model based on the first training data set to generate a target AI model, wherein the target AI model has topological structure feature information and node parameter feature information;

[0013] The topology structure characteristic information and the node parameter characteristic information are encrypted and transmitted to the first user terminal.

[0014] In a second aspect, the present application provides an electronic device, comprising:

[0015] Memory for storing computer software programs;

[0016] A processor is used to read and execute the computer software program, thereby implementing the AI ​​model generation method based on artificial intelligence described in the first aspect.

[0017] In a third aspect, the present application provides a non-transitory computer-readable storage medium, in which a computer software program is stored. When the computer software program is executed by a processor, it implements an AI model generation method based on artificial intelligence as described in any one of the first aspects.

[0018] The beneficial effects of the present invention are: by receiving the model building requirements from the user side, the first training data set is encrypted and collected according to the data crawling link provided by the user; when the data amount of the first training data set is small, the data collection task quality is generated, and the second training data set is encrypted and collected through the blockchain until the N+1th training data set; then pre-training is performed through the second training data set to the N+1th training data set to obtain a pre-trained AI model with strong versatility, and the first training data set is further used to train the target AI model to obtain an intelligent model with a high degree of integration with the first user end, thereby achieving the technical effect of increasing the amount of training samples and a high degree of integration of AI intelligent models with scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A flowchart of an AI model generation method based on artificial intelligence provided by the present invention;

[0020] Figure 2 A schematic structural diagram of the electronic device provided by the present invention;

[0021] Figure 3 A schematic structural diagram of a computer-readable storage medium provided by the present invention.

[0022] In the accompanying drawings, the components represented by the reference numerals are as follows:

[0023] Electronic device 500 , memory 510 , processor 520 , computer program 511 , computer-readable storage medium 600 , computer program 611 . DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0026] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or illustration". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0027] Example 1:

[0028] like Figure 1 As shown, an embodiment of the present invention provides an AI model generation method based on artificial intelligence, which is applied to a server and includes the following steps:

[0029] Specifically, an embodiment of the present application provides an AI model generation method based on artificial intelligence. The process is executed by a server. The server is a trusted data center. When each user terminal connected to the server needs to build an AI intelligent model, it can send a model building request to the server through the user terminal; and when receiving the data collection task instruction sent by the server, it can choose to respond as the data source according to its own situation.

[0030] S10: Interacting with the first user terminal to obtain AI model building requirements, wherein the AI ​​model building requirements include model input data attributes, model output data attributes, and data acquisition source links;

[0031] Specifically, the first user terminal is any one of several user terminals connected to the server. The AI ​​model building requirements are the constraint information for model building received from the user terminal, which at least includes model input data attributes, model output data attributes and data acquisition source links. Model input data attributes are the attributes of the processed data, such as the vibration amplitude and vibration frequency of model A electrical appliance; model output data attributes are the target data output by the model, such as the failure probability of model A electrical appliance; the data acquisition source link refers to the website used to collect training data, usually the local website of the first user terminal.

[0032] S20: Accessing the data collection source link to perform encrypted data collection based on the model input data attributes and the model output data attributes to obtain a first training data set;

[0033] Specifically, on the website, data corresponding to model input data attributes and model output data attributes should be stored in a one-to-one correspondence. Preferably, crawler technology can be used to access the data collection source link. Based on the model input data attributes and model output data attributes, a historical time zone is set for the past. After data collection, it is encrypted and transmitted back to the server for subsequent calls. Preferably, the historical time zone is user-defined, with a default of three years.

[0034] S30: When the first data volume of the first training data set is less than a convergence data volume threshold, generating a data acquisition task instruction according to the model input data attribute and the model output data attribute;

[0035] Specifically, a convergence data volume threshold is configured by model building professionals on the server to represent the minimum amount of data required for AI model training; the data volume of the first training data set is counted and set as the first data volume.

[0036] When the first data volume is less than the convergence data volume threshold, the first training data set cannot construct the AI ​​model alone. At this time, the data collection task instructions are generated through the model input data attributes and the model output data attributes. Preferably, the data collection task instructions include: description information of the model input data attributes and the model output data attributes, so that each user can know the required data type.

[0037] When the first data volume is greater than or equal to the convergence data volume threshold, the first training data set can be used to construct an AI model independently, and the target AI model can be trained directly using the first data volume and then downloaded to the first user terminal.

