A user-centered federated learning method and device based on visible light communication

Abstract

This paper provides a user - centered federated learning method and device based on visible light communication. There are multiple optical access points and multiple users within a regional scope. The federated learning method includes: determining the weights of the transmission links between users and each target optical access point according to the number of samples collected by the users and the distances between the users and each target optical access point; dividing all accessible users into multiple user clusters according to a preset physical distance; obtaining the optical access point cluster corresponding to the user cluster according to the target optical access points corresponding to the accessible users in the user cluster; if there is an intersection between multiple optical access point clusters, adjusting the optical access point clusters to obtain the adjusted optical access point clusters; adjusting the user clusters, and performing federated learning based on the adjusted optical access point clusters and the adjusted user clusters. This paper can improve the utilization rate of network resources and data transmission rate, and enhance the performance of federated learning.

Description

Technical Field

[0001] The present invention relates to the field of communications, and in particular, to a user-centered federated learning method and apparatus based on visible light communication. Background Art

[0002] Federated learning (FL) is a distributed and decentralized machine learning method that enables users to collaboratively learn a shared prediction model while keeping the collected samples on local devices. Relying on a wireless network, model parameters are transmitted and updated between a server and users, and the global model is collaboratively trained over the wireless network without aggregating the data collected by users.

[0003] Therefore, the performance of FL depends to a large extent on the total sample size of the selected users. The limited nature of wireless resources in the prior art limits the number of user devices connected to the network for training the FL model. Moreover, in traditional network-centered communication technologies, when a user moves, joins, or leaves a cell covered by a wireless signal, the cell cannot be updated, resulting in low network resource utilization and low data transmission rates.

[0004] Therefore, there is an urgent need for a user-centered federated learning method based on visible light communication that can improve network resource utilization and data transmission rates and enhance the performance of federated learning. Summary of the Invention

[0005] The purpose of the embodiments herein is to provide a user-centered federated learning method and apparatus based on visible light communication to improve network resource utilization and data transmission rates and enhance the performance of federated learning.

[0006] To achieve the above object, on the one hand, the embodiments herein provide a user-centered federated learning method based on visible light communication. A plurality of optical access points and a plurality of users are set within a regional scope. The federated learning method includes:

[0007] Regarding the optical access points among all optical access points whose incident angle to the user is less than the field of view angle of the user as the target optical access points corresponding to the user;

[0008] Determining the weights of the transmission links between the user and each target optical access point according to the number of samples collected by the user and the distances between the user and each target optical access point;

[0009] If the total number of target optical access points is greater than or equal to the total number of users, then all users are accessible users;

[0010] If the total number of target optical access points is less than the total number of users, then all users are sorted according to the weights of the transmission links between the users and each target optical access point, and the same number of users as the total number of target optical access points are obtained according to the sorting result as the accessible users;

[0011] All accessible users are divided into multiple user clusters according to a preset physical distance;

[0012] According to the target optical access points corresponding to the accessible users in the user cluster, the optical access point cluster corresponding to the user cluster is obtained;

[0013] If there is an intersection between multiple optical access point clusters, then the optical access point clusters are adjusted according to the weights of the transmission links between the accessible users and each target optical access point to obtain the adjusted optical access point clusters; wherein the target optical access points in the intersection correspond to multiple accessible users in multiple user clusters at the same time;

[0014] The user clusters are adjusted according to a set delay time to obtain the adjusted user clusters, and federated learning is performed based on the adjusted optical access point clusters and the adjusted user clusters.

[0015] Preferably, the step of dividing all accessible users into multiple user clusters according to a preset physical distance further includes:

[0016] S1: Randomly select any unlabeled accessible user as a user cluster, determine the center of the user cluster, and label the unlabeled accessible user;

[0017] S2: Select another unlabeled accessible user, and determine whether the distance between the other unlabeled accessible user and the center of the user cluster is less than or equal to the preset physical distance;

[0018] S3: If so, make the other unlabeled accessible user belong to the user cluster, label the other unlabeled accessible user, and update the center of the user cluster;

[0019] S4: If not, loop the above S2 to S4 until the distances between all unlabeled accessible users and the center of the user cluster are all greater than the preset physical distance;

[0020] S5: Loop and execute the above S1 to S5 until all accessible users are labeled.

[0021] Preferably, the step of determining the center of the user cluster further includes:

[0022] Make the unlabeled accessible user be the center of the user cluster;

[0023] The further steps of updating the center of the user cluster include:

[0024] Using the other unlabeled accessible users to perform mean processing on the center of the user cluster to update the center of the user cluster.

[0025] Preferably, the further steps of adjusting the optical access point cluster according to the weights of the transmission links between the accessible users and each target optical access point to obtain the adjusted optical access point cluster include:

[0026] Regarding the target optical access points in the intersection as the selected optical access points;

[0027] According to the weights of the transmission links between multiple accessible users and the selected optical access points, determining the uniquely corresponding selected users of the selected optical access points, and the selected user clusters to which the selected users belong;

[0028] Making the selected optical access points belong to the optical access point cluster corresponding to the selected user cluster;

[0029] According to the weights of the transmission links between the other users except the selected users among the multiple accessible users and each target optical access point, determining the other optical access points corresponding to the other users;

[0030] Making the other optical access points belong to the optical access point cluster corresponding to the other user cluster, where the other user cluster is the cluster to which the other users belong.

