CN118157999B - Model training method, device, terminal and storage medium - Google Patents

Abstract

The application provides a model training method, a device, a terminal and a storage medium, which belong to the technical field of artificial intelligence, wherein the model training method comprises the following steps: acquiring first model parameters, wherein the first model parameters are model parameters corresponding to the first network intrusion model after training based on a target sample; sending the first model parameters and the detection precision values of the first network intrusion model to a server; receiving a second model parameter sent by a server, wherein the second model parameter is a model parameter obtained by calculating a global network intrusion model of the server based on the first model parameter and the detection precision value; and updating the model parameters of the first network intrusion model based on the second model parameters to obtain a second network intrusion model. Thus, the problem of privacy disclosure caused by uploading network data of the terminal side to the server is avoided, and the privacy of private data of a user is improved.

Description

Model training method, device, terminal and storage medium

Technical Field

The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to a model training method, device, terminal and storage medium.

Background Art

The training scheme of network intrusion models in related technologies usually collects traffic data from a large number of clients and then uploads the collected traffic data to a server so that the server can train the network intrusion model based on the collected traffic data as training samples. However, in the process of uploading the client's traffic data to the server, the user's private data is also easily uploaded to the server, resulting in the risk of privacy leakage in the process of uploading traffic data.

Summary of the invention

The embodiments of the present application provide a model training method, device, terminal and storage medium to solve the problem of privacy leakage in the training scheme of the network intrusion model in the related art.

To solve the above problems, this application is implemented as follows:

In a first aspect, an embodiment of the present application provides a model training method for a terminal, comprising:

Obtaining a first model parameter, where the first model parameter is a model parameter corresponding to a first network intrusion model after training based on a target sample, where the target sample includes image network data or text network data, where the first network intrusion model is a model obtained by updating a preset initial model in the i-th iteration, and where the preset initial model is a model for performing network intrusion detection on network data, and i is a positive integer;

Sending the first model parameter and the detection accuracy value of the first network intrusion model to a server;

receiving a second model parameter sent by the server, where the second model parameter is a model parameter calculated by a global network intrusion model of the server based on the first model parameter and the detection accuracy value;

Based on the second model parameters, the model parameters of the first network intrusion model are updated to obtain a second network intrusion model;

The second network intrusion model is a model obtained by updating the preset initial model in the i+1th iteration.

In a second aspect, an embodiment of the present application provides a model training device for a terminal, comprising:

An acquisition module is used to acquire a first model parameter, wherein the first model parameter is a model parameter corresponding to a first network intrusion model after training based on a target sample, wherein the target sample includes image network data or text network data, and the first network intrusion model is a model obtained by updating a preset initial model in the i-th iteration, and the preset initial model is a model for performing network intrusion detection on network data, and i is a positive integer;

A sending module, used for sending the first model parameter and the detection accuracy value of the first network intrusion model to a server;

A receiving module, configured to receive a second model parameter sent by the server, where the second model parameter is a model parameter calculated by a global network intrusion model of the server based on the first model parameter and the detection accuracy value;

An updating module, configured to update the model parameters of the first network intrusion model based on the second model parameters to obtain a second network intrusion model;

The second network intrusion model is a model obtained by updating the preset initial model in the i+1th iteration.

In a third aspect, an embodiment of the present application further provides a terminal, comprising: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps of the method described in the first aspect above.

In a fourth aspect, an embodiment of the present application further provides a readable storage medium for storing a program, which, when executed by a processor, implements the steps in the method described in the first aspect above.

According to a fifth aspect, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the method described in the first aspect.

In an embodiment of the present application, by obtaining a first model parameter, the first model parameter is a model parameter corresponding to the first network intrusion model after training based on a target sample, the target sample includes image network data or text network data, the first network intrusion model is a model obtained by updating the preset initial model at the i-th iteration, and the preset initial model is a model for performing network intrusion detection on network data, i is a positive integer; sending the first model parameter and the detection accuracy value of the first network intrusion model to the server; receiving a second model parameter sent by the server, the second model parameter is a model parameter calculated by the global network intrusion model of the server based on the first model parameter and the detection accuracy value; updating the model parameter of the first network intrusion model based on the second model parameter to obtain a second network intrusion model; wherein the second network intrusion model is a model updated at the i+1th iteration of the preset initial model. In this way, during the model training process, the terminal and the server only interact with the model parameters, and do not involve the interaction of specific traffic data. Compared with the conventional solution that requires collecting traffic data from each terminal and training the global network intrusion model based on the collected traffic data, it can not only avoid the problem of user privacy data leakage, but also reduce the computing cost of the server during the model training process.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only 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 labor.

