A map data federated learning method and device based on bayesian expectation

Abstract

The application relates to a map data federation learning method and device based on Bayesian expectation, which comprises the following steps: obtaining map data to be learned in a federation, and dividing and reorganizing the map data by using a Vino mosaic and spatial dissimilarity; based on the divided and reorganized map data and a preset federation learning model, distributing the map data of each client, the model parameters and the distribution parameters of the corresponding sub-model by calculating a confidence value and a Bayesian expectation maximization; updating the model parameters and the distribution parameters of the sub-model of each client based on a federation average algorithm and a stochastic gradient maximization; and returning the updated distribution parameters and confidence value of the sub-model of each client to a server side until the preset federation learning model converges.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of high-precision maps and deep learning, and particularly relates to a map data federated learning method and device based on Bayesian expectation. BACKGROUND

[0002] The current federated learning and confidence value technology of high-precision map mapping has not reached the mature standard in the industry. Privacy protection is an important issue of crowdsourcing mapping, and the purpose is to protect the personal data related to the task and the participants and part of the intermediate results, so as to publish and execute the space-time task without exposing the information and physical location of the task and the personnel. This kind of method is currently mainly restricted by three aspects. First, after the crowdsourcing data is distributed, due to the difference in the source or the single distribution method, the distribution variance of the label / response variable data is too large, and the difference of the expectation side of the conditional distribution of the corresponding explanatory variable data is assumed to be zero, so that the prediction performance of the federated learning client model is degraded. Second, the connotation deviates. Even if the distribution parameters of the label data have consistency, the connotations (such as) of the explanatory variable data forming the distribution are different, so that the sub-model is difficult to fit or converge in federated learning. Third, the distribution of the overall data quality and quantity is uneven, especially for a kind of data with spatial properties such as a map, the internal correlation is easily destroyed in the distribution of federated learning, and meaningful results cannot be obtained in the sub-model fitting algorithm. SUMMARY

[0003] To solve the problem of balancing in the distribution process of the federated learning client of high-precision map mapping, in a first aspect of the application, a map data federated learning method based on Bayesian expectation is provided, comprising: acquiring map data to be federated learned, reorganizing the map data by using Veno tessellation and spatial dissimilarity; based on the reorganized map data and a preset federated learning model, distributing the map data of each client and the model parameters and distribution parameters of the corresponding sub-model by calculating the confidence value and the Bayesian expectation maximization; updating the model parameters and distribution parameters of the sub-model of each client based on the federated averaging algorithm and the stochastic gradient maximization; and returning the updated distribution parameters and confidence value of the sub-model of each client to the server side until the preset federated learning model converges.

[0004] In some embodiments of the application, the reorganizing the map data by using Veno tessellation and spatial dissimilarity comprises: rasterizing the map data and counting the trajectory density of each grid image pixel point; calculating the dissimilarity between any two points in the map data based on the trajectory density of each grid image pixel point; and reorganizing the map data based on the dissimilarity between any two points in the map data.

[0005] Further, the dividing and reorganizing the map data based on the dissimilarity between any two points in the map data comprises: clustering and dividing the map data by a preset clustering algorithm to obtain a plurality of clustered map blocks; calculating the spatial dissimilarity of each map block with other blocks, and re-merging the plurality of map blocks according to the spatial dissimilarity.

[0006] In some embodiments of the present application, the distributing the map data of each client and the model parameters and distribution parameters of the corresponding sub-model by the calculation of the confidence value and the Bayesian expectation maximization based on the divided and reorganized map data and the preset federated learning model comprises: determining the required number of iterations of the preset federated learning model and the number of blocks of the map data division; determining the optimization objective function of the corresponding sub-model of each client based on the required number of iterations and the number of blocks; calculating the confidence value of each client based on the optimization objective function and the normal prior assumption; and distributing the confidence value to the corresponding client.