[0038] S40: Publishing the data collection task instruction to the data collection blockchain to obtain the first response node up to the Nth response node;

[0039] Specifically, the data collection blockchain refers to a blockchain comprised of member companies, individuals, and other organizations connected to the server. Through the blockchain, the server can issue data collection task instructions. Upon receiving a data collection task instruction, each blockchain node decides whether to respond. If so, it must provide a data collection link. If not, the response can be rejected or ignored, and both are considered non-response. The responding node is stored as the first responding node, continuing through the Nth responding node. Because the data collection blockchain is deployed by companies, individuals, and organizations across various industries, it helps break down information silos and improve sample collection efficiency.

[0040] S50: Traverse the first response node until the Nth response node to perform data encryption collection, and obtain a second training data set until the N+1th training data set;

[0041] Specifically, data is collected based on the data collection links provided by the first response node through the Nth response node, and encrypted data is transmitted back to obtain the second training data set through the N+1th training data set. Leveraging blockchain, information silos are broken down, and encrypted transmission ensures data security.

[0042] S60: Construct a pre-trained AI model based on the second training data set to the N+1th training data set, and train the pre-trained AI model based on the first training data set to generate a target AI model, wherein the target AI model has topological structure feature information and node parameter feature information;

[0043] S70: Encrypt and transmit the topology structure characteristic information and the node parameter characteristic information to the first user terminal.

[0044] Specifically, the second training data set is first used until the N+1th training data set, and conventional model training methods, such as supervised training, semi-supervised training, unsupervised training, etc., are used to train the pre-trained AI model. After the pre-trained AI model converges, the pre-trained AI model is subjected to migration training according to the first training data set to generate a target AI model. Furthermore, the topological structure feature information and node parameter feature information of the target AI model are extracted and encrypted and transmitted to the first user terminal. The first user terminal can build a model based on the topological structure feature information and node parameter feature information. Since the entire model training process is encrypted training in the cloud, there is no need to worry about the leakage of training data, thereby ensuring the data security of each user terminal.

[0045] Furthermore, according to the model input data attributes and the model output data attributes, the data acquisition source link is accessed to perform data encryption acquisition to obtain a first training data set. Step S20 includes the following steps:

[0046] S21: Accessing the data collection source link according to the model input data attributes and the model output data attributes to collect an initial input record data set and an initial output record data set;

[0047] S22: Interacting with the first user terminal to obtain a set of input data attribute deviation thresholds;

[0048] S23: performing cluster analysis on the initial input record data set according to the input data attribute deviation threshold set to obtain multiple clusters of initial input record data;

[0049] S24: clustering the initial output record data set according to the multiple clusters of initial input record data to generate multiple clusters of initial output record data;

[0050] S25: traversing the multiple clusters of initial output record data to perform mode calculation and generate multiple output record data;

[0051] S26: randomly extracting an initial output record data from each of the multiple clusters of initial output record data to obtain a plurality of initial output record data;

[0052] S27: Encrypt and transmit the multiple initial output record data and the multiple output record data back to obtain the first training data set.

[0053] Specifically, the data collection source link is accessed, and the initial input record data set and the initial output record data set are collected one-to-one according to the model input data attributes and the model output data attributes; further, in order to improve the accuracy of the training data, the data is fitted by data clustering and fusion, as detailed below:

[0054] The interactive first user terminal configures the input data attribute deviation threshold set, that is, the deviation preset value representing each input data attribute. When it is greater than or equal to the deviation preset value, the two parameters are considered inconsistent. If it is less than the deviation preset value, the two parameters are considered consistent.

[0055] For the initial input record data set, cluster analysis is performed on the initial input record data set according to the input data attribute deviation threshold set. Any two initial input record data whose total attribute numbers are less than the input data attribute deviation threshold set are set as the same cluster data and added into the multi-cluster initial input record data.

[0056] Furthermore, multiple clusters of initial output record data are traversed to perform mode calculation, specifically, the mode calculation is performed on the output data of the same attribute in each cluster to obtain multiple output record data. An initial output record data is randomly extracted from each of the multiple clusters of initial output record data to obtain multiple initial output record data. The multiple initial output record data and the multiple output record data are encrypted and transmitted back to obtain a first training dataset. Since the individual, one-to-one correspondence between the initial input record dataset and the initial output record dataset may contain accidental errors, cluster analysis can eliminate the impact of these accidental errors and improve the accuracy of subsequent model training.