FIG1 is a schematic diagram of a flow chart of a model training method provided in an embodiment of the present application;

FIG2 is a schematic diagram of the structure of a federated quantum learning framework provided in an embodiment of the present application;

FIG3 is a schematic diagram of the structure of a model training device provided in an embodiment of the present application;

FIG4 is a schematic diagram of the structure of a terminal provided in an embodiment of the present application.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

The terms "first", "second" etc. in the embodiments of the present application are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. In addition, the terms "include" and "have" and any variation thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "and/or" is used in the present application to represent at least one of connected objects, such as A and/or B and/or C, representing the inclusion of separate A, separate B, separate C, and A and B all exist, B and C all exist, A and C all exist, and 7 situations that A, B and C all exist.

In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

The model training method provided in the embodiments of the present application is described below.

Referring to Figure 1, Figure 1 is a flow chart of a model training method provided in an embodiment of the present application. The model training method shown in Figure 1 can be executed by a terminal.

As shown in FIG1 , the model training method may include the following steps:

Step 101: Obtain first model parameters, where the first model parameters are model parameters corresponding to a first network intrusion model after training based on a target sample.

Among them, the above-mentioned target sample includes image network data or text network data, the above-mentioned first network intrusion model is a model obtained by updating the preset initial model in the i-th iteration, and the preset initial model is a model for network intrusion detection on network data, and i is a positive integer.

Since the preset initial model is a model used for performing network intrusion detection on network data, the first network intrusion model iteratively generated based on the preset initial model can also realize network intrusion detection on network data.

It is understandable that the target sample used to train the first network intrusion model may be network intrusion data. By using the network intrusion data to train the first network intrusion model, the convergence speed of the network intrusion model and the accuracy of network intrusion detection may be improved.

The above-mentioned target samples can be the local network data of the terminal, and the network intrusion model can be trained directly on the terminal side based on the local network data, avoiding the privacy leakage problem caused by uploading the local network data to the server, and improving the privacy of the user's private data.

Step 102: Send the first model parameters and the detection accuracy value of the first network intrusion model to a server.

In this embodiment, the test data may be input into the first network intrusion model to obtain a network intrusion detection result, and the detection accuracy value of the first network intrusion model may be determined based on the obtained network intrusion detection result.

For example, when the test data A is input into the first network intrusion model and the obtained network intrusion detection result is that the probability that the test data A is network intrusion data is 0.7, it can be determined that the detection accuracy value of the first network intrusion model is 0.7.

In practical applications, multiple groups of test data can be input into the first network intrusion model to obtain multiple network intrusion detection results, and the average detection accuracy corresponding to the multiple network intrusion detection results can be calculated, and finally the average value is determined as the detection accuracy value of the first network intrusion model.

For example, for test data A, test data B, and test data C, their detection accuracy values in the first network intrusion model are 0.65, 0.7, and 0.75, respectively. The detection accuracy value of the first network intrusion model can be determined as (0.65+0.7+0.75)/3, that is, the detection accuracy value of the first network intrusion model can be obtained as 0.70.

Step 103: Receive second model parameters sent by the server, where the second model parameters are model parameters calculated by the global network intrusion model of the server based on the first model parameters and the detection accuracy value.

The above-mentioned global network intrusion model can be understood as a network intrusion model on the server side, which can receive first model parameters and detection accuracy values sent from different terminals, and comprehensively update the model parameters of the global network intrusion model based on the received first model parameters and detection accuracy values, and determine the updated model parameters as the second model parameters.

In this embodiment, since the server can receive the first model parameters and detection accuracy values sent from different terminals, compared to collecting traffic data from each terminal and training the global network intrusion model based on the collected traffic data, it can not only avoid the problem of user privacy data leakage, but also reduce the computing cost of the server during the model training process.