[0007] Further, the updating the model parameters and distribution parameters of the sub-model of each client based on the federated averaging algorithm and the stochastic gradient maximization comprises: converting the optimization objective function into the maximization calculation of the distribution parameters of the sub-model based on the maximization of the lower bound of the evidence and the Monte Carlo simulation; and iteratively solving the maximization of a plurality of parameters of each sub-model by the stochastic gradient algorithm to obtain the updated distribution parameters of each sub-model.

[0008] In the above embodiments, the returning the updated distribution parameters of the sub-model of each client and the confidence value to the server side until the preset federated learning model converges comprises: returning the updated distribution parameters of the sub-model of each client and the confidence value to the server side to obtain the next iteration model parameters of each sub-model.

[0009] The second aspect of the present application provides a map data federated learning device based on Bayesian expectation, comprising: an acquisition module configured to acquire map data to be federated learned, and divide and reorganize the map data by using the tiling of the Viennois tessellation and the spatial dissimilarity; a distribution module configured to distribute the map data of each client and the model parameters and distribution parameters of the corresponding sub-model by the calculation of the confidence value and the Bayesian expectation maximization based on the divided and reorganized map data and a preset federated learning model; an update module configured to update the model parameters and distribution parameters of the sub-model of each client based on the federated averaging algorithm and the stochastic gradient maximization; and a return module configured to return the updated distribution parameters of the sub-model of each client and the confidence value to the server side until the preset federated learning model converges.

[0010] In a third aspect, the present application provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the federated learning method of map data based on Bayesian expectation provided in the first aspect of the present application.

[0011] In a fourth aspect, the present application provides a computer readable medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the federated learning method of map data based on Bayesian expectation provided in the first aspect of the present application.

[0012] The present application has the following beneficial effects:

[0013] The present application relates to a federated learning method of map data based on Bayesian expectation and a device thereof, the method comprising: obtaining map data to be federated learned, and dividing and reorganizing the map data by using a tessellation and spatial dissimilarity; based on the divided and reorganized map data and a preset federated learning model, distributing map data of each client and model parameters and distribution parameters of a corresponding sub-model by calculating confidence values and Bayesian expectation maximization; updating the model parameters and the distribution parameters of the sub-model of each client based on a federated averaging algorithm and a stochastic gradient maximization; and returning the updated distribution parameters and confidence values of the sub-model of each client to a server side until the preset federated learning model converges. It can be seen that the present application integrates spatial characteristics in the data before federated learning distribution, retains the spatial correlation of the data itself, and maximizes the correction of the balance; at the same time, local characteristics are introduced in the random distribution process of federated learning, solving problems including data label distribution deviation and connotation deviation. Finally, the present application proposes a framework of expectation maximization comprehensive federated learning for parameter updating, which reduces the heterogeneity of the global model and the sub-model. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A basic flowchart of the federated learning method of map data based on Bayesian expectation in some embodiments of the present application;

[0015] Figure 2 A specific flowchart of the federated learning method of map data based on Bayesian expectation in some embodiments of the present application;

[0016] Figure 3 A flowchart of the basic principle of the federated learning method of map data based on Bayesian expectation in some embodiments of the present application;

[0017] Figure 4 A basic structure diagram of the federated learning device of map data based on Bayesian expectation in some embodiments of the present application;

[0018] Figure 5 A specific structural schematic diagram of a map data federated learning device based on Bayesian expectation in some embodiments of the present application;

[0019] Figure 6 A structural schematic diagram of an electronic device in some embodiments of the present application. DETAILED DESCRIPTION

[0020] The principles and features of the present application are described below in conjunction with the accompanying drawings, which are only used to explain the present application and are not intended to limit the scope of the present application.