[0057] Furthermore, cluster analysis is performed on the initial input record data set according to the input data attribute deviation threshold set to obtain multiple clusters of initial input record data. Step S26 includes the following steps:

[0058] S261: Obtaining first initial input record data and second initial input record data of the initial input record data set;

[0059] S262: Compare the first initial input record data and the second initial input record data for the same attribute deviation to obtain an input data deviation set, wherein the input data deviation set corresponds to the input data attribute deviation threshold set in a one-to-one manner;

[0060] S263: When the input data deviation set satisfies the input data attribute deviation threshold set, clustering the first initial input record data and the second initial input record data into the same cluster;

[0061] S264: When any one of the input data deviation sets does not satisfy the input data attribute deviation threshold set, the first initial input record data and the second initial input record data are clustered into a heterogeneous cluster.

[0062] Specifically, any comparison in cluster analysis is exemplified as follows: First and second initial input record data are extracted from the initial input record data set; the data deviations for each attribute of the first and second initial input record data are compared one by one, and stored as an input data deviation set. Specifically, the difference between the first and second initial input record data is calculated, the absolute value is taken, and stored as the data deviation.

[0063] When the input data deviation sets are all smaller than the corresponding input data attribute deviation threshold sets, the input data deviation sets are deemed to satisfy the input data attribute deviation threshold sets, and the first initial input record data and the second initial input record data are clustered into the same cluster.

[0064] If any data deviation in the input data deviation set is greater than or equal to the corresponding input data attribute deviation threshold, it is considered unsatisfied and the first initial input record data and the second initial input record data are clustered into different clusters. The analysis is repeated repeatedly to obtain the clustering result.

[0065] Furthermore, the plurality of initial output record data and the plurality of output record data are encrypted and transmitted back to obtain the first training data set. Step S27 includes the following steps:

[0066] S271: Optimize the encryption code through the dynamic password library and generate a recommended encryption code;

[0067] S272: Encrypt the multiple initial output record data and the multiple output record data according to the recommended encryption code and transmit them back to obtain the first training data set.

[0068] Specifically, the dynamic password library refers to a group of preset passwords used to encrypt the multiple initial output record data and the multiple output record data. By optimizing the encryption codes, a recently used encryption code with low frequency is selected as the recommended encryption code. The multiple initial output record data and the multiple output record data are encrypted and transmitted back to generate a first training data set. Using encryption codes to optimize dynamically changing encryption codes helps ensure data security.

[0069] Furthermore, the encryption code is optimized by the dynamic password library to generate a recommended encryption code. Step S271 includes the following steps:

[0070] S2711: Randomly extract a first encryption code from the dynamic password library, wherein any encryption code in the dynamic password library has a preset number of bits;

[0071] S2712: Obtain the encryption code set selected for the default time zone;

[0072] S2713: Based on the first encryption code, traverse the preset time zone and select an encryption code set to perform out-of-position number statistics to generate an out-of-position number set;

[0073] S2714: Calculate the minimum value of the set of out-of-position quantities and set it as the first selection coefficient;

[0074] S2715: When the first selection coefficient is greater than or equal to a selection coefficient threshold, setting the first encryption code as the recommended encryption code;

[0075] S2716: Otherwise, update the first encryption code.

[0076] Specifically, the encryption code optimization can adopt existing optimization algorithms, such as genetic algorithm, annealing algorithm, etc. The embodiment of the present application lists an example without limitation:

[0077] Randomly extract the first encryption code from the dynamic password library; retrieve the encryption code set selected for the preset time zone, that is, the encryption code set that has been selected for the preset time zone in the past, and the preset time zone is preferably in the past three months. Based on the first encryption code, traverse the encryption code set selected for the preset time zone to perform out-of-position quantity statistics, and generate an out-of-position quantity set. Since the number of encryption code bits is the same, the alignment and comparison are performed to obtain the number of positions with different values, which are stored as out-of-position quantities. Calculate the minimum value of the out-of-position quantity set and set it as the first selection coefficient; when the first selection coefficient is greater than or equal to the selection coefficient threshold pre-identified by the server, set the first encryption code as the recommended encryption code. Otherwise, update the first encryption code. Using encryption codes to optimize dynamically changing encryption codes is conducive to ensuring data security.

[0078] Further, a pre-trained AI model is constructed based on the second training data set to the N+1th training data set, and the pre-trained AI model is trained based on the first training data set to generate a target AI model. Step S60 includes the following steps:

[0079] S61: Determine whether a second data volume of the second training data set up to the N+1th training data set is greater than or equal to the convergence data volume threshold;

[0080] S62: If greater than or equal to, construct a pre-trained AI model based on the second training data set to the N+1th training data set;

[0081] S63: Train the pre-trained AI model according to the first training data set to generate the target AI model.

