Modeling method applied to mobile authentication, coordination node and client node

Abstract

The application discloses a modeling method applied to mobile authentication, a coordination node and a client node, and relates to the field of mobile authentication. The modeling method applied to mobile authentication is applied to the coordination node and comprises the following steps: classifying a plurality of client nodes to obtain K types of client nodes; and performing the following steps in a loop until the system model converges: for a first type of client node, fusing model parameters uploaded by the first type of client node to obtain parameters of an intra-class fusion model of the first type of client node; fusing parameters of intra-class fusion models of the K types of client nodes to obtain system model parameters; and distributing the system model parameters to the plurality of client nodes; wherein K is an integer greater than or equal to 1, and the first type of client node is any type of client node among the K types of client nodes. The above scheme can solve the non-independent and identically distributed phenomenon and improve the precision of the model.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of mobile authentication, and particularly relates to a modeling method applied to mobile authentication, a coordination node and a client node. BACKGROUND

[0002] The basic idea of federated learning is to build a high-performance model shared by multiple participants under the condition of guaranteeing user privacy and data security, aiming to solve the problems of privacy protection and data island. The specific method is to use the data of multiple parties to jointly model, and the data of each party does not leave the local and is not exposed to each modeling party during the modeling process, so as to achieve the purpose of protecting user privacy and data security, and at the same time, a high-performance model is established by using the data of each party.

[0003] Horizontal federated learning is an important part of federated learning, and is mainly applied to the case where the data structures of multiple participants are the same, but the data quantity is insufficient, and a high-performance model is expected to be obtained by joint modeling under the condition of not exposing the data privacy to each other. In horizontal federated learning, the algorithm model and data characteristics of each participant are consistent, each participant trains a local model using local data, and the parameters are transmitted to the coordination party, the coordination party aggregates the model parameters uploaded by each party into a global model, and then transmits the global model to each participant for local update.

[0004] However, due to the difference in user behavior, there is an inevitable phenomenon of non-independent and identically distributed when collecting user data by different terminals. The traditional horizontal federation directly uses mobile terminals as modeling nodes for joint modeling, ignores the solution to the non-independent and identically distributed phenomenon, and thus the model precision is often very low, which cannot achieve the purpose of solving problems by establishing a model through federated learning. SUMMARY

[0005] The embodiments of the application provide a modeling method applied to mobile authentication, a coordination node and a client node, which can solve the problem that the existing federated learning model establishment method ignores the non-independent and identically distributed phenomenon, resulting in low model precision.

[0006] To solve the above technical problem, the embodiments of the application provide a modeling method applied to mobile authentication, which is applied to a coordination node and includes the following steps.

[0007] Classify a plurality of client nodes to obtain K types of client nodes;

[0008] Cyclically execute the following steps until the system model converges:

[0009] For the first type of client node, fuse the model parameters uploaded by the first type of client node to obtain the parameters of the intra-class fusion model of the first type of client node;

[0010] fusing parameters of the intra-class fusion model of the K types of client nodes to obtain system model parameters;

[0011] downloading the system model parameters to the plurality of client nodes;

[0012] wherein K is an integer greater than or equal to 1, and the first type of client node is any one of the K types of client nodes.

[0013] Optionally, the fusing, for the first type of client node, of the model parameters uploaded by the first type of client node to obtain the parameters of the intra-class fusion model of the first type of client node comprises:

[0014] cyclically performing the following steps until the number of iterative updates of the local model by the client node reaches a first threshold value:

[0015] fusing the model parameters uploaded by the first type of client node to obtain the parameters of the intra-class fusion model of the first type of client node;

[0016] sending the parameters of the intra-class fusion model to each of the first type of client nodes;

[0017] receiving first parameters sent by each of the first type of client nodes;

[0018] wherein the first parameters are model parameters obtained by the client node after iterative updating of the local model according to the parameters of the intra-class fusion model.

[0019] Optionally, the fusing of the parameters of the intra-class fusion model of the K types of client nodes to obtain system model parameters comprises:

[0020] determining model accuracies corresponding to the parameters of the intra-class fusion model of each type of client node;

[0021] obtaining parameters of a first intra-class fusion model with a model accuracy less than a second threshold value and parameters of a second intra-class fusion model with a model accuracy greater than or equal to the second threshold value;

[0022] determining a first number of the parameters of the first intra-class fusion model and the parameters of each of the second intra-class fusion models whose Euclidean distances are higher than a third threshold value;

[0023] fusing the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to a fourth threshold value to obtain system model parameters.

[0024] Optionally, after the first number of the Euclidean distances between the parameters of the first intra-class fusion model and the parameters of each second intra-class fusion model is determined to be higher than a third threshold, the method further comprises:

[0025] deleting the parameters of the first intra-class fusion model whose first number is greater than a fourth threshold.

[0026] Optionally, the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold are fused to obtain the system model parameters, comprising:

[0027] The model accuracy corresponding to the parameters of the intra-class fusion model is determined as a weight, and the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold are weighted and averaged to perform inter-class model parameter fusion to obtain the system model parameters.

[0028] Optionally, the method of determining the model accuracy corresponding to the parameters of the intra-class fusion model of each class of client nodes comprises:

[0029] obtaining the accuracy of the local behavior data sent by the client nodes belonging to the same class;

[0030] According to the accuracy of the local behavior data, the model accuracy corresponding to the parameters of the intra-class fusion model of the client nodes belonging to the same class is determined.

[0031] Optionally, the method of classifying the plurality of client nodes to obtain K classes of client nodes comprises:

[0032] obtaining the encrypted feature distribution vectors sent by the plurality of client nodes;

[0033] According to the feature distribution vectors, the plurality of client nodes are classified to obtain K classes of client nodes.

[0034] Optionally, the method of classifying the plurality of client nodes to obtain K classes of client nodes according to the feature distribution vectors comprises:

[0035] The plurality of client nodes corresponding to the feature distribution vectors are divided into K classes by clustering.

[0036] Embodiments of the present application also provide a modeling method applied to mobile authentication, applied to a client node, comprising:

[0037] collecting feature data generated by each event trigger;

[0038] generating a local model according to the feature data;

[0039] sending the model parameters corresponding to the local model to a coordination node;

[0040] receiving the system model parameter sent by the coordination node;

[0041] updating the local model according to the system model parameter.

[0042] Optionally, before the sending of the model parameter corresponding to the local model to the coordination node, the method further comprises:

[0043] obtaining a feature distribution vector;

[0044] encrypting the feature distribution vector;

[0045] sending the encrypted feature distribution vector to the coordination node.

[0046] Optionally, the obtaining of the feature distribution vector comprises:

[0047] sorting the data of each feature field in the feature data within a preset statistical time;

[0048] obtaining the data located at a preset sorting position in the sorting;

[0049] determining the feature distribution vector according to the data at the preset sorting position.

[0050] Optionally, the encrypting of the feature distribution vector comprises:

[0051] encrypting the feature distribution vector by using noise subject to Laplace distribution to obtain the encrypted feature distribution vector.

[0052] Optionally, the method further comprises:

[0053] receiving the parameter of the intra-class fusion model sent by the coordination node;

[0054] iteratively updating the local model according to the parameter of the intra-class fusion model.

[0055] Optionally, after the sending of the model parameter corresponding to the local model to the coordination node, the method further comprises:

[0056] obtaining the accuracy of the local behavior data by the local model;

[0057] sending the accuracy of the local behavior data to the coordination node.

[0058] Optionally, the method further comprises:

[0059] obtaining behavior data of a to-be-authenticated event;

[0060] matching the behavior data with the local model to output an authentication result of the to-be-authenticated event.

[0061] The embodiment of the application further provides a coordination node, comprising:

[0062] a classification module, configured to classify a plurality of client nodes, and obtain K types of client nodes

[0063] a first fusion module, configured to, for a first type of client node, fuse model parameters uploaded by the first type of client node, and obtain parameters of an intra-class fusion model of the first type of client node;

[0064] a second fusion module, configured to fuse parameters of the intra-class fusion models of the K types of client nodes to obtain system model parameters;

[0065] a first sending module, configured to distribute the system model parameters to the plurality of client nodes;

[0066] wherein K is an integer greater than or equal to 1, and the first type of client node is any type of client node among the K types of client nodes.