[0031] Preferably, the further steps of determining the other optical access points corresponding to the other users according to the weights of the transmission links between the other users except the selected users among the multiple accessible users and each target optical access point include:

[0032] Obtaining all the target optical access points corresponding to the other users;

[0033] Removing the selected optical access points from all the target optical access points to obtain the remaining all target optical access points after removal;

[0034] Comparing the weights of the transmission links between the other users and the remaining target optical access points after removal, and regarding the target optical access point corresponding to the largest weight as the other optical access point corresponding to the other users.

[0035] Preferably, the further steps of adjusting the user cluster according to the set delay time to obtain the adjusted user cluster include:

[0036] Calculating the data transmission time of the accessible users according to the uplink rate, uplink data volume, downlink rate, and downlink data volume between the accessible users in the user cluster and the federated learning server;

[0037] When the data transmission time of the accessible user is greater than the set delay time, the accessible user is removed from the user cluster to obtain an adjusted user cluster.

[0038] Preferably, the calculating the data transmission time of the accessible user according to the uplink rate, uplink data volume, downlink rate, and downlink data volume between the accessible user in the user cluster and the federated learning server further includes:

[0039] The data transmission time of the accessible user is calculated by the following formula:

[0040]

[0041] where t is the data transmission time of the accessible user, δ u is the downlink data volume of the accessible user, is the downlink rate of the accessible user, δ G is the uplink data volume of the accessible user, is the uplink rate of the accessible user.

[0042] On the other hand, the embodiments of the present invention provide a user-centered federated learning device based on visible light communication. A plurality of optical access points and a plurality of users are arranged within a regional range. The federated learning device includes:

[0043] A target optical access point determination module, configured to use the optical access points among all optical access points whose incident angle to the user is less than the field of view angle of the user as the target optical access points corresponding to the user;

[0044] A weight determination module, configured to determine the weights of the transmission links between the user and each target optical access point according to the number of samples collected by the user and the distances between the user and each target optical access point;

[0045] An accessible user determination module, configured to: if the total number of target optical access points is greater than or equal to the total number of users, all users are accessible users; if the total number of target optical access points is less than the total number of users, all users are sorted according to the weights of the transmission links between the user and each target optical access point, and users with the same number as the total number of target optical access points are obtained according to the sorting result as the accessible users;

[0046] A user cluster division module, configured to divide all accessible users into a plurality of user clusters according to a preset physical distance;

[0047] An optical access point determination module, configured to obtain an optical access point cluster corresponding to the user cluster according to the target optical access points corresponding to the accessible users in the user cluster;

[0048] An optical access point cluster adjustment module, which is used to adjust the optical access point cluster according to the weights of the transmission links between the accessible users and each target optical access point if there is an intersection between multiple optical access point clusters, so as to obtain an adjusted optical access point cluster; wherein the target optical access points in the intersection correspond to multiple accessible users in multiple user clusters at the same time;

[0049] A user cluster adjustment module, which is used to adjust the user cluster according to the set delay time to obtain an adjusted user cluster, and perform federated learning based on the adjusted optical access point cluster and the adjusted user cluster.

[0050] On the other hand, an embodiment of this article also provides a computer device, including a memory, a processor, and a computer program stored on the memory. When the computer program is run by the processor, it executes the instructions of any one of the above methods.

[0051] On the other hand, an embodiment of this article also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by the processor of a computer device, it executes the instructions of any one of the above methods.

[0052] As can be seen from the technical solutions provided by the embodiments of this article above, through the method of the embodiments of this article, users are first divided into clusters, and then an optical access point cluster is obtained according to the user clusters. After dividing the users into clusters, by adjusting the optical access point cluster and the user cluster, the optical access points in the final optical access point cluster and the users in the user cluster can perform many-to-many communication in a vector transmission manner, reducing the occurrence of communication interference and ensuring the communication quality. And the users in the final user cluster are all users with higher transmission efficiency and better communication quality, improving the performance of federated learning.

[0053] To make the above and other purposes, features and advantages of this article more obvious and understandable, the following specifically enumerates preferred embodiments and cooperates with the attached drawings to make a detailed description as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of this article or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of this article. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 Shows a schematic flowchart of a user-centered federated learning method based on visible light communication provided by an embodiment of this article;

[0056] Figure 2 Shows a schematic flow chart for obtaining an adjusted optical access point cluster provided by the embodiments herein;

[0057] Figure 3 Shows a schematic flow chart for determining other optical access points corresponding to other users provided by the embodiments herein;

[0058] Figure 4 Shows a schematic flow chart for adjusting a user cluster according to a set delay time provided by the embodiments herein;

[0059] Figures 5a - 5d Shows a schematic diagram of the number of users selected by the traditional method and the method herein when the user's field of view angle ranges from 90° to 120° provided by the embodiments herein;

[0060] Figure 6 Shows a schematic diagram of the recognition accuracy corresponding to different numbers of connected users and different sample numbers for each user provided by the embodiments herein;

[0061] Figure 7 Shows a schematic module structure diagram of a user-centered visible light communication-based federated learning device provided by the embodiments herein;

[0062] Figure 8 Shows a schematic structure diagram of a computer device provided by the embodiments herein.