[0082] Furthermore, the method further includes step S64: if it is less than, re-collecting data according to the data collection task instruction.

[0083] Furthermore, step S80 is also included: encrypting and transmitting the pre-trained model topology structure feature information and the pre-trained model node parameter feature information of the pre-trained AI model to the first response node until the Nth response node.

[0084] Specifically, it is necessary to determine whether the second data volume of the second training dataset up to the N+1th training dataset is greater than or equal to the convergence data volume threshold, where the second data volume is the sum of the data volumes of the second training dataset up to the N+1th training dataset. If so, a pre-trained AI model is constructed based on the second training dataset up to the N+1th training dataset; if not, data is re-collected according to the data collection task instructions. The pre-trained AI model is trained based on the first training dataset to generate the target AI model.

[0085] Preferably, the pre-trained model topology structure feature information and the pre-trained model node parameter feature information of the pre-trained AI model are encrypted and transmitted to the first response node until the Nth response node. The first response node until the Nth response node shares data and can also share models, reflecting humanization.

[0086] The embodiment of the present invention provides an AI model generation method based on artificial intelligence, which has at least the following technical effects:

[0087] By receiving the model building requirements from the user side, the first training data set is encrypted and collected according to the data crawling link provided by the user; when the data amount of the first training data set is small, the data collection task quality is generated, and the second training data set is encrypted and collected through the blockchain until the N+1th training data set; then pre-training is performed through the second training data set to the N+1th training data set to obtain a pre-trained AI model with strong versatility, and the first training data set is further used to train the target AI model to obtain an intelligent model with a high degree of integration with the first user end, thereby achieving the technical effect of increasing the amount of training samples and a high degree of integration of AI intelligent models with scenarios.

[0088] Example 2:

[0089] See also Figure 2 , Figure 2 Schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 2 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, the following steps are implemented:

[0090] Interacting with the first user terminal to obtain AI model building requirements, wherein the AI ​​model building requirements include model input data attributes, model output data attributes, and a data acquisition source link;

[0091] Accessing the data collection source link to perform data encryption collection according to the model input data attributes and the model output data attributes to obtain a first training data set;

[0092] When a first data volume of the first training data set is less than a convergence data volume threshold, generating a data acquisition task instruction according to the model input data attribute and the model output data attribute;

[0093] Publishing the data collection task instruction to the data collection blockchain, obtaining the first response node up to the Nth response node;

[0094] Traversing the first response node until the Nth response node performs data encryption collection to obtain a second training data set until the N+1th training data set;

[0095] Constructing a pre-trained AI model based on the second training data set up to the N+1th training data set, and training the pre-trained AI model based on the first training data set to generate a target AI model, wherein the target AI model has topological structure feature information and node parameter feature information;

[0096] The topology structure characteristic information and the node parameter characteristic information are encrypted and transmitted to the first user terminal.

[0097] Example 3:

[0098] See also Figure 3 , Figure 3 Schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. Figure 3 As shown, this embodiment provides a computer-readable storage medium 600 on which a computer program 611 is stored. When the computer program 611 is executed by a processor, the following steps are implemented:

[0099] Interacting with the first user terminal to obtain AI model building requirements, wherein the AI ​​model building requirements include model input data attributes, model output data attributes, and a data acquisition source link;

[0100] Accessing the data collection source link to perform data encryption collection according to the model input data attributes and the model output data attributes to obtain a first training data set;

[0101] When a first data volume of the first training data set is less than a convergence data volume threshold, generating a data acquisition task instruction according to the model input data attribute and the model output data attribute;

[0102] Publishing the data collection task instruction to the data collection blockchain, obtaining the first response node up to the Nth response node;

[0103] Traversing the first response node until the Nth response node performs data encryption collection to obtain a second training data set until the N+1th training data set;

[0104] Constructing a pre-trained AI model based on the second training data set up to the N+1th training data set, and training the pre-trained AI model based on the first training data set to generate a target AI model, wherein the target AI model has topological structure feature information and node parameter feature information;

[0105] The topology structure characteristic information and the node parameter characteristic information are encrypted and transmitted to the first user terminal.