[0067] The embodiment of the application further provides a coordination node, comprising a transceiver and a processor;

[0068] the processor is configured to classify a plurality of client nodes, and obtain K types of client nodes;

[0069] the following steps are cyclically executed until a system model converges:

[0070] for a first type of client node, fuse model parameters uploaded by the first type of client node, and obtain parameters of an intra-class fusion model of the first type of client node;

[0071] fuse parameters of the intra-class fusion models of the K types of client nodes to obtain system model parameters;

[0072] the transceiver is configured to distribute the system model parameters to the plurality of client nodes;

[0073] wherein K is an integer greater than or equal to 1, and the first type of client node is any type of client node among the K types of client nodes.

[0074] The embodiment of the application further provides a coordination node, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the modeling method applied to mobile authentication when executing the program.

[0075] The embodiment of the application further provides a client node, comprising:

[0076] The collection module is configured to collect feature data generated by each event trigger;

[0077] The generation module is configured to generate a local model according to the feature data;

[0078] The second sending module is configured to send model parameters corresponding to the local model to a coordination node;

[0079] The first receiving module is configured to receive system model parameters sent by the coordination node;

[0080] The first updating module is configured to update the local model according to the system model parameters.

[0081] Embodiments of the present application also provide a client node, which comprises a transceiver and a processor;

[0082] The processor is configured to collect feature data generated by each event trigger;

[0083] The processor is configured to generate a local model according to the feature data;

[0084] The transceiver is configured to send model parameters corresponding to the local model to a coordination node;

[0085] The transceiver is configured to receive system model parameters sent by the coordination node;

[0086] The processor is configured to update the local model according to the system model parameters.

[0087] Embodiments of the present application also provide a client node, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the modeling method for mobile authentication when executing the program.

[0088] Embodiments of the present application also provide a readable storage medium, which stores a computer program, and the program implements the steps of the modeling method for mobile authentication when executed by a processor.

[0089] The present application has the following beneficial effects:

[0090] The above scheme can solve the non-independent and identically distributed phenomenon and improve the precision of the model by first classifying a plurality of client nodes, then cyclically performing fusion on model parameters uploaded by client nodes belonging to the same class to obtain parameters of an intra-class fusion model of the client nodes belonging to the same class, fusing parameters of the intra-class fusion models of the multiple client nodes to obtain system model parameters, and finally sending the system model parameters corresponding to the converged system model to the client nodes. BRIEF DESCRIPTION OF DRAWINGS

[0091] Figure 1 is a flowchart of a modeling method for mobile authentication applied to a coordination node according to an embodiment of the present application;

[0092] Figure 2 is an interaction framework diagram according to an embodiment of the present application;

[0093] Figure 3 is a flowchart of a method for removing a poor intra-fusion model according to an embodiment of the present application;

[0094] Figure 4 is a general flowchart according to an embodiment of the present application;

[0095] Figure 5 is a module diagram of a coordination node according to an embodiment of the present application;

[0096] Figure 6 is a structural diagram of a coordination node according to an embodiment of the present application;

[0097] Figure 7 is a flowchart of a modeling method for mobile authentication applied to a client node according to an embodiment of the present application;

[0098] Figure 8 is a module diagram of a client node according to an embodiment of the present application;

[0099] Figure 9 is a structural diagram of a client node according to an embodiment of the present application. DETAILED DESCRIPTION

[0100] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0101] The terms "first", "second", and the like in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than that illustrated or described herein, and the objects distinguished by "first", "second", and the like are generally a class, and are not limited to the number of objects, for example, the first object can be one or more. In addition, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / ", generally indicates that the objects before and after are in an "or" relationship.

[0102] It is worth noting that the technology described in the embodiments of the present application is not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA) and other systems. The terms "system" and "network" in the embodiments of the present application are often used interchangeably, and the described technology can be used in the above-mentioned systems and radio technologies, as well as other systems and radio technologies. The following description describes a New Radio (NR) system for example purposes, and NR terminology is used in most of the following description, but these technologies can also be applied to applications other than NR system applications, such as 6th Generation (6G) communication systems.

[0103] The modeling method applied to mobile authentication, the coordination node and the client node provided by the embodiments of the present application will be described in detail below in combination with the drawings, specific embodiments and application scenarios.

[0104] As shown in Figure 1 At least one embodiment of the present application provides a modeling method applied to mobile authentication, applied to a coordination node, comprising:

[0105] Step 101, classifying a plurality of client nodes to obtain K classes of client nodes;

[0106] It should be noted that the client node can be regarded as a modeling node for modeling, and K is an integer greater than or equal to 1.

[0107] The following steps are executed in a loop until the system model converges:

[0108] Step 102, for the first class of client nodes, fusing the model parameters uploaded by the first class of client nodes to obtain the parameters of the intra-class fusion model of the first class of client nodes;

[0109] It should be noted that the first type of client node is any type of K type of client node. It should be noted here that the client node determines the local model by using the feature data generated by each event trigger, uses the feature data to determine the local model, obtains the model parameters corresponding to the local model, and sends the model parameters to the coordination node. After obtaining the model parameters corresponding to the local models of the plurality of client nodes, the coordination node does not directly fuse these model parameters, but first fuses the model parameters of the client nodes belonging to the same type according to the classification information of the client nodes, so as to avoid the non-independent and identically distributed phenomenon of the client nodes.

[0110] Step 103, fusing the parameters of the intra-class fusion model of the K type of client nodes to obtain system model parameters;

[0111] It should be noted that this step is to fuse the parameters of the plurality of intra-class fusion models obtained by intra-class fusion to obtain an inter-class fusion model, that is, a final system model. The system model has system model parameters. By determining the intra-class fusion model corresponding to the client nodes of the same class first, and then determining the inter-class fusion model, the accuracy of the model can be improved.

[0112] By cyclically executing steps 102 and 103, a relatively stable system model can be finally obtained, and the accuracy of the system model can be ensured.

[0113] Step 104, distributing the system model parameters to the plurality of client nodes;

[0114] It should be noted that after the coordination node obtains the system model parameters, the coordination node sends the system model parameters to the client nodes. The client nodes can update the local model according to the system model parameters. Further, in the case that the client nodes collect authentication events, the client nodes need to obtain behavior data of the to-be-authenticated event, and then match the behavior data with the local model to output an authentication result of the to-be-authenticated event.

[0115] Optionally, in at least one embodiment of the present application, the optional implementation of step 102 is:

[0116] The following steps are executed in a loop until the number of iterative updates of the local model by the client nodes reaches a first threshold value:

[0117] Step 1021, fusing the model parameters uploaded by the first type of client nodes to obtain the parameters of the intra-class fusion model of the first type of client nodes;

[0118] Step 1022, sending the parameters of the intra-class fusion model to each of the first type of client nodes;

[0119] Step 1023, receiving the first parameter sent by each of the first type of client nodes;

[0120] The first parameter is the model parameter obtained by the client node after iteratively updating the local model according to the parameters of the intra-class fusion model.

[0121] That is, the coordination node and the client node cyclically execute steps 1021 to 1023 until the parameters of the intra-class fusion model meet the requirements of parameter fusion of the inter-class model. It should be noted that because the client node iteratively updates the local model once, the reporting of the model parameter corresponding to the local model is triggered, and the coordination node also updates the intra-class fusion model once, that is, the number of times of iteratively updating the local model by the client node is consistent with the number of times of iteratively updating the intra-class fusion model by the coordination node.

[0122] It should be noted that the model parameters corresponding to the local models of the client nodes belonging to the same class are fused by using the federated learning (fedAvg) method in the embodiments of the present application, and the specific implementation process is as follows:

[0123] First, the client nodes belonging to the same class upload the model parameters obtained after completing one round of iteration of the local model using local data to the coordination node;

[0124] Second, the coordination node averages the model parameters corresponding to the local models of the client nodes belonging to the same class in the last time, and returns them to the client nodes of the class;

[0125] Finally, each client node belonging to the same class substitutes the model parameters obtained by averaging into the local model update to perform the next round of model iteration.

[0126] Further, in at least one embodiment of the embodiments of the present application, the optional implementation process of step 103 is as follows:

[0127] Step 1031, determining the model accuracy corresponding to the parameters of the intra-class fusion model of each class of client nodes;

[0128] Optionally, the further implementation process of this step is as follows: obtaining the accuracy of the local behavior data sent by the client nodes belonging to the same class; determining the model accuracy corresponding to the parameters of the intra-class fusion model of the client nodes belonging to the same class according to the accuracy of the local behavior data.