Moreover, since the second model parameters are determined by the server based on the first model parameters and detection accuracy values sent by different terminals, the network intrusion model obtained after the first network intrusion model on the terminal side is updated based on the second model parameters can have the network intrusion detection performance of the global network intrusion model, which greatly improves the network intrusion detection performance of the network intrusion model on the terminal side.

Furthermore, since the information interaction between the terminal and the server is mainly the interaction of model parameters, compared with the conventional solution that requires uploading a large amount of traffic data, the communication pressure between the terminal and the server can be effectively reduced.

Step 104: Update the model parameters of the first network intrusion model based on the second model parameters to obtain a second network intrusion model.

The second network intrusion model can be understood as a model obtained by updating the preset initial model in the i+1th iteration.

In one embodiment, obtaining the first model parameter includes:

Obtain target samples;

Obtaining a fairness coefficient of the first network intrusion model;

Determining model parameters of the first network intrusion model based on the fairness coefficient;

The first network intrusion model is trained using the target sample to update the model parameters of the first network intrusion model, and the updated model parameters of the first network intrusion model are determined as the first model parameters.

In this embodiment, the fairness coefficient of the first network intrusion model can be obtained, and its model parameters are updated based on its fairness parameter, thereby reducing the impact of abnormal samples on model training.

The above-mentioned target samples can be understood as network intrusion data detected on the terminal side, including but not limited to network intrusion content such as pictures and texts.

In one embodiment, obtaining the fairness coefficient of the first network intrusion model includes:

Obtaining a distance coefficient between the first network intrusion model and a third network intrusion model, wherein the third network intrusion model is a model obtained by updating the preset initial model at the i-1th iteration, and the distance coefficient is used to characterize the degree of change of the model;

The fairness coefficient of the first network intrusion model is calculated based on the distance coefficient and a preset constant term.

The first network intrusion model can be understood as a network intrusion model at the current moment, and the third network intrusion model can be understood as a network intrusion model at the previous moment.

In this embodiment, the distance coefficients between the first network intrusion model and the third network intrusion model can be obtained to determine the degree of change of the model, and then the degree of change of the model can be memorized to calculate the fairness coefficient of the first network intrusion model, thereby reducing the impact of abnormal samples on model training.

For example, based on the formula:

Calculate the distance coefficient d _k , where d _k represents the distance coefficient between the first network intrusion model and the third network intrusion model, and the smaller d _{k is} , the greater the degree of change of the model is, w _r-1 represents the model parameters of the third network intrusion model, w _r represents the model parameters of the first network intrusion model, k can be understood as the network model on the terminal side, r can be understood as the number of communications between the terminal and the server, and r=1,2,3….

After determining the distance coefficient d _k , it can be based on the formula:

Calculate the fairness coefficient q _k of the first network intrusion model, where: Represents a constant term, which can be a preset value.

After determining the fairness coefficient q _k , it can be based on the formula:

The fairness coefficient area average is calculated to obtain q _ave and the loss function Loss(W _r ) of the first network intrusion model.

After determining the loss function Loss(W _r ) of the first network intrusion model, the first model parameter w _r+1 can be calculated based on the model parameter w _r of the first network intrusion model and the loss function Loss(W _r ). Specifically, it can be based on the formula:

Where n represents the learning rate.

Through the above process, the fairness coefficient can be used to update the model parameters of the first network intrusion model, thereby obtaining the above first model parameters.

In one embodiment, obtaining a target sample includes:

Acquire target network data of the terminal, where the target network data is local network intrusion data of the terminal;

Generate false data corresponding to the target network data based on a quantum generative adversarial network;

The target network data and the false data are merged in random order to obtain a target sample.

In this embodiment, false data corresponding to the target network data can be generated based on the quantum generative adversarial network, and then the target network data and the false data are randomly merged to obtain a target sample, thereby expanding the number of target samples.

In one embodiment, the acquired target network data may be preprocessed first, and the preprocessing includes but is not limited to data cleaning, digitization, normalization, data random splitting and other processing methods.

Moreover, when the target network data obtained is small, that is, the number of samples is insufficient, quantum generative adversarial networks (QGAN) can be used to expand the sample data, which solves the problems of small data samples and unbalanced distribution, improves the efficiency of updating the network intrusion model on the terminal side and the robustness of the intrusion detection model, and improves the imbalance of sample data.