[0063] Explanation of reference signs in the drawings:

[0064] 100, target optical access point determination module;

[0065] 200, weight determination module;

[0066] 300, accessible user determination module;

[0067] 400, user cluster division module;

[0068] 500, optical access point determination module;

[0069] 600, optical access point cluster adjustment module;

[0070] 700, user cluster adjustment module;

[0071] 802, computer device;

[0072] 804, processor;

[0073] 806, memory;

[0074] 808, drive mechanism;

[0075] 810, input / output module;

[0076] 812. Input device;

[0077] 814. Output device;

[0078] 816. Presentation device;

[0079] 818. Graphical user interface;

[0080] 820. Network interface;

[0081] 822. Communication link;

[0082] 824. Communication bus. Detailed implementation manner

[0083] Next, the technical solutions in the embodiments of this article will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this article. Obviously, the described embodiments are only a part of the embodiments of this article, rather than all the embodiments. Based on the embodiments in this article, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this article.

[0084] The performance of federated learning depends to a large extent on the total sample size of the selected users. In the prior art, the limited wireless resources limit the number of user devices connected to the network for training the FL model. Moreover, in traditional network-centric communication technologies, when users move, join, or leave the cell covered by the wireless signal, the cell cannot be updated, resulting in low network resource utilization and low data transmission rate.

[0085] To solve the above problems, the embodiments of this article provide a user-centric federated learning method based on visible light communication. Figure 1 It is a flowchart of a user-centric federated learning method based on visible light communication provided by the embodiments of this article. This specification provides the method operation steps as described in the embodiments or flowcharts, but based on routine or non-creative labor, there may be more or fewer operation steps. The step order listed in the embodiments is only one way among the execution orders of numerous steps and does not represent the only execution order. When the actual system or device product executes, it can be executed in the order of the method shown in the embodiments or the drawings or executed in parallel.

[0086] It should be noted that the terms "first", "second", etc. in the specification, claims and the above-mentioned drawings of this article are used to distinguish similar objects, and do not necessarily 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 this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or equipment.

[0087] Federated Learning (FL) is a distributed and decentralized machine learning method that enables users to collaboratively learn a shared prediction model while keeping the collected samples on the users' local devices. Relying on the wireless network, the model parameters are transmitted and updated between the server and the users, and the global model is collaboratively trained over the wireless network without the need to aggregate the data collected by the users.

[0088] For user u, the FL training aims to find the parameter ω u The minimum loss function under the model: where K u is the number of samples collected by user u, and f uk (ω) is the loss function that obtains the performance of the FL algorithm, which is related to the input vector and the output vector For example, the performance of a regression model can be represented by the root mean square error; while the performance of a classification model can be measured by accuracy.

[0089] Minimizing the loss function is where is the sum of the data samples. Once the FL algorithm converges, all users and the server will share the same model ω 1 = ω 2 = … = ω U = ω G , where ω G is the global variable parameter.

[0090] To utilize more data to train the learning model and thus improve its accuracy, by implementing user selection in FL, more users are associated in the visible light network. When the transmit power of the users is fixed, minimizing the loss function can be transformed into an optimization problem of maximizing the total sample size of the selected users, which can be expressed as the maximum value of the training sample numbers of all users within the cluster, denoted as

[0091] where, The achievable data rate of the downlink optical link for user u Indicates whether user u is selected to be in user cluster C m , where U is the set of all users that can access the optical access points.

[0092] To obtain the user clusters, an exhaustive search method needs to be used, but it has extremely high computational complexity when the number of users increases. To reduce the computational complexity, the user-centric visible light communication-based federated learning method of this embodiment can be utilized.

[0093] Refer to Figure 1 , this paper provides a user-centric visible light communication-based federated learning method. There are multiple optical access points and multiple users set within a regional scope. The federated learning method includes:

[0094] S101: Regarding the optical access points among all optical access points whose incident angle to the user is less than the field of view angle of the user as the target optical access points corresponding to the user;

[0095] S102: Determine the weights of the transmission links between the user and each target optical access point according to the number of samples collected by the user and the distances between the user and each target optical access point;

[0096] S103: If the total number of target optical access points is greater than or equal to the total number of users, then all users are accessible users;

[0097] S104: If the total number of target optical access points is less than the total number of users, then sort all users according to the weights of the transmission links between the user and each target optical access point, and obtain the same number of users as the total number of target optical access points according to the sorting result as the accessible users;

[0098] S105: Divide all accessible users into multiple user clusters according to a preset physical distance;

[0099] S106: Obtain the optical access point cluster corresponding to the user cluster according to the target optical access points corresponding to the accessible users in the user cluster;

[0100] S107: If there is an intersection among multiple optical access point clusters, then adjust the optical access point clusters according to the weights of the transmission links between the accessible users and each target optical access point to obtain the adjusted optical access point clusters; where the target optical access points in the intersection correspond to multiple accessible users in multiple user clusters at the same time;

[0101] S108: Adjust the user clusters according to a set delay time to obtain the adjusted user clusters, and perform federated learning based on the adjusted optical access point clusters and the adjusted user clusters.