[0129] It should be noted that the client node first obtains the accuracy of the local behavior data through the local model, and then sends the accuracy of the local behavior data to the coordination node. The coordination node calculates the average accuracy of each intra-class fusion model according to the accuracy of the local behavior data of the client node, as the model accuracy corresponding to the parameter of each intra-class fusion model. For example, the coordination node adds and averages the accuracy of the local behavior data of each client node belonging to the same class to determine the average value as the model accuracy corresponding to the parameter of the intra-class fusion model.

[0130] In step 1032, the parameters of the first intra-class fusion model with the model accuracy less than the second threshold value and the parameters of the second intra-class fusion model with the model accuracy greater than or equal to the second threshold value are obtained.

[0131] It should be noted that this step is to divide the parameters of the first intra-class fusion model and the parameters of the second intra-class fusion model among all the parameters of the intra-class fusion model. Optionally, the obtained parameters of the first intra-class fusion model can be multiple.

[0132] In step 1033, a first number of the parameters of the first intra-class fusion model with the Euclidean distance higher than a third threshold value from the parameters of each second intra-class fusion model is determined.

[0133] It should be noted that this process is to calculate the Euclidean distance between each parameter of the first intra-class fusion model and each parameter of the second intra-class fusion model, and then determine the number of the Euclidean distance higher than the third threshold value.

[0134] In step 1034, the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model with the first number less than or equal to a fourth threshold value are fused to obtain the system model parameters.

[0135] Optionally, if the first number is greater than the fourth threshold value, the parameters of the first intra-class fusion model with the first number greater than the fourth threshold value are deleted, that is, in this case, the parameters of the first intra-class fusion model are not considered, only the parameters of the first intra-class fusion model with the first number less than or equal to the fourth threshold value are considered. If there is no parameter of the first intra-class fusion model with the first number less than or equal to the fourth threshold value, the coordination node only fuses the parameters of the multiple second intra-class fusion models to obtain the system model parameters.

[0136] For example, there are four in-class fusion models in total, the model accuracy corresponding to the parameters of the in-class fusion model A is A1, the model accuracy corresponding to the parameters of the in-class fusion model B is B1, the model accuracy corresponding to the parameters of the in-class fusion model C is C1, and the model accuracy corresponding to the parameters of the in-class fusion model D is D1, and only the model accuracy of the parameters of the in-class fusion model A is less than the second threshold value, then the Euclidean distances of the parameters of the in-class fusion model A with the parameters of the in-class fusion model B, the in-class fusion model C and the in-class fusion model D are calculated respectively, which are d1, d2 and d3; and d2 and d3 are higher than the third threshold value, then the first number is 2, if the first number is 2 and less than or equal to the fourth threshold value (for example, the fourth threshold value is set to 1), the parameters of the in-class fusion model A are retained, and the parameters of the in-class fusion model A, the in-class fusion model B, the in-class fusion model C and the in-class fusion model D are fused to obtain the system model parameters; if the first number is 2 and greater than the fourth threshold value (for example, the fourth threshold value is set to 3), the parameters of the in-class fusion model A are deleted, and only the parameters of the in-class fusion model B, the in-class fusion model C and the in-class fusion model D are fused to obtain the system model parameters.

[0137] Optionally, in at least one embodiment of the present application, the specific implementation of fusing the parameters of the second in-class fusion model and the parameters of the first in-class fusion model whose first number is less than or equal to the fourth threshold value to obtain the system model parameters comprises:

[0138] The model accuracy corresponding to the parameters of the in-class fusion model is determined as a weight, and the parameters of the second in-class fusion model and the parameters of the first in-class fusion model whose first number is less than or equal to the fourth threshold value are weighted and averaged to perform inter-class model parameter fusion and obtain the system model parameters.

[0139] For example: the parameters of the second in-class fusion model and the parameters of the first in-class fusion model whose first number is less than or equal to the fourth threshold value are three, which are M1, M2 and M3, and the model accuracy of the parameters of the corresponding in-class fusion model is P1, P2 and P3, then the three models are fused according to the following formula to obtain M4, that is:

[0140] M4=(P1*M1+P2*M2+P3*M3) / (P1+P2+P3).

[0141] Optionally, in at least one embodiment of the present application, the implementation of classifying the client nodes comprises:

[0142] Step 1011, obtaining the encrypted feature distribution vector sent by the plurality of client nodes;

[0143] It should be noted that the encrypted feature distribution vector is usually sent from the client node to the coordinating node. Specifically, the client node first sorts the data of each feature field in the feature data within a preset statistical time period; obtains the data located at the preset sorting position; then determines the feature distribution vector based on the data at the preset sorting position; then encrypts the feature distribution vector using noise that follows a Laplace distribution to obtain the encrypted feature distribution vector, and finally sends the encrypted feature distribution vector to the coordinating node.

[0144] Step 1012: Classify the multiple client nodes according to the feature distribution vector to obtain K types of client nodes;

[0145] Optionally, this step can be implemented by: using clustering to divide the multiple client nodes corresponding to the feature distribution vector into K classes; where K is an integer greater than or equal to 1.

[0146] It should be noted that the clustering method can be K-means clustering, HAC hierarchical agglomerative clustering, etc.

[0147] For example, a software development kit (SDK) deployed on a client node (e.g., a terminal) collects the following characteristics from each key press and release by the user:

[0148] The data includes user ID, timestamp, area of ​​the user's finger touching the screen, pressure applied to the screen, speed of keystrokes, acceleration, angular acceleration, finger movement trajectory, etc. The sampling rate is generally selected from 100Hz to 180Hz, and the collection period is generally selected from about 7 days.

[0149] Secondly, each client node calculates the 0%, 25%, 50%, 75%, and 100% quantiles (i.e., feature values ​​ranked at the 0%, 25%, 50%, 75%, and 100% positions) of each feature field in its historical data (it should be noted that the area of ​​the user's finger touching the screen, the force of pressing the screen, the speed of pressing the keyboard, acceleration, angular acceleration, and finger movement trajectory are all feature fields). The five quantiles of all feature fields are combined in order of the feature fields to form a data record, which serves as the feature distribution vector for that client node. A Laplace-distributed noise 'l' is added to the feature distribution vector to achieve ε-differential privacy. This vector is then passed to the coordinating node, which uses clustering to divide the encrypted feature distribution vector into k classes, thus classifying the corresponding terminal client nodes into k classes.

[0150] Example as follows:

[0151] Category 1: First terminal, second terminal... tenth terminal;

[0152] The second type: the eleventh terminal, the twelfth terminal …… the fifteenth terminal;

[0153] …

[0154] It should be noted that the application embodiment changes the authentication technology interaction mode of the past, that is, the user's terminal extracts the user's original privacy data and uploads it to the business platform for modeling. The user queries the authentication result through the interface to the business platform every time he logs in. The improved federated learning technology is applied to the authentication technology to prevent user privacy leakage, shorten the event ID authentication time, and improve efficiency.

[0155] The application embodiment is applied to the mobile authentication interaction mode, that is, the user terminal collects data through SDK and stores it locally. Each modeling node establishes a local model using local data, only uploads encrypted model parameters to the business platform, the business platform classifies the terminals through classification technology, and deletes the model parameters of the poor effect class after using the model accuracy of each class and the difference between the model parameters of other classes. The weighted average method is used to aggregate the inter-class model parameters as system model parameters, and broadcast to the terminal to update the local model. The terminal deletes the historical record data and only retains the model. The user generates time ID behavior data every time he authenticates, matches it with the stored local model, and outputs the authentication result.

[0156] The following takes the mobile authentication technology as an example (this is a typical modeling node sample data set, which is not independent and identically distributed. Each terminal collects data with different dimensional data distribution differences) to illustrate the application process of at least one embodiment of the application as follows: the user terminal (i.e. the modeling node, also corresponding to the client node mentioned above) collects data through SDK and stores it locally. Each terminal establishes a local model using local data, only uploads encrypted model parameters to the business platform (corresponding to the coordination node above), the business platform groups the terminals through grouping technology, and deletes the model parameters of the poor effect group after using the anomaly detection technology. The model parameters of the terminals in the group are aggregated using aggregation technology and broadcast to the terminal to update the local model. The terminal deletes the historical record data and only retains the model. The user generates time ID behavior data every time he authenticates, matches it with the stored local model, and outputs the authentication result. The specific interaction mode is as shown in Figure 2 .