[0021] Reference Figure 1 With Figure 2 In a first aspect of the present application, a map data federated learning method based on Bayesian expectation is provided, comprising: S100. obtaining map data to be federated learned, and dividing and reorganizing the map data by using Vino tessellation and spatial dissimilarity; S200. based on the divided and reorganized map data and a preset federated learning model, distributing the map data of each client and the model parameters and distribution parameters of the corresponding sub-model by calculating the confidence value and the Bayesian expectation maximization; S300. updating the model parameters and distribution parameters of the sub-model of each client based on the federated averaging algorithm and the stochastic gradient maximization; S400. returning the updated distribution parameters and confidence value of the sub-model of each client to the server side until the preset federated learning model converges.

[0022] It can be understood that, as a federated learning method of map data, the basic elements of the present application include map data and a federated learning model, the federated learning model at least includes one server side and one client, the server side distributes sample data and necessary parameters (gradient or loss) of a machine learning model to one or more clients, each client trains by using the sample data and the necessary parameters in each round (each time) iteration, and returns the training data to the server side, the server side receives the returned data, updates the necessary parameters and the sample data, and distributes the updated necessary parameters and the sample data to each client again until the entire federated learning model converges.

[0023] In step S100 of some embodiments of the present application, the dividing and reorganizing of the map data by using Vino tessellation and spatial dissimilarity comprises:

[0024] S101. rasterizing the map data and counting the trajectory density of each raster image pixel; specifically, the trajectory density is counted by using the rasterized map. The collected map data is divided into 5x5km 2 square region of interest (Region of Interest) into a raster image, and then the raster image is converted into a 1x1m2 The size is as a pixel point of a grid map, and the size of the region of interest and the pixel point can be adjusted according to actual map data production and computing power. The pixel coordinate is denoted as: {s i | = 1…N}, where N is the number of pixel points on the grid map, s i That is, the coordinates of the i-th grid pixel, and again: y(s i ) is the trajectory density at the coordinates,

[0025] S102. Based on the trajectory density of each grid image pixel, the dissimilarity between any two points in the map data is calculated;

[0026] Specifically, the dissimilarity between two points located at s k , s l coordinates is defined as:

[0027]

[0028] Where |y(s k )-y(s l )| is the absolute value of the difference between the trajectory densities of the two points, and ||s k , -s l || is the Euclidean distance between the two points. Then, in order to integrate spatial information, we establish a Voronoi tessellation for the points at coordinates s1…s M with trajectory density greater than 0. The specific algorithm is maturely applied (such as Centroidal Voronoi Tessellation), which will not be repeated here.

[0029] S103. Based on the dissimilarity between any two points in the map data, the map data is divided and reorganized.

[0030] Further, in step S103, the dissimilarity between any two points in the map data is calculated, and the map data is divided and reorganized, including: clustering the map data by a preset clustering algorithm of the map data to obtain a plurality of clustered map blocks; calculating the spatial dissimilarity of each map block and other blocks, and re-merging the plurality of map blocks according to the spatial dissimilarity. Specifically, first define the mathematical symbols needed. After the Voronoi tessellation establishes a polygon, two points s k , s l are defined as adjacent, and the adjacent relationship is denoted as: s k ~ s lIf the total number of map clustering cut blocks is denoted as K (the total block number is an adjustable parameter, which is obtained according to experience, a rough clustering algorithm, or cross-validation of a test set through a later federated learning algorithm), and the pixel point sets contained in the two clustered map blocks k1 and k2 are denoted as: The difference between the blocks is:

[0031]

[0032] Wherein is the number of pixel points contained in the kth cluster, is the average trajectory density of the kth cluster block, represents The average Euclidean distance between all pairs of points with adjacent relationship in

[0033] Then, the initial clustering and merging are iterated to form new map blocks in the following specific way:

[0034] 1. Directly take all M points as initial clusters, that is: u u u = 1 … M.

[0035] 2. In the order of u = 0, …, M-K+1, search for two clusters u1 and u2, which need to satisfy:

[0036] a) The point in the iteration order is the smallest and The points in have adjacent relationship.b) If u1 < u2 in the above case, the clusters u1 and u2 are merged into u1, that is: And Conversely.