Among them, applying QGAN to expand sample data includes the following steps:

Step a: Initialize QGAN, and the quantum generative adversarial network includes a generator G, a discriminator D and a conditional label .

Step b: Train QGAN, for the number of iterations ,include:

For attack type ,have:

Eigenvalue normalization: Each network data has K features, and each eigenvalue is , in order to better input data into the quantum generative adversarial network, there is

In the formula, the sum of all β values is 1, and q satisfies The minimum value in the case of .

Among them, the network data as the input of the quantum generative adversarial network can be expressed as:

The generator can generate fake data GD _{(i,a) based on the formula GD (i,a) =} G _i (n,l _a ). After generating the fake data GD _(i,a) , the fake data GD _(i,a) can be merged with the preprocessed data D', and the flag of the fake data GD _(i,a) can be set to 0, the flag of the real data D' can be set to 1, and the fake data GD _(i,a) and the real data D' can be merged in random order into data _(i,a) to obtain the expanded sample data.

After obtaining the expanded sample data data _{(i, a)} , the data can be identified by the discriminator D, and if the given data is real, the return value is 1, otherwise the return value is 0.

Specifically, it can be based on the formula:

Calculate the objective function value V( _D ,G) in the i-th iteration of QGAN; _{where EG} is the objective function of the generator, ED is the objective function of the discriminator, D ( G ( Z )) represents the predicted output of the discriminator for the generated sample, and D ( x ) represents the predicted output of the discriminator for the real sample; then determine whether V(D,G) tends to be stable. If so, generate fake data; if not, set i=i+1, and based on the formula:

Calculate the objective function value and update the QGAN model parameters w _(QGAN,i+1) under the learning rate n.

Step c: The generator synthesizes fake data. In the process of generating fake data, β _i can be regarded as the probability of occurrence of the corresponding eigenvalue. The quantum generator can generate new data according to this probability distribution. After iterative training of the quantum generative adversarial network, the synthesized data Constantly looking to real data Close to where the data synthesized by the generator as follows:

In one embodiment, sending the first model parameter and the detection accuracy value of the first network intrusion model to the server includes:

Obtaining a detection accuracy value of the first network intrusion model;

When the detection accuracy value is greater than or equal to a preset accuracy value, sending the first model parameter and the detection accuracy value to a server;

The preset precision value is determined based on the highest precision value and the lowest precision value.

In this embodiment, by setting the sending condition, that is, determining whether the detection accuracy value is greater than or equal to the preset accuracy value, to determine whether to send the first model parameter and the detection accuracy value to the server, the influence of invalid data on the global network intrusion model on the server side is avoided, and the processing of invalid data by the global network intrusion model on the server side is also avoided, thereby reducing the computing cost of the server.

In one embodiment, the model parameters may be updated based on a Federated Quantum Learning (FQL) framework.

The federated quantum learning framework shown in Figure 2 includes a global network intrusion model on the server side and N local models on the terminal side. The local model on the terminal side can be understood as the first network intrusion model mentioned above. The global network intrusion model can send model parameters GW ₀ , learning rate ρ, batch size B, loss function f, and total number of communication rounds R to each local model participating in the training, and divide the data into N independent and identically distributed parts, and distribute them to each local model participating in the training. It is the local model communication status, and the initial value is 1.

For the number of communications ,have:

Calculate the fairness coefficient and update the local model parameters.

The global network intrusion model can upload some local model parameters to the global model according to the filtering weighting strategy. Specifically, the detection accuracy Acc of the local network intrusion model on the terminal side can be calculated. If the detection accuracy Acc<κ, its communication state is set to error and the value is assigned , do not upload local model parameters to the global network intrusion model, and only upload local model parameters to the global network intrusion model when the detection accuracy Acc ≥ κ to reduce the impact of invalid data on the global network intrusion model on the server side.

Moreover, through such a setting, the impact of poorly performing local models on the global network intrusion model can be reduced, the communication overhead can be reduced, and the performance of the global intrusion detection model can be improved.

Among them, under normal circumstances, the threshold κ _i defaults to κ _0. When the local model has serious abnormalities, its accuracy will be greatly different from the normal client accuracy. If the difference between the highest accuracy and the lowest accuracy in the local model is higher than 0.4, the expression of κ is as follows:

It is understandable that the difference between the highest accuracy and the lowest accuracy in the local model may also be set higher than 0.45 or 0.35, and the specific value may be set based on actual needs.