[0157] First of all, it needs to be pointed out that the coordination node will get an intra-class fusion model after fusing the model parameters of the same class of client nodes, and the intra-class fusion model corresponds to an intra-class fusion model parameter (model parameter). When using the intra-class fusion model for fusion, the intra-class fusion model parameter is used for fusion to obtain a system model, and the system model corresponds to a system model parameter. Therefore, in the subsequent description, if it is for the model or the comparison between the models, it is uniformly described as a model, and the fusion process is described as a model parameter (or model parameter).

[0158] Process one, intra-class modeling process

[0159] After classifying the modeling nodes, intra-class modeling can be realized, and each class of modeling nodes can perform intra-class modeling according to the following steps:

[0160] Step 1: Each class of modeling nodes uses local data to train a local model, and transmits the model parameters obtained in each round to the coordination node for encryption;

[0161] Step 2: The coordination node fuses the model parameters of the same class of modeling nodes according to the class to which the modeling nodes belong, and obtains k fused model parameters by using the FedAvg algorithm;

[0162] Step 3: The coordination node transmits the k fused model parameters to the corresponding intra-class modeling nodes, and each modeling node uses the fused model parameters as the initial value of the next round of local model iteration.

[0163] Step 4: Repeat steps 1-3 until the number of iterations reaches the number of inter-class model parameter fusion requirements, and perform the model screening process.

[0164] Process two, model screening process

[0165] It should be noted that this process is to eliminate the model corresponding to the class with low accuracy and large difference between the model parameters and the model parameters of other classes, that is, to eliminate the model with poor effect, and to retain the high-quality model to improve the overall model precision. The specific implementation process includes:

[0166] Step 5: When the number of iterations of each class reaches the number of inter-class model parameter fusion requirements, each modeling node calculates the accuracy of the local behavior data and uploads it to the business platform (coordination node), and the business platform calculates the average accuracy of each class of modeling nodes as the model accuracy corresponding to each class of model parameters.

[0167] Step 6: select the intra-class fusion model with an average accuracy lower than the set accuracy threshold, and count the number s of other intra-class fusion models with a Euclidean distance higher than the set distance threshold, if the number s is higher than the set distance number threshold, the intra-class fusion model of the class is removed, otherwise, it is saved. The specific flow chart of hypothesis testing of the intra-class fusion model is shown in Figure 3 .

[0168] Process three, inter-class model fusion process

[0169] The business platform retains each class model after removing the model, and performs weighted average according to the accuracy of each class model as the weight to obtain the entire system model, and broadcasts it to each modeling node. Continue the cycle of process one to process three until the entire system model converges and stabilizes.

[0170] Process four, event ID authentication process

[0171] It should be noted that each modeling node retains the system model and deletes the historical data. The user generates the behavior data of the event ID to be authenticated, matches it with the local stored model to output the authentication result of the event ID.

[0172] It should be noted that, as shown in Figure 4 , the overall implementation process of the embodiment of the application is as follows:

[0173] First, all terminals are classified, and each intra-class terminal trains a local model using local data and uploads it to the business platform. Second, the business platform aggregates the model on each intra-class modeling node to obtain the model of each class, referred to as the first class model, the second class model, the kth class model, and the like.

[0174] Next, each class model is retained on the business platform through process one and process two.

[0175] Then, the retained class model uses process three to fuse the parameters of the inter-class model to obtain the system model, and broadcasts the system model parameters corresponding to the system model to each terminal.

[0176] Next, each terminal uses it as the initial model of the next round of local model iteration, and continues to execute process one to process three until the entire system model converges and stabilizes.

[0177] As can be seen from the above, at least one embodiment of the application can achieve the following beneficial effects:

[0178] 1. The participating modeling nodes are clustered, the nodes with similar data distribution are organized together for modeling, which effectively improves the accuracy of the model; on the other hand, the poor group model is removed to prevent abnormal terminal data from affecting the overall model.

[0179] 2. Protect the privacy of users, prevent the leakage of user original data in transmission, and protect the privacy of users compared with previous authentication technologies.

[0180] 3. Compared with previous authentication technologies, the pressure on the service platform is reduced, and the system crash caused by the authentication of all users at the same time is prevented.

[0181] 4. The problem of seriously affecting the accuracy of the federated learning model due to the non-independent and identically distributed phenomenon of each node data can be solved.

[0182] As shown in Figure 5 At least one embodiment of the present application also provides a coordination node 500, comprising:

[0183] The classification module 501 is configured to classify a plurality of client nodes, and obtain K-class client nodes

[0184] The first fusion module 502 is configured to, for the first-class client nodes, fuse the model parameters uploaded by the first-class client nodes to obtain the parameters of the intra-class fusion model of the first-class client nodes.

[0185] The second fusion module 503 is configured to fuse the parameters of the intra-class fusion models of the K-class client nodes to obtain system model parameters.

[0186] It should be noted that the first fusion module 502 and the second fusion module 503 are cyclically operated until the system model converges.

[0187] The first sending module 504 is configured to distribute the system model parameters to the plurality of client nodes.

[0188] Wherein, K is an integer greater than or equal to 1, and the first-class client nodes are any one of the K-class client nodes.

[0189] Optionally, the first fusion module 502 comprises:

[0190] The following units are cyclically operated until the number of iteration updates of the local model by the client nodes reaches a first threshold value:

[0191] The first fusion unit is configured to fuse the model parameters uploaded by the first-class client nodes to obtain the parameters of the intra-class fusion model of the first-class client nodes.

[0192] The first sending unit is configured to send the parameters of the intra-class fusion model to each of the first-class client nodes.

[0193] The first receiving unit is configured to receive the first parameters sent by each of the first-class client nodes.

[0194] The first parameter is a model parameter obtained by iteratively updating a local model according to a parameter of the intra-class fusion model by the client node.

[0195] Optionally, the second fusion module 503 comprises:

[0196] The first determination unit is configured to determine a model accuracy corresponding to the parameter of the intra-class fusion model of each class of client nodes.

[0197] The first acquisition unit is configured to acquire the parameter of the first intra-class fusion model with the model accuracy less than the second threshold value and the parameter of the second intra-class fusion model with the model accuracy greater than or equal to the second threshold value.

[0198] The second determination unit is configured to determine a first number of intra-class fusion models whose parameters have a Euclidean distance higher than a third threshold value from the parameter of each second intra-class fusion model.

[0199] The second fusion unit is configured to fuse the parameters of the second intra-class fusion models and the parameters of the first intra-class fusion models whose first numbers are less than or equal to a fourth threshold value, to obtain the system model parameter.

[0200] Optionally, after the second determination unit determines the first number of intra-class fusion models whose parameters have a Euclidean distance higher than a third threshold value from the parameter of each second intra-class fusion model, the method further comprises:

[0201] The deletion unit is configured to delete the parameters of the first intra-class fusion models whose first numbers are greater than the fourth threshold value.

[0202] Optionally, the second fusion unit is configured to:

[0203] determine the model accuracy corresponding to the parameter of the intra-class fusion model as a weight, and perform weighted average on the parameters of the second intra-class fusion models and the parameters of the first intra-class fusion models whose first numbers are less than or equal to the fourth threshold value, to perform inter-class model parameter fusion and obtain the system model parameter.

[0204] Optionally, the first determination unit is configured to:

[0205] acquire an accuracy of local behavior data sent by the client nodes belonging to the same class;

[0206] determine the model accuracy corresponding to the parameter of the intra-class fusion model of the client nodes belonging to the same class according to the accuracy of the local behavior data.

[0207] Optionally, the classification module 501 comprises:

[0208] a second obtaining unit, configured to obtain encrypted feature distribution vectors sent by a plurality of client nodes;

[0209] a third obtaining unit, configured to classify the plurality of client nodes according to the feature distribution vectors, and obtain K classes of client nodes.

[0210] Optionally, the third obtaining unit is configured to:

[0211] divide the plurality of client nodes corresponding to the feature distribution vectors into K classes by using a clustering manner.

[0212] It should be noted that the coordination node provided by at least one embodiment of the present application is a coordination node capable of executing the modeling method applied to mobile authentication, and all embodiments of the modeling method applied to mobile authentication are applicable to the coordination node and can achieve the same or similar beneficial effects.

[0213] At least one embodiment of the present application further provides a coordination node, which comprises a transceiver and a processor.

[0214] The processor is configured to classify the plurality of client nodes and obtain K classes of client nodes.

[0215] The following steps are cyclically executed until the system model converges:

[0216] For the first class of client nodes, the model parameters uploaded by the first class of client nodes are fused to obtain the parameters of an intra-class fusion model of the first class of client nodes.

[0217] The parameters of the intra-class fusion models of the K classes of client nodes are fused to obtain system model parameters.