[0037] 3. Until the iteration ends when u takes the value of M-K+1.

[0038] At this point, J clusters can be formed on the original map, and when entering the confidence and Bayesian expectation maximization algorithm, each cluster will be distributed to different client computing sub-model gradients.

[0039] In step S200 of some embodiments of the present application, based on the reorganized map data after division and the preset federated learning model, the confidence value and the Bayesian expectation maximization calculation are used to distribute the map data of each client and the model parameters and distribution parameters of the corresponding sub-model, including:

[0040] S201. Determine the number of iterations required by the preset federated learning model and the number of blocks of the map data division;

[0041] ​​Specifically, we set the total number of iterations required for the map algorithm (such as neural network) to be calculated as: T, and during the iterations from 0 to T-1 times, the parameters to be optimized are distributed to J blocks according to the previous clustering.

[0042] S202. Based on the number of required iterations and the number of blocks, determine the optimization objective function of the sub-model corresponding to each client;

[0043] Specifically, let the total model parameter be a vector ω, and the parameters of the J sub-models be ω 1：J The optimization objective function is:

[0044]

[0045] Where D is the map data, D j is the data distributed in J parts, f j (ω j ;D j ) is the learning algorithm (neural network or other convex optimization, etc.) of the sub-model corresponding to the parameter ω on each client.

[0046] S203. Based on the optimization objective function and the normal prior assumption, calculate the confidence value of each client; distribute the confidence value to the corresponding client.

[0047] Specifically, the parameter ω j of the sub-model has a normal prior assumption: And is not available, it needs to be approximated by distribution, that is, expectation maximization. At the same time, the sub-model confidence value ρ j is introduced here.

[0048] Then, when t=0, initialize the parameter ω as the iteration result, distribute the parameters to 1…J sub-models according to the clustering block distribution, and fit the multivariate normal distribution to estimate And calculate the confidence value:

[0049]

[0050] Where is the trace of the covariance matrix , representing the uncertainty, is the square of the l2 norm of the difference between the distribution mean and the model parameter vector, representing the model offset.

[0051] In step S300 of some embodiments of the application, the model parameters and distribution parameters of the sub-model of each client are updated based on the federated average algorithm and the stochastic gradient maximization, comprising:

[0052] S301. Based on the maximization of the lower bound of evidence and Monte Carlo simulation, the optimization objective function is converted into the calculation of the maximization of the distribution parameters of the sub-models;

[0053] Specifically, starting from the (t+1)th iteration, the ELBO (evidence lower bound) is maximized, that is:

[0054]

[0055] Since It is not easy to find the approximation formula, we first calculate the proportional lower bound of the KL divergence, that is:

[0056]

[0057] Then use Monte Carlo simulation L times ω j Substitute into the sub-model objective function:

[0058]

[0059] S302. Through the stochastic gradient algorithm, the maximization of multiple parameters of each sub-model is iteratively solved, and the updated distribution parameters of each sub-model are obtained.

[0060] Specifically, the stochastic gradient algorithm is used to iteratively maximize the ω j，l The average and covariance are calculated to obtain the updated distribution parameters:

[0061] In step S400 of the above embodiment, the distribution parameters and confidence values of the updated sub-models of each client are returned to the server side until the preset federated learning model converges, including: returning the distribution parameters and confidence values of the updated sub-models of each client to the server side to obtain the next iteration model parameters of each sub-model. Specifically, after the J clients are updated, the current confidence value and the updated distribution mean are returned to the server side, and the updated parameters are normalized:

[0062]

[0063] Iterate until T times or the algorithm converges to update the parameters after federated learning.

[0064] Reference Figure 3 ​In a specific embodiment of the present application, the map data federated learning method based on Bayesian expectation comprises the following steps: step 1, reorganizing the map data by using Veno tessellation and dissimilarity; step 2, distributing the client model confidence calculation; step 3, federated learning client model parameter update; and step 4, returning the server-side global model parameter update.