[0106] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0107] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0109] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0111] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An AI model generation method based on artificial intelligence, characterized in that: Applicable to servers, including: Interacting with the first user terminal to obtain AI model building requirements, wherein the AI ​​model building requirements include model input data attributes, model output data attributes, and a data acquisition source link; Accessing the data collection source link to perform data encryption collection according to the model input data attributes and the model output data attributes to obtain a first training data set; When a first data volume of the first training data set is less than a convergence data volume threshold, generating a data acquisition task instruction according to the model input data attribute and the model output data attribute; Publishing the data collection task instruction to the data collection blockchain, obtaining the first response node up to the Nth response node; Traversing the first response node until the Nth response node performs data encryption collection to obtain a second training data set until the N+1th training data set; Constructing a pre-trained AI model based on the second training data set to the N+1th training data set, and training the pre-trained AI model based on the first training data set to generate a target AI model, wherein the target AI model has topological structure feature information and node parameter feature information; encrypting and transmitting the topology structure characteristic information and the node parameter characteristic information to the first user terminal; Accessing the data collection source link to perform encrypted data collection according to the model input data attributes and the model output data attributes to obtain a first training data set includes: Accessing the data collection source link according to the model input data attributes and the model output data attributes to collect an initial input record data set and an initial output record data set; Interacting with the first user terminal to obtain a set of input data attribute deviation thresholds; Performing cluster analysis on the initial input record data set according to the input data attribute deviation threshold set to obtain multiple clusters of initial input record data; Clustering the initial output record data set according to the multiple clusters of initial input record data to generate multiple clusters of initial output record data; Traversing the multiple clusters of initial output record data to perform mode calculation and generate multiple output record data; randomly extracting one piece of initial output record data from each of the plurality of clusters of initial output record data to obtain a plurality of initial output record data; The plurality of initial output record data and the plurality of output record data are encrypted and transmitted back to obtain the first training data set.

2. The method according to claim 1, wherein Performing cluster analysis on the initial input record data set according to the input data attribute deviation threshold set to obtain multiple clusters of initial input record data, including: Obtaining first initial input record data and second initial input record data of the initial input record data set; Comparing the same attribute deviations of the first initial input record data and the second initial input record data to obtain an input data deviation set, wherein the input data deviation set corresponds one-to-one to the input data attribute deviation threshold set; When the input data deviation set satisfies the input data attribute deviation threshold set, clustering the first initial input record data and the second initial input record data into the same cluster; When any one of the input data deviation sets does not satisfy the input data attribute deviation threshold set, the first initial input record data and the second initial input record data are clustered into a heterogeneous cluster.

3. The method according to claim 2, wherein Encrypting and transmitting the plurality of initial output record data and the plurality of output record data to obtain the first training data set includes: Optimize encryption codes through the dynamic password library and generate recommended encryption codes; The plurality of initial output record data and the plurality of output record data are encrypted and transmitted back according to the recommended encryption code to obtain the first training data set.

4. The method according to claim 3, wherein Optimize encryption codes through the dynamic password library and generate recommended encryption codes, including: Randomly extracting a first encryption code from the dynamic password library, wherein any encryption code in the dynamic password library has a preset number of bits; Get the default time zone encryption code set; Based on the first encryption code, traverse the preset time zone and select an encryption code set to perform out-of-position quantity statistics to generate an out-of-position quantity set; Calculate the minimum value of the set of out-of-position quantities and set it as the first selection coefficient; When the first selection coefficient is greater than or equal to a selection coefficient threshold, setting the first encryption code as the recommended encryption code; Otherwise, the first encryption code is updated.

5. The method according to claim 1, wherein Constructing a pre-trained AI model based on the second training data set to the N+1th training data set, and training the pre-trained AI model based on the first training data set to generate a target AI model, including: Determining whether a second data volume of the second training data set up to the N+1th training data set is greater than or equal to the convergence data volume threshold; If it is greater than or equal to, construct a pre-trained AI model based on the second training data set to the N+1th training data set; The pre-trained AI model is trained according to the first training data set to generate the target AI model.

6. The method according to claim 5, wherein Also includes: If it is less than, data is collected again according to the data collection task instruction.

7. The method according to claim 5, wherein Also includes: The pre-trained model topology structure feature information and the pre-trained model node parameter feature information of the pre-trained AI model are encrypted and transmitted to the first response node until the Nth response node.

8. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor, configured to read and execute the computer software program, thereby implementing the artificial intelligence-based AI model generation method described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium, characterized in that The storage medium stores a computer software program, which, when executed by a processor, implements an AI model generation method based on artificial intelligence as described in any one of claims 1 to 7.

Images