The server can receive a parameter set sent by a terminal in a correct state, where the parameter set includes information such as model parameters and detection accuracy values of a local model on the terminal side.

Among them, it can be based on the formula:

Calculate contribution rate , where η is the number of clients with correct receiving status, and 0<η≤N, is the number of samples of the local terminal c in the rth round of communication; then, it can be based on the formula:

Calculate the contribution rate of the local terminal in this round of communication , is the detection accuracy value of the local terminal in this round of communication; finally, based on the formula:

The model parameters of the global network intrusion model are calculated, such as the second model parameters, and then the calculated model parameters are sent to each local terminal, so that each local terminal can update the local model based on the received model parameters.

It can be understood that r<R can be judged. If so, r=r+1 is set, and the model parameters of the global network intrusion model are calculated, and its communication state is set to correct and assigned , to enter the next round of model training until r = R. And when r = R, the training ends.

In this embodiment, the user-side terminal can regularly interact with the server and summarize model parameters or gradient information to collaboratively train the global network intrusion model. During the training, the user's sensitive data will not be involved, effectively avoiding the problem of user privacy data leakage.

The model training method provided in the embodiment of the present application obtains a first model parameter, wherein the first model parameter is a model parameter corresponding to the first network intrusion model after training based on a target sample, wherein the target sample includes image network data or text network data, wherein the first network intrusion model is a model obtained by updating the preset initial model at the i-th iteration, and the preset initial model is a model for performing network intrusion detection on network data, wherein i is a positive integer; sends the first model parameter and the detection accuracy value of the first network intrusion model to a server; receives a second model parameter sent by the server, wherein the second model parameter is a model parameter calculated by the global network intrusion model of the server based on the first model parameter and the detection accuracy value; updates the model parameter of the first network intrusion model based on the second model parameter to obtain a second network intrusion model; wherein the second network intrusion model is a model obtained by updating the preset initial model at the i+1th iteration. This avoids the problem of privacy leakage caused by uploading the network data on the terminal side to the server, and improves the privacy of the user's private data.

The various optional implementation modes introduced in the embodiments of the present application may be implemented in combination with each other or may be implemented separately if they do not conflict with each other, and the embodiments of the present application are not limited to this.

See Figure 3, which is a structural diagram of a model training device provided in an embodiment of the present application. As shown in Figure 3, the model training device 300 includes:

The acquisition module 301 is used to acquire a first model parameter, wherein the first model parameter is a model parameter corresponding to a first network intrusion model after training based on a target sample, wherein the target sample includes image network data or text network data, and the first network intrusion model is a model obtained by updating a preset initial model in the i-th iteration, and the preset initial model is a model for performing network intrusion detection on network data, and i is a positive integer;

A sending module 302, configured to send the first model parameter and the detection accuracy value of the first network intrusion model to a server;

A receiving module 303 is used to receive a second model parameter sent by the server, where the second model parameter is a model parameter calculated by the global network intrusion model of the server based on the first model parameter and the detection accuracy value;

An updating module 304, configured to update the model parameters of the first network intrusion model based on the second model parameters to obtain a second network intrusion model;

The second network intrusion model is a model obtained by updating the preset initial model in the i+1th iteration.

Optionally, the acquisition module 301 is specifically configured to:

Obtain target samples;

Obtaining a fairness coefficient of the first network intrusion model;

Determining model parameters of the first network intrusion model based on the fairness coefficient;

The first network intrusion model is trained using the target sample to update the model parameters of the first network intrusion model, and the updated model parameters of the first network intrusion model are determined as the first model parameters.

Optionally, the acquisition module 301 is specifically configured to:

Obtaining a distance coefficient between the first network intrusion model and a third network intrusion model, wherein the third network intrusion model is a model obtained by updating the preset initial model at the i-1th iteration, and the distance coefficient is used to characterize the degree of change of the model;

The fairness coefficient of the first network intrusion model is calculated based on the distance coefficient and a preset constant term.

Optionally, the acquisition module 301 is specifically configured to:

Acquire target network data of the terminal, where the target network data is local network intrusion data of the terminal;

Generate false data corresponding to the target network data based on a quantum generative adversarial network;

The target network data and the false data are merged in random order to obtain a target sample.