[0102] Take the optical access points among all optical access points whose incident angle to the user is less than the user's field of view angle as the target optical access points corresponding to the user. Among them, the target optical access points are the optical access points that can communicate with the user. For the optical access points whose incident angle to the user is greater than or equal to the user's field of view angle, these optical access points cannot communicate with the user and thus are not the target optical access points corresponding to the user.

[0103] To ensure that the number of total target optical access points is greater than or equal to the total number of accessible users, when the number of total target optical access points is less than the total number of users, sort all users according to the weights of the transmission links between the users and each target optical access point, and obtain the same number of users as the total target optical access points according to the sorting result as the accessible users. When sorting, it can be sorted from large to small or from small to large according to the weights, and select the user corresponding to the larger weight according to the sorting result as the accessible user. Among them, the accessible users are the users who can access the target optical access points. In the embodiments of this article, according to the number of samples collected by the user and the distance between the user and each target optical access point, the weight of the transmission link between the user and each target optical access point can be determined, and the weight can be calculated by the following formula:

[0104]

[0105] where l(u, a * (u)) represents the physical distance between user u and target optical access point a * (u), and K u is the number of samples collected by user u.

[0106] Furthermore, all accessible users can be divided into multiple user clusters according to a preset physical distance. The specific division method includes:

[0107] S1: Randomly select any unmarked accessible user as a user cluster, determine the center of the user cluster, and mark the unmarked accessible user;

[0108] S2: Select another unmarked accessible user, and judge whether the distance between the other unmarked accessible user and the center of the user cluster is less than or equal to the preset physical distance;

[0109] S3: If so, make the other unmarked accessible user belong to the user cluster, mark the other unmarked accessible user, and update the center of the user cluster;

[0110] S4: If not, loop S2 to S4 above until the distances between all unmarked accessible users and the center of the user cluster are all greater than the preset physical distance;

[0111] S5: Repeat the above S1 to S5 until all accessible users are marked.

[0112] In the initial state, all accessible users are unmarked. Randomly select any unmarked accessible user a to form a user cluster A alone. At this time, make the unmarked accessible user the center of the user cluster, that is, use the position of the accessible user a as the position of the center o1, and mark user a. Select another unmarked accessible user b, and determine whether the distance between the accessible user b and the center o1 is less than or equal to a preset physical distance. The preset physical distance can be determined according to the actual working conditions and will not be elaborated here. If it is less than or equal to, the accessible user b belongs to the user cluster A, mark the accessible user b, and perform mean processing on the center of the user cluster according to the other unmarked accessible user to update the center of the user cluster, that is, calculate the position of the center point of the two according to the position of the accessible user b and the position of the center o1, and the position of this center point is the position of the new center o2. If it is greater than, the accessible user b does not belong to the user cluster A, then select the accessible user c for judgment until the distances between all unmarked accessible users and the center of the user cluster are greater than the preset physical distance, so as to obtain the final user cluster A. Similarly, repeat the above S1 to S5 until all accessible users are marked, so that all accessible users can be divided into multiple user clusters. For any user cluster, there may be multiple accessible users or only one accessible user.

[0113] After obtaining the user clusters, the optical access point clusters corresponding to the user clusters can be obtained according to the target optical access points corresponding to the accessible users in the user clusters. The target optical access points corresponding to the accessible users are the optical access points that can communicate with the accessible users. Assume that the user cluster A includes accessible users a, b, and c. The accessible user a corresponds to the target optical access point 1, the accessible user b corresponds to the target optical access point 2, and the accessible user c corresponds to the target optical access points 3 and 4. Then the optical access point cluster M corresponding to the user cluster A includes optical access points 1, 2, 3, and 4.

[0114] Inevitably, there may be an optical access point that belongs to the optical access point clusters corresponding to two or more user clusters at the same time. For example, the target optical access point 1 belongs to both the optical access point cluster M corresponding to the user cluster A and the optical access point cluster N corresponding to the user cluster B. This situation means that there is an intersection between multiple optical access point clusters, and then the optical access point clusters need to be adjusted according to the weights of the transmission links between the accessible users and the respective target optical access points to obtain the adjusted optical access point clusters.

[0115] Specifically, refer to Figure 2, adjusting the optical access point cluster according to the weights of the transmission links between the accessible users and each target optical access point, and the further steps for obtaining the adjusted optical access point cluster include:

[0116] S201: Regarding the target optical access points in the intersection as the selected optical access points;

[0117] S202: According to the weights of the transmission links between multiple accessible users and the selected optical access points, determining the selected user uniquely corresponding to the selected optical access point, and the selected user cluster to which the selected user belongs;

[0118] S203: Making the selected optical access points belong to the optical access point cluster corresponding to the selected user cluster;

[0119] S204: According to the weights of the transmission links between the other users except the selected users among the multiple accessible users and each target optical access point, determining the other optical access points corresponding to the other users;

[0120] S205: Making the other optical access points belong to the optical access point cluster corresponding to the other user cluster, where the other user cluster is the cluster to which the other users belong.