[0218] The transceiver is configured to distribute the system model parameters to the plurality of client nodes.

[0219] K is an integer greater than or equal to 1, and the first class of client nodes is any class of client nodes in the K classes of client nodes.

[0220] Optionally, the processor is configured to:

[0221] The following steps are cyclically executed until the number of iterative updates of the local model by the client nodes reaches a first threshold value:

[0222] The model parameters uploaded by the first class of client nodes are fused to obtain the parameters of an intra-class fusion model of the first class of client nodes.

[0223] The parameters of the intra-class fusion model are sent to each client node in the first class of client nodes.

[0224] The transceiver is configured to receive the first parameter sent by each of the first type of client nodes.

[0225] The first parameter is a model parameter obtained by iteratively updating a local model according to a parameter of the intra-class fusion model.

[0226] Optionally, the processor is configured to:

[0227] determine a model accuracy corresponding to the parameter of the intra-class fusion model of each type of client node;

[0228] obtain the parameter of the first intra-class fusion model whose model accuracy is less than a second threshold value and the parameter of the second intra-class fusion model whose model accuracy is greater than or equal to the second threshold value;

[0229] determine a first number of the parameters of the first intra-class fusion model whose Euclidean distance with the parameter of each second intra-class fusion model is higher than a third threshold value;

[0230] fuse the parameter of the second intra-class fusion model and the parameter of the first intra-class fusion model whose first number is less than or equal to a fourth threshold value to obtain a system model parameter.

[0231] Optionally, the processor is further configured to delete the parameter of the first intra-class fusion model whose first number is greater than the fourth threshold value.

[0232] Optionally, the processor is configured to determine the model accuracy corresponding to the parameter of the intra-class fusion model as a weight, and perform weighted averaging on the parameter of the second intra-class fusion model and the parameter of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value to obtain the system model parameter.

[0233] Optionally, the processor is configured to:

[0234] obtain an accuracy rate of local behavior data sent by the client nodes belonging to the same class;

[0235] determine a model accuracy corresponding to the parameter of the intra-class fusion model of the client nodes belonging to the same class according to the accuracy rate of the local behavior data.

[0236] Optionally, the processor is configured to:

[0237] obtain an encrypted feature distribution vector sent by the plurality of client nodes;

[0238] classify the plurality of client nodes according to the feature distribution vector to obtain K types of client nodes.

[0239] Optionally, the processor is configured to:

[0240] The multiple client nodes corresponding to the feature distribution vector are divided into K classes using clustering.

[0241] like Figure 6 As shown, this embodiment of the invention also provides a coordination node, including a processor 600, a transceiver 610, a memory 620, and a program stored in the memory 620 and executable on the processor 600; wherein the transceiver 610 is connected to the processor 600 and the memory 620 via a bus interface, and the processor 600 is used to read the program in the memory and execute the following processes:

[0242] Classify multiple client nodes to obtain K types of client nodes;

[0243] Repeat the following steps until the system model converges:

[0244] For the first type of client node, the model parameters uploaded by the first type of client node are fused to obtain the parameters of the intra-class fused model of the first type of client node;

[0245] The system model parameters are obtained by fusing the parameters of the intra-class fusion model of K-type client nodes;

[0246] The system model parameters are sent to the multiple client nodes via transceiver 610;

[0247] Where K is an integer greater than or equal to 1, and the first type of client node is any one of the K types of client nodes.

[0248] Transceiver 610 is used to receive and send data under the control of processor 600.

[0249] Among them, Figure 6 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 600 and memory represented by memory 620 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 610 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium, including wireless channels, wired channels, optical fibers, and other transmission media.

[0250] The processor 600 is responsible for managing the bus architecture and general processing, while the memory 620 can store the data used by the processor 600 when performing operations.

[0251] Optionally, the processor 600 can be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) or a CPLD (Complex Programmable Logic Device), and the processor can also adopt a multi-core architecture.

[0252] The processor is configured to execute any of the methods provided by the embodiments of the present application by invoking the computer program stored in the memory. The processor and the memory can also be physically arranged separately.

[0253] Optionally, the processor 600 is configured to read the program in the memory and further execute the following processes:

[0254] The following steps are repeatedly executed until the number of iterative updates of the local model by the client nodes reaches a first threshold value:

[0255] The model parameters uploaded by the first type of client nodes are fused to obtain the parameters of the intra-class fusion model of the first type of client nodes;

[0256] The parameters of the intra-class fusion model are sent to each of the first type of client nodes;

[0257] The first parameter sent by each of the first type of client nodes is received through the transceiver 610;

[0258] The first parameter is the model parameter obtained by the client node after iteratively updating the local model according to the parameters of the intra-class fusion model.

[0259] Optionally, the processor 600 is configured to read the program in the memory and further execute the following processes:

[0260] The model accuracy corresponding to the parameters of the intra-class fusion model of each type of client node is determined;

[0261] The parameters of the first intra-class fusion model with a model accuracy less than a second threshold value and the parameters of the second intra-class fusion model with a model accuracy greater than or equal to the second threshold value are obtained;

[0262] A first number of the parameters of the first intra-class fusion model and the parameters of each of the second intra-class fusion model whose Euclidean distance is higher than a third threshold value is determined;

[0263] Fuse the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value, to obtain the system model parameters.

[0264] Optionally, the processor 600 is configured to read a program in the memory, and further perform the following process:

[0265] Delete the parameters of the first intra-class fusion model whose first number is greater than the fourth threshold value.

[0266] Optionally, the processor 600 is configured to read a program in the memory, and further perform the following process:

[0267] Determine the model accuracy corresponding to the parameters of the intra-class fusion model as a weight, and perform weighted average on the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value, to perform inter-class model parameter fusion and obtain the system model parameters.

[0268] Optionally, the processor 600 is configured to read a program in the memory, and further perform the following process:

[0269] Obtain the accuracy of the local behavior data sent by the client nodes belonging to the same class;

[0270] According to the accuracy of the local behavior data, determine the model accuracy corresponding to the parameters of the intra-class fusion model of the client nodes belonging to the same class.

[0271] Optionally, the processor 600 is configured to read a program in the memory, and further perform the following process:

[0272] Obtain the encrypted feature distribution vector sent by the plurality of client nodes;

[0273] According to the feature distribution vector, classify the plurality of client nodes to obtain K classes of client nodes.

[0274] Optionally, the processor 600 is configured to read a program in the memory, and further perform the following process:

[0275] Divide the plurality of client nodes corresponding to the feature distribution vector into K classes by using clustering.

[0276] At least one embodiment of the present application also provides a coordination node, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements each process in the modeling method for mobile authentication applied to the coordination node and achieves the same technical effects. To avoid repetition, details are not repeated here.

[0277] At least one embodiment of the present application also provides a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements each process in the modeling method for mobile authentication applied to a coordination node as described above and achieves the same technical effects. To avoid repetition, details are not described here. The computer readable storage medium is, for example, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, etc.

[0278] Corresponding to the implementation of the coordination node side, as shown in Figure 7 At least one embodiment of the present application provides a modeling method for mobile authentication applied to a client node, comprising:

[0279] Step 701, collecting feature data generated by each event trigger;

[0280] Step 702, generating a local model according to the feature data;

[0281] Step 703, sending model parameters corresponding to the local model to the coordination node;

[0282] Step 704, receiving system model parameters sent by the coordination node;

[0283] Step 705, updating the local model according to the system model parameters.

[0284] Optionally, before the model parameters corresponding to the local model are sent to the coordination node, the method further comprises:

[0285] Obtaining a feature distribution vector;

[0286] Encrypting the feature distribution vector;

[0287] Sending the encrypted feature distribution vector to the coordination node.

[0288] Optionally, the obtaining of the feature distribution vector comprises:

[0289] Sorting data of each feature field in the feature data within a preset statistical time;

[0290] Obtaining data located at a preset sorting position in the sorting;

[0291] Determining the feature distribution vector according to the data at the preset sorting position.

[0292] Optionally, the encryption of the feature distribution vector comprises:

[0293] The feature distribution vector is encrypted by using noise subject to a Laplace distribution to obtain an encrypted feature distribution vector.

[0294] Optionally, the method further comprises:

[0295] receiving the parameters of the intra-class fusion model sent by the coordination node;

[0296] performing iterative updating of the local model according to the parameters of the intra-class fusion model.