Optionally, the sending module 302 is specifically configured to:

Obtaining a detection accuracy value of the first network intrusion model;

When the detection accuracy value is greater than or equal to a preset accuracy value, sending the first model parameter and the detection accuracy value to a server;

The preset precision value is determined based on the highest precision value and the lowest precision value.

The model training device 300 can implement each process of the method embodiment of Figure 1 in the embodiment of the present application, and achieve the same beneficial effects. To avoid repetition, it will not be repeated here.

See Figure 4, which is a schematic diagram of the structure of a terminal provided in an embodiment of the present application. As shown in Figure 4, an embodiment of the present application further provides a terminal, including a bus 401, a transceiver 402, an antenna 403, a bus interface 404, a processor 405 and a memory 406.

The processor 405 is configured to:

Obtaining a first model parameter, where the first model parameter is a model parameter corresponding to a first network intrusion model after training based on a target sample, where the target sample includes image network data or text network data, where the first network intrusion model is a model obtained by updating a preset initial model in the i-th iteration, and where the preset initial model is a model for performing network intrusion detection on network data, and i is a positive integer;

Sending the first model parameter and the detection accuracy value of the first network intrusion model to a server;

receiving a second model parameter sent by the server, where the second model parameter is a model parameter calculated by a global network intrusion model of the server based on the first model parameter and the detection accuracy value;

Based on the second model parameters, the model parameters of the first network intrusion model are updated to obtain a second network intrusion model;

The second network intrusion model is a model obtained by updating the preset initial model in the i+1th iteration.

Optionally, the processor 405 is configured to:

Obtain target samples;

Obtaining a fairness coefficient of the first network intrusion model;

Determining model parameters of the first network intrusion model based on the fairness coefficient;

The first network intrusion model is trained using the target sample to update the model parameters of the first network intrusion model, and the updated model parameters of the first network intrusion model are determined as first model parameters.

Optionally, the processor 405 is configured to:

Obtaining a distance coefficient between the first network intrusion model and a third network intrusion model, wherein the third network intrusion model is a model obtained by updating the preset initial model in the i-1th iteration, and the distance coefficient is used to characterize a degree of change of the model;

The fairness coefficient of the first network intrusion model is calculated based on the distance coefficient and a preset constant term.

Optionally, the processor 405 is configured to:

Acquire target network data of the terminal, where the target network data is local network intrusion data of the terminal;

Generate false data corresponding to the target network data based on a quantum generative adversarial network;

The target network data and the false data are merged in random order to obtain a target sample.

Optionally, the processor 405 is configured to:

Obtaining a detection accuracy value of the first network intrusion model;

When the detection accuracy value is greater than or equal to a preset accuracy value, sending the first model parameter and the detection accuracy value to a server;

The preset precision value is determined based on the highest precision value and the lowest precision value.

In FIG. 4 , a bus architecture (represented by bus 401) is shown. Bus 401 may include any number of interconnected buses and bridges. Bus 401 links various circuits including one or more processors represented by processor 405 and memory represented by memory 406. Bus 401 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 404 provides an interface between bus 401 and transceiver 402. Transceiver 402 may be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. Data processed by processor 405 is transmitted on a wireless medium via antenna 403. Further, antenna 403 also receives data and transmits the data to processor 405.

Processor 405 is responsible for managing bus 401 and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. Memory 406 can be used to store data used by processor 405 when performing operations.

Optionally, the processor 405 may be a CPU, an ASIC, an FPGA or a CPLD.

The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, each process of the above-mentioned model training method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the computer-readable storage medium is such as a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

An embodiment of the present application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the various processes of the above-mentioned model training method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.