[0121] Assume that target optical access point 1 is both the target optical access point corresponding to accessible user a, where accessible user a belongs to user cluster A, and the target optical access point corresponding to accessible user s, and accessible user s belongs to user cluster B. Then target optical access point 1 is the selected optical access point, and it is necessary to further determine the selected user uniquely corresponding to the selected optical access point 1 and the selected user cluster, where the selected user belongs to the accessible users. In this way, during the signal transmission process, the number of accessible users in the user cluster is not greater than the number of optical access points in its corresponding optical access point cluster, ensuring the communication quality and data transmission rate.

[0122] When determining the selected user uniquely corresponding to the selected optical access point, it is necessary to rely on the weights of the transmission links between multiple accessible users and the selected optical access point. Specifically, select the accessible user corresponding to the largest weight of the transmission link as the selected user, and the user cluster to which the selected user belongs is the selected user cluster. Since the weight of the transmission link is related to the number of samples and the physical distance, the larger the number of samples, the better the effect of the corresponding federated learning, and the smaller the physical distance, the better the communication effect. Therefore, using the weight of the transmission link as the criterion for selecting the user can ensure the efficiency of federated learning and the communication effect.

[0123] In this way, the selected optical access points belong to the optical access point cluster corresponding to the selected user cluster. In this way, the other users except the selected users need to re-determine the corresponding other optical access points.

[0124] For example, if the selected optical access point 1 belongs to the optical access point cluster M corresponding to the user cluster A, the accessible user s needs to re-determine the corresponding optical access point. The determination basis is also the weights of the transmission links between other users and each target optical access point.

[0125] Specifically, referring to Figure 3 , determining the other optical access points corresponding to the other users according to the weights of the transmission links between the other users (excluding the selected user) among the multiple accessible users and each target optical access point further includes:

[0126] S301: Obtain all target optical access points corresponding to the other users;

[0127] S302: Exclude the selected optical access point from all the target optical access points to obtain all the target optical access points after exclusion;

[0128] S303: Compare the weights of the transmission links between the other users and the target optical access points after exclusion, and use the target optical access point corresponding to the largest weight as the other optical access point corresponding to the other users.

[0129] For example, all the target optical access points corresponding to the accessible user s include: 1, 7, 9. Among them, the target optical access point 1 already belongs to the optical access point cluster M. At this time, one needs to be selected from the target optical access points 7 and 9 as the other optical access point corresponding to the accessible user s. The method is the same as above, that is, using the target optical access point corresponding to the largest weight of the transmission link as the corresponding other optical access point. Suppose the weight of the transmission link between the accessible user s and the target optical access point 7 is larger, then the other optical access point corresponding to the accessible user s is 7.

[0130] In this way, the adjustment of the optical access point cluster can be achieved.

[0131] Adjust the user cluster according to the set delay time, referring to Figure 4 , specifically:

[0132] S401: Calculate the data transmission time of the accessible users according to the uplink rate, uplink data volume, downlink rate, and downlink data volume between the accessible users in the user cluster and the federated learning server;

[0133] S402: When the data transmission time of the accessible users is greater than the set delay time, exclude the accessible users from the user cluster to obtain the adjusted user cluster.

[0134] The set delay time can be determined according to the actual working conditions and will not be elaborated here. The data transmission time of the accessible users can be calculated through the following formula:

[0135]

[0136] Among them, t is the data transmission time of the accessible user, and δ u is the downlink data volume of the accessible user, is the downlink rate of the accessible user, and δ G is the uplink data volume of the accessible user, is the uplink rate of the accessible user.

[0137] When the transmission time of the accessible user is too long, that is, greater than the set delay time, it means that the transmission efficiency of the accessible user is too low. To ensure the overall transmission efficiency, it can be removed.

[0138] Through the method of the embodiments of this article, first divide the accessible users into clusters, then obtain the optical access point clusters according to the user clusters. After dividing the users into clusters, by adjusting the optical access point clusters and the user clusters, the optical access points in the final optical access point clusters and the accessible users in the user clusters can adopt a vector transmission method for many-to-many communication, reducing the occurrence of communication interference and ensuring the communication quality. And the accessible users in the final user clusters are all users with higher transmission efficiency and better communication quality, improving the performance of federated learning.