[0297] Optionally, after the model parameters corresponding to the local model are sent to the coordination node, the method further comprises:

[0298] obtaining the accuracy of the local behavior data by the local model;

[0299] sending the accuracy of the local behavior data to the coordination node.

[0300] Optionally, the method further comprises:

[0301] obtaining behavior data of an event to be authenticated;

[0302] matching the behavior data with the local model to output an authentication result of the event to be authenticated.

[0303] It should be noted that all the descriptions about the client node in the above embodiments are applicable to the embodiments of the modeling method applied to mobile authentication, and the same technical effects can be achieved.

[0304] As shown in Figure 8 at least one embodiment of the present application further provides a client node 800, comprising:

[0305] a collection module 801, configured to collect feature data generated by each event trigger;

[0306] a generation module 802, configured to generate a local model according to the feature data;

[0307] a first sending module 803, configured to send model parameters corresponding to the local model to a coordination node;

[0308] a first receiving module 804, configured to receive system model parameters sent by the coordination node;

[0309] a first updating module 805, configured to update the local model according to the system model parameters.

[0310] Optionally, before the first sending module 803 sends the model parameters corresponding to the local model to the coordination node, the method further comprises:

[0311] The first obtaining module is configured to obtain a feature distribution vector.

[0312] The encryption module is configured to encrypt the feature distribution vector.

[0313] The third sending module is configured to send the encrypted feature distribution vector to a coordination node.

[0314] Optionally, the first obtaining module comprises:

[0315] The sorting unit is configured to sort data of each feature field in feature data within a preset statistical time.

[0316] The fourth obtaining unit is configured to obtain data located at a preset sorting position in the sorting.

[0317] The third determining unit is configured to determine the feature distribution vector according to the data located at the preset sorting position.

[0318] Optionally, the encryption module is configured to:

[0319] encrypt the feature distribution vector by using noise subject to a Laplace distribution to obtain an encrypted feature distribution vector.

[0320] Optionally, the client node further comprises:

[0321] The second receiving module is configured to receive parameters of an intra-class fusion model sent by the coordination node.

[0322] The second updating module is configured to perform iterative updating of the local model according to the parameters of the intra-class fusion model.

[0323] Optionally, after the first sending module 803 sends the model parameters corresponding to the local model to the coordination node, the first sending module 803 further comprises:

[0324] The second obtaining module is configured to obtain an accuracy rate of local behavior data by using the local model.

[0325] The fourth sending module is configured to send the accuracy rate of the local behavior data to the coordination node.

[0326] Optionally, the client node further comprises:

[0327] The third obtaining module is configured to obtain behavior data of a to-be-authenticated event.

[0328] The output module is configured to match the behavior data with the local model and output an authentication result of the to-be-authenticated event.

[0329] It should be noted that the client node provided by at least one embodiment of the present application is a client node capable of performing the transmission method described above, and all embodiments of the transmission method described above are applicable to the client node and can achieve the same or similar beneficial effects.

[0330] At least one embodiment of the present application further provides a client node, comprising a transceiver and a processor;

[0331] The processor is configured to collect feature data generated by each event trigger;

[0332] According to the feature data, a local model is generated;

[0333] The transceiver is configured to send model parameters corresponding to the local model to a coordination node;

[0334] The system model parameters sent by the coordination node are received;

[0335] The processor is configured to update the local model according to the system model parameters.

[0336] Optionally, the processor is further configured to:

[0337] Obtain a feature distribution vector;

[0338] The feature distribution vector is encrypted;

[0339] The transceiver is further configured to send the encrypted feature distribution vector to the coordination node.

[0340] Optionally, the processor is configured to:

[0341] Sort the data of each feature field in the feature data within a preset statistical time;

[0342] Obtain data located at a preset sorting position in the sorting;

[0343] According to the data at the preset sorting position, a feature distribution vector is determined.

[0344] Optionally, the processor is configured to:

[0345] The feature distribution vector is encrypted using noise subject to Laplace distribution to obtain an encrypted feature distribution vector.

[0346] The transceiver is further configured to:

[0347] Receive parameters of an intra-class fusion model sent by the coordination node;

[0348] The processor is further configured to perform iterative updating of the local model according to the parameters of the intra-class fusion model.

[0349] The processor is further configured to acquire an accuracy of the local behavior data by the local model.

[0350] The transceiver is further configured to send the accuracy of the local behavior data to a coordination node.

[0351] The processor is further configured to acquire behavior data of an event to be authenticated.

[0352] The behavior data is matched with the local model, and an authentication result of the event to be authenticated is output.

[0353] As shown in Figure 9 The embodiment of the application further provides a client node, comprising a processor 900, a transceiver 910, a memory 920, and a program stored in the memory 920 and executable on the processor 900; wherein the transceiver 910 is connected with the processor 900 and the memory 920 through a bus interface, and the processor 900 is configured to read the program in the memory and perform the following processes:

[0354] Collecting feature data generated by each event trigger;

[0355] Generating a local model according to the feature data;

[0356] Sending model parameters corresponding to the local model to a coordination node through the transceiver 910;

[0357] Receiving system model parameters sent by the coordination node through the transceiver 910;

[0358] Updating the local model according to the system model parameters.

[0359] The transceiver 910 is configured to receive and send data under the control of the processor 900.

[0360] Wherein, in Figure 9In particular embodiments, the bus architecture can include any number of interconnecting buses and bridges, depending on the specific application of the processor 900. The bus architecture can link various circuits of the various circuitries represented by the processor 900 and the memory 920, which is represented by one or more processors and memories, respectively. The bus architecture can also link various other circuitries, such as peripheral devices, voltage regulators, and power management circuitries, which are well known in the art and thus, are not further described herein. The bus interface provides an interface. The transceiver 910 can be a plurality of elements, including a transmitter and a receiver, which provides a means for communicating with various other apparatus over a transmission medium, including a wireless channel, a wired channel, optical cable, and the like. The processor 900 is responsible for managing the bus architecture and general processing, and the memory 920 can store data used by the processor 900 in executing operations. The user interface 930 can also be an interface that can be externally or internally connected to a device as needed for different user devices, including but not limited to a keypad, a display, a speaker, a microphone, a joystick, and the like.

[0361] The processor 900 can be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD), and the processor can also adopt a multi-core architecture.

[0362] Optionally, the processor 900 is configured to read a program in the memory and further perform the following process:

[0363] Obtain a feature distribution vector;

[0364] Encrypt the feature distribution vector;

[0365] Send the encrypted feature distribution vector to a coordination node through the transceiver 910.

[0366] Optionally, the processor 900 is configured to read a program in the memory and further perform the following process:

[0367] Sort data of each feature field in feature data within a preset statistical time;

[0368] Obtain data located at a preset sorting position in the sorting;

[0369] Determine a feature distribution vector according to the data at the preset sorting position.

[0370] Optionally, the processor 900 is configured to read a program in the memory and further perform the following process:

[0371] The feature distribution vector is encrypted by using noise subject to a Laplace distribution to obtain an encrypted feature distribution vector.

[0372] Optionally, the processor 900 is configured to read a program in the memory, and further perform the following process:

[0373] The parameter of the intra-class fusion model sent by the coordination node is received by the transceiver 910;

[0374] The local model is iteratively updated according to the parameter of the intra-class fusion model.

[0375] Optionally, the processor 900 is configured to read a program in the memory, and further perform the following process:

[0376] The accuracy of the local behavior data is obtained by the local model;

[0377] The accuracy of the local behavior data is sent to the coordination node by the transceiver 910.

[0378] Optionally, the processor 900 is configured to read a program in the memory, and further perform the following process:

[0379] The behavior data of the event to be authenticated is obtained;

[0380] The behavior data is matched with the local model to output an authentication result of the event to be authenticated.

[0381] At least one embodiment of the present application also provides a client node comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements each process in the modeling method for mobile authentication applied to the client node and achieves the same technical effects. To avoid repetition, details are not described here.

[0382] At least one embodiment of the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement each process in the modeling method for mobile authentication applied to the client node and achieve the same technical effects. To avoid repetition, details are not described here. The computer readable storage medium may be, for example, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.

[0383] It should be noted that, in the present document, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element. Furthermore, it is to be understood that the method and apparatus of the present application can be carried out by more than one process, method, article, or apparatus either simultaneously, concurrently, or with intervening action that are carried out at the same time, either in a simultaneous fashion or in a fashion that is interleaved in time. For example, the described methods can be performed in a different order from that described, and / or various steps can be combined or omitted, and / or additional steps can be added, without departing from the scope of the present application. Also, features described with respect to certain examples can be combined in other examples.