It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

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, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims (9)

1. A model training method, characterized in that it is used in a terminal, comprising: Obtaining a first model parameter, where the first model parameter is a model parameter corresponding to a first network intrusion model after training based on a target sample, where the target sample includes image network data or text network data, where the first network intrusion model is a model obtained by updating a preset initial model in the i-th iteration, and where the preset initial model is a model for performing network intrusion detection on network data, and i is a positive integer; Sending the first model parameter and the detection accuracy value of the first network intrusion model to a server; receiving a second model parameter sent by the server, where the second model parameter is a model parameter calculated by a global network intrusion model of the server based on the first model parameter and the detection accuracy value; Based on the second model parameters, the model parameters of the first network intrusion model are updated to obtain a second network intrusion model; The second network intrusion model is a model obtained by updating the preset initial model at the i+1th iteration; The obtaining of the first model parameter comprises: Obtain target samples; Obtaining a fairness coefficient of the first network intrusion model; Determine a loss function of the first network intrusion model based on the fairness coefficient; Using the target sample to train the first network intrusion model, updating the model parameters of the first network intrusion model based on the loss function, and determining the updated model parameters of the first network intrusion model as first model parameters; The obtaining of the fairness coefficient of the first network intrusion model includes: Obtaining a distance coefficient between the first network intrusion model and a third network intrusion model, wherein the third network intrusion model is a model obtained by updating the preset initial model in the i-1th iteration, and the distance coefficient is used to characterize a degree of change of the model; The fairness coefficient of the first network intrusion model is calculated based on the distance coefficient and a preset constant term. 2. The method according to claim 1, characterized in that the obtaining of the target sample comprises: Acquire target network data of the terminal, where the target network data is local network intrusion data of the terminal; Generate false data corresponding to the target network data based on a quantum generative adversarial network; The target network data and the false data are merged in random order to obtain a target sample. 3. The method according to claim 1 or 2, characterized in that the sending the first model parameter and the detection accuracy value of the first network intrusion model to the server comprises: Obtaining a detection accuracy value of the first network intrusion model; When the detection accuracy value is greater than or equal to a preset accuracy value, sending the first model parameter and the detection accuracy value to a server; The preset precision value is determined based on the highest precision value and the lowest precision value. 4. A model training device, characterized in that it is used in a terminal, comprising: An acquisition module is used to acquire a first model parameter, wherein the first model parameter is a model parameter corresponding to a first network intrusion model after training based on a target sample, wherein the target sample includes image network data or text network data, and the first network intrusion model is a model obtained by updating a preset initial model in the i-th iteration, and the preset initial model is a model for performing network intrusion detection on network data, and i is a positive integer; A sending module, used for sending the first model parameter and the detection accuracy value of the first network intrusion model to a server; A receiving module, configured to receive a second model parameter sent by the server, where the second model parameter is a model parameter calculated by a global network intrusion model of the server based on the first model parameter and the detection accuracy value; An updating module, configured to update the model parameters of the first network intrusion model based on the second model parameters to obtain a second network intrusion model; The second network intrusion model is a model obtained by updating the preset initial model at the i+1th iteration; The acquisition module is specifically used for: Obtain target samples; Obtaining a fairness coefficient of the first network intrusion model; Determine a loss function of the first network intrusion model based on the fairness coefficient; Using the target sample to train the first network intrusion model to update the model parameters of the first network intrusion model, and determining the updated model parameters of the first network intrusion model as the first model parameters; The acquisition module is specifically used for: Obtaining a distance coefficient between the first network intrusion model and a third network intrusion model, wherein the third network intrusion model is a model obtained by updating the preset initial model in the i-1th iteration, and the distance coefficient is used to characterize a degree of change of the model; The fairness coefficient of the first network intrusion model is calculated based on the distance coefficient and a preset constant term. 5. The device according to claim 4, characterized in that the acquisition module is specifically used to: Acquire target network data of the terminal, where the target network data is local network intrusion data of the terminal; Generate false data corresponding to the target network data based on a quantum generative adversarial network; The target network data and the false data are merged in random order to obtain a target sample. 6. The device according to claim 4 or 5, characterized in that the sending module is specifically used to: Obtaining a detection accuracy value of the first network intrusion model; When the detection accuracy value is greater than or equal to a preset accuracy value, sending the first model parameter and the detection accuracy value to a server; The preset precision value is determined based on the highest precision value and the lowest precision value. 7. A terminal comprising: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that the processor is used to read the program in the memory to implement the steps in the model training method as described in any one of claims 1 to 3. 8. A readable storage medium for storing a program, characterized in that when the program is executed by a processor, the steps in the model training method as described in any one of claims 1 to 3 are implemented. 9. A computer program product, characterized in that it comprises computer instructions, which, when executed by a processor, implement the steps of the model training method as described in any one of claims 1 to 3.