[0139] In the embodiments of this article, the optical access point can be an LED lamp, and the specific parameters of the LED lamp can be as shown in Table 1 below:

[0140] Table 1

[0141]

[0142] For example, in a 15m×15m×3m room model indoors, a (4×4) optical access point with a height of 2.5m is evenly distributed to cover the optical communication downlink. The parameters of the optical access point (AP) array are shown in Table 1. Only considering line-of-sight propagation, since the field of view (FOV) of each user is limited, when one or more optical access points are within the user's FOV, the user can only receive information from the AP. When the incident angle ψ from the AP to the user is less than the user's FOV ψF, the total direct current (DC) attenuation of the optical channel is:

[0143]

[0144] Among them, m depends on the half-angle φ at half power 1 / 2 , m = -1 / log 2 (cosφ 1 / 2 ); D PA is the physical area of the light-emitting diode; r is the distance between the user and the optical access point; T s (ψ) and g(ψ) are the gains of the optical filter and the optical concentrator, g(ψ) = n2 / sin 2 ψ F ; n is the refractive index of the lens at the photodiode (PD).

[0145] To serve multiple users simultaneously and eliminate the inter-cell interference of multiple optical access points within a single user cluster, a zero-forcing (ZF)-based vector transmission technique is adopted.

[0146] Each user-centric (UC) cluster C m , consists of a set of APs with a cardinality of and a set of users with a cardinality of . constitutes.

[0147] and represent the vectors of the transmitted signal and the received signal respectively. On the premise of vector transmission, Y r = γ·P t ·H·G·Ω·X t +N, where γ and P t are the optical / electrical (O / E) conversion efficiency and the transmitted optical power respectively; N represents noise; the channel matrix represents the attenuation between user and AP . When the matrix G = H H ·(H·H H ) -1 meets the ZF criterion, an anti-interference identity matrix

[0148] is introduced. The matrix constrains the power of each AP, where G(i,:) is the i-th row of G. The signal transmission power constraint of the i-th AP can be obtained as Considering asymmetrically clipped optical OFDM (ACO-OFDM)

[0149] So To make each AP meet the power limit, it is required that

[0150] The signal-to-interference-plus-noise ratio (SINR) is defined as the ratio of the total electrical power within the bandwidth B to the sum of the noise power and the electrical power received by other nearby optical access points. Since the electrical power is proportional to the square of the current amplitude, and the interference between clusters and within clusters can be mitigated, the SINR within the user μ cluster C m is ​Among them, it is the interference caused by the reflected light. Considering that the interference power received by the cluster is affected by the ZF-based vector transmission within other clusters, assuming that the applied interference is always the maximum value, N 0 is the noise power spectral density N shot is the noise power spectral density of the optical signal, with an order of magnitude of 10 -22 , where q is the electron charge, I a (P r ) is the photocurrent at the user side.

[0151] As Figures 5a - 5d shown, it is compared with the traditional network-centric user selection algorithm, that is, the user is served by the optical access point that can provide the strongest signal. Among different FOV scenarios, that is, different degrees of interference, users are randomly distributed in the room and run independently 500 times, and the average results are taken. As the FOV increases, since the UC structure effectively reduces interference, the method proposed in the present invention can cover more users. For the traditional network-centric scheme, as the FOV increases from 90° to 110°, the number of connected users increases. However, when the FOV increases from 110° to 120°, due to the increase in the FOV resulting in a decrease in the received signal power and an increase in the received noise power, the number of connected users no longer increases.

[0152] As Figure 6 shown, as the number of connected users increases, the data available for FL training increases, and the recognition accuracy also increases. The same trend is also observed when the number of samples collected by users ranges from 300 to 900.

[0153] In summary, the method in the embodiments of this article fully considers the number of samples collected by each user and the latency requirements to construct user clusters. By applying the embodiments of this article, more users can be jointly used for FL training, reducing the complexity while improving the recognition accuracy.

[0154] Based on the above-mentioned user-centered federated learning method based on visible light communication, an embodiment of this article also provides a user-centered federated learning device based on visible light communication. The described device may include a system (including a distributed system), software (application), module, component, server, client, etc. that uses the method described in the embodiment of this article and combines the necessary implementation hardware. Based on the same innovative concept, the devices in one or more embodiments provided in the embodiments of this article are as described in the following embodiments. Since the implementation solutions for the device to solve problems are similar to the method, the implementation of the specific device in the embodiments of this article can refer to the implementation of the foregoing method, and the repeated parts will not be elaborated. As used hereinafter, the term "unit" or "module" may be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0155] Specifically, Figure 7 is a schematic diagram of the module structure of an embodiment of a user-centered federated learning device based on visible light communication provided in an embodiment of this article. Refer to Figure 7 As shown, in a user-centered federated learning device based on visible light communication provided in an embodiment of this article, a plurality of optical access points and a plurality of users are set within a regional range. The federated learning device includes: a target optical access point determination module 100, a weight determination module 200, an accessible user determination module 300, a user cluster division module 400, an optical access point determination module 500, an optical access point cluster adjustment module 600, and a user cluster adjustment module 700.