[0065] Embodiment 2

[0066] Reference Figure 4 In a second aspect of the present application, a map data federated learning device 1 based on Bayesian expectation is provided, comprising: an acquisition module 11 configured to acquire map data to be federated learned, and reorganize the map data by using Veno tessellation and spatial dissimilarity; a distribution module 12 configured to distribute, based on the reorganized map data and a preset federated learning model, the map data of each client and the model parameters and distribution parameters of the corresponding sub-model through the calculation of confidence value and Bayesian expectation maximization; an update module 13 configured to update the model parameters and distribution parameters of the sub-model of each client based on federated averaging algorithm and stochastic gradient maximization; and a return module 14 configured to return the updated distribution parameters and confidence value of the sub-model of each client to the server side until the preset federated learning model converges.

[0067] Further, the acquisition module 11 comprises: a statistical unit configured to rasterize the map data and count the trajectory density of each grid image pixel; a calculation unit configured to calculate the dissimilarity between any two points in the map data based on the trajectory density of each grid image pixel; and a division unit configured to reorganize the map data based on the calculation of the dissimilarity between any two points in the map data.

[0068] Reference Figure 5 In a specific embodiment of the present application, the map data federated learning device 1 based on Bayesian expectation comprises: a trajectory density statistical module, a spatial information clustering module, a client confidence value module, a client model parameter update module, and a server-side update module. It can be understood that the cooperation of the above-mentioned trajectory density statistical module, spatial information clustering module and client confidence value module is equivalent to the functions realized by the acquisition module and the distribution module.

[0069] Embodiment 3

[0070] Reference Figure 6 In a third aspect of the present application, an electronic device is provided, comprising: one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the map data federated learning method based on Bayesian expectation of the first aspect of the present application.

[0071] The electronic device 500 can include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501 that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 502 or loaded into a random access memory (RAM) 503 from a storage device 508. Various programs and data required for the operation of the electronic device 500 are also stored in the RAM 503. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0072] In general, the following devices can be connected to the I / O interface 505: input devices 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 508 including, for example, a hard disk, etc.; and communication devices 509. The communication devices 509 can allow the electronic device 500 to communicate wirelessly or wired with other devices to exchange data. Although Figure 6 The electronic device 500 is shown with various devices, but it should be understood that all of the shown devices are not required to be implemented or present. More or less devices can alternatively be implemented or present. Figure 6 Each block shown in the middle can represent a device or, as desired, multiple devices.

[0073] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the methods of embodiments of the present disclosure are executed. It should be noted that the computer readable medium described in embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination of the two. The computer readable storage medium may, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, device or apparatus. In embodiments of the present disclosure, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take a variety of forms, including but not limited to, an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, device or apparatus. The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to, wire, cable, RF (radio frequency), or the like, or any suitable combination of the above.

[0074] The computer readable medium described above can be included in the electronic device described above; or can exist separately from the electronic device and be not assembled in the electronic device. The computer readable medium described above carries one or more computer programs, which, when executed by the electronic device, cause the electronic device to:

[0075] Computer program code for carrying out operations of embodiments of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++, Python, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0076] The computer program instructions can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0077] The foregoing is merely illustrative of the principles of the application, and the application should not be limited to such detail. Rather, the spirit and scope of the application are limited solely by the claims.

Claims

1. A Bayesian expectation-based map data federated learning method, characterized in that, The method comprises the following steps: obtaining map data to be federated learning, and dividing and reorganizing the map data by using a tiling and spatial dissimilarity; based on the divided and reorganized map data and a preset federated learning model, distributing the map data of each client and the model parameters and distribution parameters of the corresponding sub-model by calculating a confidence value and a Bayesian expectation maximum value: determining the number of iterations required by the preset federated learning model and the number of blocks of the divided map data; based on the number of iterations required and the number of blocks, determining the optimization objective function of the corresponding sub-model of each client; based on the optimization objective function and a normal prior assumption, calculating the confidence value of each client; distributing the confidence value to the corresponding client; based on a federated averaging algorithm and a stochastic gradient maximum value, updating the model parameters and distribution parameters of the sub-model of each client: based on the maximum value of the evidence lower bound and Monte Carlo simulation, converting the optimization objective function into the calculation of the maximum value of the distribution parameters of the sub-model; by using a stochastic gradient algorithm, iteratively solving the maximum value of multiple parameters of each sub-model to obtain the updated distribution parameters of each sub-model; returning the updated distribution parameters and confidence value of the sub-model of each client to the server side until the preset federated learning model converges.