[0384] From the above description of the embodiments, it is apparent that the above-described method of the embodiments can be realized by means of software and general-purpose hardware platforms, and of course, can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such an understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a ROM / RAM, a magnetic disk, or an optical disk) and includes a number of instructions for causing a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods described in the various embodiments of the present application.

[0385] The embodiments of the present application are described above with reference to the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, rather than limiting, and a person of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the scope protected by the claims.

Claims

1. A modeling method applied to mobile authentication, applied to a coordination node, characterized in that, The method comprises the following steps: classifying a plurality of client nodes to obtain K classes of client nodes; performing the following steps in a loop until the system model converges: for a first class of client nodes, fusing model parameters uploaded by the first class of client nodes to obtain parameters of an intra-class fusion model of the first class of client nodes; fusing the parameters of the intra-class fusion models of the K classes of client nodes to obtain system model parameters; downloading the system model parameters to the plurality of client nodes; wherein K is an integer greater than or equal to 1, and the first class of client nodes is any class of client nodes among the K classes of client nodes; wherein the step of fusing, for a first class of client nodes, model parameters uploaded by the first class of client nodes to obtain parameters of an intra-class fusion model of the first class of client nodes comprises the following steps: performing the following steps in a loop until the number of iterative updates of the local model by the client nodes reaches a first threshold value: fusing model parameters uploaded by the first class of client nodes to obtain parameters of an intra-class fusion model of the first class of client nodes; sending the parameters of the intra-class fusion model to each client node in the first class of client nodes; receiving first parameters sent by each client node in the first class of client nodes; wherein the first parameters are model parameters obtained by the client nodes after iterative updates of a local model according to the parameters of the intra-class fusion model; wherein the step of fusing the parameters of the intra-class fusion models of the K classes of client nodes to obtain system model parameters comprises the following steps: determining model accuracies corresponding to the parameters of the intra-class fusion models of each class of client nodes; obtaining parameters of a first intra-class fusion model with a model accuracy less than a second threshold value and parameters of a second intra-class fusion model with a model accuracy greater than or equal to the second threshold value; determining a first number of first intra-class fusion models whose Euclidean distances with each second intra-class fusion model are higher than a third threshold value; fusing the parameters of the second intra-class fusion models and the parameters of the first intra-class fusion models with a first number less than or equal to a fourth threshold value to obtain system model parameters; wherein after the step of determining the first number of first intra-class fusion models whose Euclidean distances with each second intra-class fusion model are higher than the third threshold value, the method further comprises the following step: deleting the parameters of the first intra-class fusion models with the first number greater than the fourth threshold value; wherein the step of fusing the parameters of the second intra-class fusion models and the parameters of the first intra-class fusion models with the first number less than or equal to the fourth threshold value to obtain system model parameters comprises the following step: determining model accuracies corresponding to the parameters of the intra-class fusion models as weights, and performing weighted averaging on the parameters of the second intra-class fusion models and the parameters of the first intra-class fusion models with the first number less than or equal to the fourth threshold value to obtain system model parameters.

2. The method of claim 1, wherein, The step of determining model accuracies corresponding to the parameters of the intra-class fusion models of each class of client nodes comprises the following steps: obtaining an accuracy rate of local behavior data sent by client nodes belonging to the same class. According to the accuracy of the local behavior data, determine the model accuracy corresponding to the parameters of the intra-class fusion model of the client nodes belonging to the same class.

3. The method of claim 1, wherein, The classification of the plurality of client nodes to obtain K classes of client nodes comprises: Obtain the encrypted feature distribution vector sent by the plurality of client nodes; According to the feature distribution vector, classify the plurality of client nodes to obtain K classes of client nodes.

4. The method of claim 3, wherein, According to the feature distribution vector, classify the plurality of client nodes to obtain K classes of client nodes, comprising: The feature distribution vector corresponding to the plurality of client nodes is divided into K classes by clustering.

5. A modeling method applied to mobile authentication, applied to a client node, characterized in that, Comprise: Collect feature data generated by each event trigger; According to the feature data, generate a local model; Send the model parameters corresponding to the local model to the coordination node; Receive the system model parameters sent by the coordination node; According to the system model parameters, update the local model; Wherein, the system model parameters are the parameters of the intra-class fusion model of the first class of client nodes obtained by the coordination node for the first class of client nodes, and the parameters of the intra-class fusion model of the K classes of client nodes are fused to obtain; The K classes of client nodes are obtained by classifying the plurality of client nodes by the coordination node; K is an integer greater than or equal to 1, and the first class of client nodes is any class of client nodes in the K classes of client nodes; Wherein, the parameters of the intra-class fusion model of the first class of client nodes obtained by fusing the model parameters uploaded by the first class of client nodes, comprising: Loop the following steps until the number of iteration updates of the local model by the client node reaches the first threshold value: Fuse the model parameters uploaded by the first class of client nodes to obtain the parameters of the intra-class fusion model of the first class of client nodes; Send the parameters of the intra-class fusion model to each of the first class of client nodes; Receive the first parameter sent by each of the first class of client nodes; Wherein, the first parameter is the model parameter obtained by the client node after iteratively updating the local model according to the parameters of the intra-class fusion model; Wherein, the system model parameters are obtained by fusing the parameters of the intra-class fusion model of the K classes of client nodes, comprising: Determine the model accuracy corresponding to the parameters of the intra-class fusion model of each class of client nodes; Obtain the parameters of the first intra-class fusion model with a model accuracy less than a second threshold value and the parameters of the second intra-class fusion model with a model accuracy greater than or equal to the second threshold value; Determine the first number whose Euclidean distance between the parameters of the first intra-class fusion model and each of the parameters of the second intra-class fusion model is higher than a third threshold value; Fuse the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model less than or equal to a fourth threshold value to obtain the system model parameters; Further comprising: wherein, after the first number of the determined Euclidean distances between the parameters of the first intra-class fusion model and the parameters of each second intra-class fusion model is higher than a third threshold, further comprising: deleting the parameters of the first intra-class fusion model whose first number is greater than a fourth threshold; wherein, the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold are fused to obtain the system model parameters, comprising:

6. The method of claim 5, wherein, determining the model accuracy corresponding to the parameters of the intra-class fusion model as a weight, and performing weighted average on the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold to obtain the system model parameters. Further comprising: obtaining a feature distribution vector; encrypting the feature distribution vector; 7. The method of claim 6, wherein, sending the encrypted feature distribution vector to the coordination node. The method for obtaining the feature distribution vector comprises: sorting the data of each feature field in the feature data within the preset statistical time; obtaining the data located at the preset sorting position in the sorting; 8. The method of claim 6, wherein, determining the feature distribution vector according to the data at the preset sorting position. The method for encrypting the feature distribution vector comprises:

9. The method of claim 5, wherein, encrypting the feature distribution vector by using noise subject to Laplace distribution to obtain the encrypted feature distribution vector. Further comprising: receiving the parameters of the intra-class fusion model sent by the coordination node; 10. The method of claim 5, wherein, performing iterative update of the local model according to the parameters of the intra-class fusion model. Further comprising: obtaining the accuracy of the local behavior data by the local model; 11. The method of claim 5, wherein, sending the accuracy of the local behavior data to the coordination node. Further comprising: obtaining behavior data of an event to be authenticated; 12. A coordination node, characterized by matching the behavior data with the local model to output an authentication result of the event to be authenticated. Comprise: a classification module, configured to classify a plurality of client nodes to obtain K classes of client nodes a first fusion module, configured to fuse the model parameters uploaded by the first class of client nodes to obtain the parameters of the intra-class fusion model of the first class of client nodes; a second fusion module, configured to fuse the parameters of the intra-class fusion models of the K classes of client nodes to obtain system model parameters; a first sending module, configured to send the system model parameters to the plurality of client nodes; wherein, K is an integer greater than or equal to 1, and the first class of client nodes is any class of client nodes in the K classes of client nodes; wherein, the first fusion module comprises: cyclically running the following units until the number of iterative updates of the local model by the client nodes reaches a first threshold: a first fusion unit, configured to fuse the model parameters uploaded by the first class of client nodes to obtain the parameters of the intra-class fusion model of the first class of client nodes; a first sending unit, configured to send the parameters of the intra-class fusion model to each client node in the first class of client nodes; The first receiving unit is configured to receive first parameters sent by each of the first type of client nodes; The first parameters are model parameters obtained by the client nodes after iterative updating of the local model according to the parameters of the intra-class fusion model; The second fusion module comprises: The first determining unit is configured to determine model accuracy corresponding to the parameters of the intra-class fusion model of each class of client nodes. The first obtaining unit is configured to obtain the parameters of the first intra-class fusion model whose model accuracy is less than the second threshold value and the parameters of the second intra-class fusion model whose model accuracy is greater than or equal to the second threshold value. The second determining unit is configured to determine a first number of intra-class fusion models whose Euclidean distance between the parameters of the first intra-class fusion model and the parameters of each second intra-class fusion model is higher than a third threshold value. The second fusion unit is configured to fuse the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to a fourth threshold value to obtain system model parameters. After the second determining unit determines the first number of intra-class fusion models whose Euclidean distance between the parameters of the first intra-class fusion model and the parameters of each second intra-class fusion model is higher than the third threshold value, the method further comprises: The deleting unit is configured to delete the parameters of the first intra-class fusion model whose first number is greater than the fourth threshold value. The second fusion unit is configured to: Determine the model accuracy corresponding to the parameters of the intra-class fusion model as a weight, and perform weighted averaging on the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value to obtain system model parameters.