[0156] The target optical access point determination module 100 is configured to use the optical access points among all optical access points whose incident angle to the user is less than the field of view angle of the user as the target optical access points corresponding to the user;

[0157] The weight determination module 200 is configured to determine the weights of the transmission links between the user and each target optical access point according to the number of samples collected by the user and the distances between the user and each target optical access point;

[0158] The accessible user determination module 300 is configured to, if the total number of target optical access points is greater than or equal to the total number of users, all users are accessible users; if the total number of target optical access points is less than the total number of users, then sort all users according to the weights of the transmission links between the users and each target optical access point, and obtain the same number of users as the total number of target optical access points according to the sorting result as the accessible users;

[0159] The user cluster division module 400 is configured to divide all accessible users into multiple user clusters according to a preset physical distance;

[0160] An optical access point determination module 500 is configured to obtain an optical access point cluster corresponding to the user cluster according to the target optical access points corresponding to the accessible users in the user cluster;

[0161] An optical access point cluster adjustment module 600 is configured to, if there is an intersection between multiple optical access point clusters, adjust the optical access point cluster according to the weights of the transmission links between the accessible users and the respective target optical access points to obtain an adjusted optical access point cluster; wherein the target optical access points in the intersection correspond to multiple accessible users in multiple user clusters at the same time;

[0162] A user cluster adjustment module 700 is configured to adjust the user cluster according to a set delay time to obtain an adjusted user cluster, and perform federated learning based on the adjusted optical access point cluster and the adjusted user cluster.

[0163] Referring to Figure 8 As shown, based on the above-mentioned user-centered federated learning method based on visible light communication, in an embodiment of this article, a computer device 802 is further provided, where the above method runs on the computer device 802. The computer device 802 may include one or more processors 804, such as one or more central processing units (CPUs) or graphics processing units (GPUs), and each processing unit may implement one or more hardware threads. The computer device 802 may further include any memory 806, which is used to store any kind of information such as code, settings, data, etc. In a specific implementation manner, a computer program stored on the memory 806 and executable on the processor 804, when the computer program is run by the processor 804, may execute instructions according to the above method. Non-limitingly, for example, the memory 806 may include any one or a combination of the following: any type of RAM, any type of ROM, flash memory devices, hard disks, optical discs, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 802. In one case, when the processor 804 executes the associated instructions stored in any memory or combination of memories, the computer device 802 may perform any operation of the associated instructions. The computer device 802 further includes one or more drive mechanisms 808 for interacting with any memory, such as a hard disk drive mechanism, an optical disc drive mechanism, etc.

[0164] The computer device 802 may also include an input / output module 810 (I / O) for receiving various inputs (via the input device 812) and for providing various outputs (via the output device 814). A specific output mechanism may include a presentation device 816 and an associated graphical user interface 818 (GUI). In other embodiments, the input / output module 810 (I / O), the input device 812, and the output device 814 may not be included, and it may only be a computer device in the network. The computer device 802 may also include one or more network interfaces 820 for exchanging data with other devices via one or more communication links 822. One or more communication buses 824 couple the components described above together.

[0165] The communication link 822 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 822 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0166] Corresponding to Figures 1 - 6 In the method, embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the above method.

[0167] Embodiments of the present invention also provide a computer-readable instruction. When the processor executes the instruction, the program therein causes the processor to execute the method as Figures 1 to 6 shown.

[0168] It should be understood that in various embodiments of the present invention, the magnitudes of the serial numbers of the above processes do not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0169] It should also be understood that in the embodiments of the present invention, the term "and / or" is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present invention generally represents an "or" relationship between the associated objects before and after.

[0170] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this article.

[0171] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0172] In the several embodiments provided in this article, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can also be electrical, mechanical, or other forms of connection.

[0173] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments in this article.

[0174] In addition, the functional units in the various embodiments of this article can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0175] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution herein, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments herein. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0176] Specific embodiments are used in this article to elaborate on the principles and implementation manners of this article. The description of the above embodiments is only used to help understand the method and its core idea herein; at the same time, for those of ordinary skill in the art, according to the idea herein, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this article.

Claims

1. A user - centered federated learning method based on visible light communication, characterized in that, a plurality of optical access points and a plurality of users are set within a region, and the federated learning method includes: Regarding the optical access points among all optical access points whose incident angle to the user is less than the field - of - view angle of the user as the target optical access points corresponding to the user; Determining the weights of the transmission links between the user and each target optical access point according to the number of samples collected by the user and the distances between the user and each target optical access point; If the total number of target optical access points is greater than or equal to the total number of users, then all users are accessible users; If the total number of target optical access points is less than the total number of users, then sort all users according to the weights of the transmission links between the user and each target optical access point, and obtain the same number of users as the total number of target optical access points according to the sorting result as accessible users; Dividing all accessible users into multiple user clusters according to a preset physical distance; Obtaining the optical access point cluster corresponding to the user cluster according to the target optical access points corresponding to the accessible users in the user cluster; If there is an intersection between multiple optical access point clusters, then adjust the optical access point clusters according to the weights of the transmission links between the accessible users and each target optical access point to obtain the adjusted optical access point clusters; where the target optical access points in the intersection simultaneously correspond to multiple accessible users in multiple user clusters; Adjusting the user clusters according to a set delay time to obtain the adjusted user clusters, and performing federated learning based on the adjusted optical access point clusters and the adjusted user clusters.