2. The Bayesian expectation-based map data federated learning method according to claim 1, characterized in that, The division and reorganization of the map data by using the tiling and spatial dissimilarity comprises: rasterizing the map data and counting the trajectory density of each grid image pixel; and establishing a tiling on the points with coordinates having a trajectory density greater than 0; based on the trajectory density of each grid image pixel, calculating the dissimilarity between any two points in the map data; based on the calculation of the dissimilarity between any two points in the map data, dividing and reorganizing the map data.

3. The Bayesian expectation-based map data federated learning method according to claim 2, characterized in that, The division and reorganization of the map data based on the calculation of the dissimilarity between any two points in the map data comprises: by using a preset clustering algorithm of the map data, clustering and dividing the map data to obtain a plurality of clustered map blocks; calculating the spatial dissimilarity between each map block and other blocks, and recombining the plurality of map blocks according to the spatial dissimilarity.

4. The Bayesian expectation-based map data federated learning method according to any one of claims 1 to 3, characterized in that, The returning of the updated distribution parameters and confidence value of the sub-model of each client to the server side until the preset federated learning model converges comprises: returning the updated distribution parameters and confidence value of the sub-model of each client to the server side to obtain the next iteration model parameters of each sub-model. 5.A Bayesian expectation-based map data federated learning device, characterized in that, The method comprises the following steps: an acquisition module is configured to obtain map data to be federated learning, and divide and reorganize the map data by using a tiling and spatial dissimilarity; a distribution module is configured to distribute the map data of each client and the model parameters and distribution parameters of the corresponding sub-model by calculating a confidence value and a Bayesian expectation maximum value based on the divided and reorganized map data and a preset federated learning model: determine the number of iterations required by the preset federated learning model and the number of blocks of the divided map data; based on the number of iterations required and the number of blocks, determine the optimization objective function of the corresponding sub-model of each client; based on the optimization objective function and a normal prior assumption, calculate the confidence value of each client; distribute the confidence value to the corresponding client; an updating module configured to update model parameters and distribution parameters of the sub-models of each client based on a federated averaging algorithm and a stochastic gradient maximization, convert an optimization objective function into a calculation of maximization of the distribution parameters of the sub-models based on maximization of a lower bound of evidence and Monte Carlo simulation, and iteratively solve maximization of a plurality of parameters of each sub-model by a stochastic gradient algorithm to obtain updated distribution parameters of each sub-model; a back transmission module configured to back transmit the updated distribution parameters of the sub-models of each client and the confidence value to the server side until a preset federated learning model converges.

6. The Bayesian expectation-based map data federated learning apparatus according to claim 5, characterized in that, The acquisition module comprises: a statistical unit configured to rasterize the map data and count a trajectory density of each grid image pixel point; a calculation unit configured to calculate a dissimilarity between any two points in the map data based on the trajectory density of each grid image pixel point; a division unit configured to divide and reorganize the map data based on the dissimilarity between any two points in the map data.

7. An electronic device, comprising: one or more processors; a storage device configured to store one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the federated learning method of map data based on Bayesian expectation as claimed in any one of claims 1 to 4.

8. A computer readable medium having stored thereon a computer program, wherein, The computer program, when executed by a processor, implements the federated learning method of map data based on Bayesian expectation as claimed in any one of claims 1 to 4.

Images