13. A coordination node, characterized by The transceiver and the processor are included; The processor is configured to classify the plurality of client nodes to obtain K classes of client nodes. The following steps are repeatedly executed until the system model converges: For the first type of client nodes, fuse the model parameters uploaded by the first type of client nodes to obtain the parameters of the intra-class fusion model of the first type of client nodes. Fuse the parameters of the intra-class fusion models of the K classes of client nodes to obtain system model parameters. The transceiver is configured to send the system model parameters to the plurality of client nodes. K is an integer greater than or equal to 1, and the first type of client nodes are any class of client nodes in the K classes of client nodes. The processor is configured to: The following steps are repeatedly executed until the number of iterative updates of the local model by the client nodes reaches a first threshold value: Fuse the model parameters uploaded by the first type of client nodes to obtain the parameters of the intra-class fusion model of the first type of client nodes. Send the parameters of the intra-class fusion model to each of the first type of client nodes. The transceiver is configured to receive first parameters sent by each of the first type of client nodes. The first parameters are model parameters obtained by the client nodes after iterative updating of the local model according to the parameters of the intra-class fusion model. The processor is configured to: determine a model accuracy corresponding to the parameters of the intra-class fusion model of each type of client node; obtain the parameters of a first intra-class fusion model whose model accuracy is less than a second threshold value and the parameters of a second intra-class fusion model whose model accuracy is greater than or equal to the second threshold value; determine a first number of intra-class fusion models whose Euclidean distance between the parameters of the first intra-class fusion model and the parameters of each second intra-class fusion model is higher than a third threshold value; fuse the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to a fourth threshold value to obtain system model parameters; wherein the processor is further configured to delete the parameters of the first intra-class fusion model whose first number is greater than the fourth threshold value; wherein the processor is configured to determine the model accuracy corresponding to the parameters of the intra-class fusion model as a weight, and to perform weighted averaging on the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value to obtain the system model parameters.

14. A coordination node, characterized by A computer program product comprising a memory, a processor and a computer program stored on the memory and loadable on the processor, the processor implementing the steps of the modeling method for mobile authentication according to any one of claims 1-4 when executing the program.

15. A client node, characterized by comprising: a collection module configured to collect feature data generated by each event trigger; a generation module configured to generate a local model according to the feature data; a second sending module configured to send model parameters corresponding to the local model to a coordination node; a first receiving module configured to receive system model parameters sent by the coordination node; a first updating module configured to update the local model according to the system model parameters; wherein the system model parameters are obtained by the coordination node fusing model parameters uploaded by a first type of client node to obtain parameters of an intra-class fusion model of the first type of client node, and fusing parameters of intra-class fusion models of K types of client nodes; the K types of client nodes are obtained by the coordination node classifying a plurality of client nodes; K is an integer greater than or equal to 1, and the first type of client node is any one of the K types of client nodes; wherein the fusing of the model parameters uploaded by the first type of client node to obtain the parameters of the intra-class fusion model of the first type of client node comprises: recursively performing the following steps until the number of iterative updates of the local model by the client node reaches a first threshold value: fusing the model parameters uploaded by the first type of client node to obtain the parameters of the intra-class fusion model of the first type of client node; sending the parameters of the intra-class fusion model to each client node in the first type of client node; receiving first parameters sent by each client node in the first type of client node; wherein the first parameters are model parameters obtained by the client node after iteratively updating the local model according to the parameters of the intra-class fusion model; The parameters of the intra-class fusion model of the Kth client node are fused to obtain system model parameters, including: Determine the model accuracy corresponding to the parameters of the intra-class fusion model of each class of client nodes; Obtain the parameters of the first intra-class fusion model whose model accuracy is less than the second threshold and the parameters of the second intra-class fusion model whose model accuracy is greater than or equal to the second threshold; Determine the first number of the intra-class fusion model whose Euclidean distance with the parameters of each second intra-class fusion model is higher than the third threshold; Fuse the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold to obtain system model parameters; Wherein, after determining the first number of the intra-class fusion model whose Euclidean distance with the parameters of each second intra-class fusion model is higher than the third threshold, it further includes: Delete the parameters of the first intra-class fusion model whose first number is greater than the fourth threshold; Wherein, the system model parameters obtained by fusing the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold, include: Determine the model accuracy corresponding to the parameters of the intra-class fusion model as the weight, and perform weighted average on the parameters of the second intra-class fusion model and the parameters of the first intra-class fusion model whose first number is less than or equal to the fourth threshold to perform inter-class model parameter fusion and obtain system model parameters.

16. A client node, characterized by Including a transceiver and a processor; The processor is configured to collect feature data generated by each event trigger; According to the feature data, a local model is generated; The transceiver is configured to send the model parameters corresponding to the local model to the coordination node; Receive the system model parameters sent by the coordination node; The processor is configured to update the local model according to the system model parameters; Wherein, the system model parameters are obtained by the coordination node fusing the model parameters uploaded by the first class of client nodes, and the parameters of the intra-class fusion model of the first class of client nodes; the parameters of the intra-class fusion model of the Kth class of client nodes are fused; the Kth class of client nodes is obtained by the coordination node classifying a plurality of client nodes; K is an integer greater than or equal to 1, and the first class of client nodes is any class of client nodes in the Kth class of client nodes; Wherein, the parameters of the intra-class fusion model of the first class of client nodes are obtained by fusing the model parameters uploaded by the first class of client nodes, including: Loop the following steps until the number of iteration updates of the local model by the client node reaches the first threshold: Fuse the model parameters uploaded by the first class of client nodes to obtain the parameters of the intra-class fusion model of the first class of client nodes; Send the parameters of the intra-class fusion model to each client node in the first class of client nodes; Receive the first parameters sent by each client node in the first class of client nodes; The first parameter is a model parameter obtained by iteratively updating a local model according to a parameter of the intra-class fusion model by the client node; The parameter of the intra-class fusion model of the K client nodes is fused to obtain the system model parameter, including: Determining the model accuracy corresponding to the parameter of the intra-class fusion model of each client node; Obtaining the parameter of the first intra-class fusion model with the model accuracy less than the second threshold value and the parameter of the second intra-class fusion model with the model accuracy greater than or equal to the second threshold value; Determining the first number of the first intra-class fusion model whose Euclidean distance with the parameter of each second intra-class fusion model is higher than the third threshold value; Fusing the parameter of the second intra-class fusion model and the parameter of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value to obtain the system model parameter; After determining the first number of the first intra-class fusion model whose Euclidean distance with the parameter of each second intra-class fusion model is higher than the third threshold value, the method further includes: Deleting the parameter of the first intra-class fusion model whose first number is greater than the fourth threshold value; The fusing the parameter of the second intra-class fusion model and the parameter of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value to obtain the system model parameter includes: Determining the model accuracy corresponding to the parameter of the intra-class fusion model as a weight, and performing weighted average on the parameter of the second intra-class fusion model and the parameter of the first intra-class fusion model whose first number is less than or equal to the fourth threshold value to perform inter-class model parameter fusion and obtain the system model parameter.

17. A client node, characterized by The computer program is stored in the memory and executable on the processor, and the processor executes the program to realize the steps of the modeling method applied to mobile authentication according to any one of claims 5-11.

18. A readable storage medium, having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the steps in the modeling method applied to mobile authentication according to any one of claims 1-11.

Images