2. The user - centered federated learning method based on visible light communication according to claim 1, characterized in that, The step of dividing all accessible users into multiple user clusters according to a preset physical distance further includes: S1: Randomly select any unlabeled accessible user as a user cluster, determine the center of the user cluster, and label the unlabeled accessible user; S2: Select another unlabeled accessible user, and determine whether the distance between the other unlabeled accessible user and the center of the user cluster is less than or equal to the preset physical distance; S3: If so, make the other unlabeled accessible user belong to the user cluster, label the other unlabeled accessible user, and update the center of the user cluster; S4: If not, loop through the above S2 to S4 until the distances between all unlabeled accessible users and the center of the user cluster are all greater than the preset physical distance; S5: Loop through the above S1 to S5 until all accessible users are labeled.

3. The user - centered federated learning method based on visible light communication according to claim 2, characterized in that, The step of determining the center of the user cluster further includes: Making the unlabeled accessible user be the center of the user cluster; The step of updating the center of the user cluster further includes: Using the other unlabeled accessible user to perform a mean process on the center of the user cluster to update the center of the user cluster.

4. The user-centered visible light communication-based federated learning method according to claim 1, wherein, the adjusting the optical access point cluster according to the weights of the transmission links between the accessible users and each target optical access point to obtain an adjusted optical access point cluster further includes: Regarding the target optical access points in the intersection as selected optical access points; Determining the selected user uniquely corresponding to the selected optical access point and the selected user cluster to which the selected user belongs according to the weights of the transmission links between multiple accessible users and the selected optical access point; Making the selected optical access point belong to the optical access point cluster corresponding to the selected user cluster; Determining the other optical access points corresponding to the other users according to the weights of the transmission links between the other users except the selected users among the multiple accessible users and each target optical access point; Making the other optical access points belong to the optical access point cluster corresponding to the other user cluster, where the other user cluster is the cluster to which the other users belong.

5. The user-centered visible light communication-based federated learning method according to claim 4, wherein, the determining the other optical access points corresponding to the other users according to the weights of the transmission links between the other users except the selected users among the multiple accessible users and each target optical access point further includes: Obtaining all the target optical access points corresponding to the other users; Removing the selected optical access points from all the target optical access points to obtain the remaining all target optical access points after removal; Comparing the weights of the transmission links between the other users and the remaining target optical access points after removal, and regarding the target optical access point corresponding to the largest weight as the other optical access point corresponding to the other users.

6. The user-centered visible light communication-based federated learning method according to claim 1, wherein, the adjusting the user cluster according to the set delay time to obtain an adjusted user cluster further includes: Calculating the data transmission time of the accessible users according to the uplink rate, uplink data volume, downlink rate, and downlink data volume between the accessible users in the user cluster and the federated learning server; When the data transmission time of the accessible users is greater than the set delay time, removing the accessible users from the user cluster to obtain an adjusted user cluster.

7. The user-centered visible light communication-based federated learning method according to claim 6, wherein, the calculating the data transmission time of the accessible users according to the uplink rate, uplink data volume, downlink rate, and downlink data volume between the accessible users in the user cluster and the federated learning server further includes: Calculating the data transmission time of the accessible users through the following formula: Among them, t is the data transmission time of the accessible user, and δ u is the downlink data volume of the accessible user, is the downlink rate of the accessible user, and δ G is the uplink data volume of the accessible user, is the uplink rate of the accessible user.

8. A user-centered visible light communication-based federated learning device, wherein, A plurality of optical access points and a plurality of users are arranged within the regional range, and the federated learning device includes: A target optical access point determination module, configured to use, as the target optical access point corresponding to the user, the optical access points among all optical access points whose incident angle to the user is less than the field of view angle of the user; A weight determination module, configured to determine the weights of the transmission links between the user and each target optical access point according to the number of samples collected by the user and the distances between the user and each target optical access point; An accessible user determination module, configured to: if the total number of target optical access points is greater than or equal to the total number of users, then all users are accessible users; if the total number of target optical access points is less than the total number of users, then sort all users according to the weights of the transmission links between the user and each target optical access point, and obtain, as the accessible users, the same number of users as the total number of target optical access points according to the sorting result; A user cluster division module, configured to divide all accessible users into multiple user clusters according to a preset physical distance; An optical access point determination module, configured to obtain the optical access point cluster corresponding to the user cluster according to the target optical access points corresponding to the accessible users in the user cluster; An optical access point cluster adjustment module, configured to: if there is an intersection between multiple optical access point clusters, then adjust the optical access point clusters according to the weights of the transmission links between the accessible users and each target optical access point to obtain the adjusted optical access point clusters; wherein the target optical access points in the intersection correspond to multiple accessible users in multiple user clusters at the same time; A user cluster adjustment module, configured to adjust the user clusters according to a set delay time to obtain the adjusted user clusters, and perform federated learning based on the adjusted optical access point clusters and the adjusted user clusters.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory, wherein, when the computer program is run by the processor, it executes the instructions of the method according to any one of claims 1-7.

10. A computer-readable storage medium, on which a computer program is stored, wherein, when the computer program is run by the processor of the computer device, it executes the instructions of the method according to any one of claims 1-7.

Images