A Federated Learning Method and System Based on RSSI Principle
By using RSSI localization and centroid algorithms to aggregate model parameters in federated learning, the problem of decreased model accuracy caused by non-independent and identically distributed data is solved. This achieves improved model training accuracy and recognition precision without affecting performance and is applicable to various data distributions.
Patent Information
- Application Number
- CN202211665585.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-23
- Publication Date
- 2025-10-28
- Estimated Expiration
- 2042-12-23
AI Technical Summary
Existing federated learning methods suffer from poor model accuracy and performance when faced with non-independent and identically distributed terminal data, and data transmission also poses privacy risks and additional costs.
The RSSI localization algorithm and the centroid algorithm are used to aggregate model parameters. The aggregated global model parameters are determined by anchor node coordinates and the centroid algorithm, which improves the accuracy and adaptability of model training.
Without affecting the model's time complexity and performance, it improves the model's training accuracy and recognition precision, is applicable to various data distributions, and reduces the risk of privacy leaks during data transmission.
Smart Images
Figure CN116090582B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machine learning, and in particular to a federated learning method and system based on the RSSI principle. Background Technology
[0002] In recent years, tablets, smartphones, and smartwatches have become increasingly popular and indispensable parts of people's lives. These devices often store various types of data from multiple sensors, such as images, voice, personal motion, and health status. These mobile devices use this local data to train machine learning models to provide users with richer and more convenient functions (image recognition, voice recognition, etc.). However, the data upload process often involves issues such as privacy leaks, uneven data quality, and additional data transmission costs. To address these problems, Google proposed Federated Learning in 2016. Federated Learning aims to solve the privacy leaks and uneven data quality issues that may exist in traditional machine learning, and it has attracted widespread attention from scholars both domestically and internationally. Federated Learning is a distributed machine learning framework. Compared to traditional machine learning frameworks, Federated Learning does not require terminals to upload data to a central server. Each terminal only needs to use the initial model transmitted from the central server and its local data for training. After training, the model parameters are uploaded to the server, where parameter fusion is performed to update the global model. The server then transmits the updated model to each terminal to begin the next round of training. This new machine learning training method largely avoids the privacy leaks that may occur when terminals upload data to the server. However, federated learning suffers from inconsistent data quality and non-independent, identically distributed data due to different data sources. These factors can lead to a decrease in the accuracy of the integrated model and affect its performance.
[0003] Therefore, there is an urgent need to propose a new federated learning method or system that can improve the accuracy of the model with almost no impact on the overall model performance; and is applicable to various data distributions, with good universality. Summary of the Invention
[0004] The purpose of this invention is to provide a federated learning method and system based on the RSSI principle, which can improve the accuracy of training models with almost no impact on model time complexity, performance and efficiency, thereby improving the accuracy of data recognition.
[0005] To achieve the above object, the present invention provides the following solutions:
[0006] A federated learning method based on the RSSI principle includes:
[0007] The server determines the initial recognition model and training parameters; the training parameters include: the number of training rounds and the target accuracy of the model after training.
[0008] Each terminal uses the initial recognition model obtained from the server and trains the initial recognition model based on the corresponding data, updates the parameters of the recognition model, and uploads the updated parameters of the recognition model to the server.
[0009] The server uses RSSI positioning algorithm and centroid algorithm to aggregate the parameters of the updated model, and uses the aggregated model parameters to update the global model. The updated global model is then transmitted to each terminal. The server then returns the initial recognition model obtained from the server to each terminal, and repeats this process of training the initial recognition model based on the corresponding data, updating the parameters of the recognition model, and uploading the updated parameters of the recognition model to the server until training is completed and the final global model is determined.
[0010] Each terminal uses the final recognition model to identify the corresponding data.
[0011] Optionally, the server determines the initial model and training parameters, specifically including:
[0012] The server determines the target accuracy of the trained model and the parameters of the initial model based on the test dataset.
[0013] Optionally, the test dataset includes: the MNIST image recognition dataset or the CIFAR-10 image recognition dataset.
[0014] Optionally, the server uses the RSSI positioning algorithm and the centroid algorithm to aggregate the parameters of the updated model, updates the recognition model with the aggregated recognition model parameters, and transmits the updated model to each terminal, specifically including:
[0015] The parameters of the updated model determined by each terminal are used as the anchor node coordinates in the RSSI algorithm, and the aggregated global model parameters are determined by combining the centroid algorithm.
[0016] The aggregated recognition model parameters are used to update the global model, and the updated global model is then transmitted to each terminal.
[0017] A federated learning system based on RSSI principles includes:
[0018] An initialization module is used by the server to determine the initial model and training parameters; the training parameters include: the number of training epochs and the target accuracy of the model after training.
[0019] The model parameter update module is used by each terminal to update the parameters of the initial model obtained from the server and trained based on the corresponding data, and then upload the updated model parameters to the server.
[0020] The model update module is used by the server to aggregate the parameters of the updated model using the RSSI localization algorithm and the centroid algorithm, update the global model with the aggregated model parameters, and transmit the updated global model to each terminal; and return to the model parameter update module until training is completed and the final global model is determined.
[0021] The terminal data recognition module is used by each terminal to recognize the corresponding data using the final recognition model.
[0022] A federated learning system based on RSSI principles includes: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein the computer program instructions are executed by the processor to implement the federated learning method based on RSSI principles.
[0023] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0024] This invention provides a federated learning method and system based on RSSI principles. The server uses RSSI localization and centroid algorithms to aggregate the parameters of the updated model, updates the global model with the aggregated parameters, and transmits the updated global model to each terminal. This novel parameter aggregation method allows data from any terminal to participate in training regardless of its distribution. The accuracy of the trained model is used to differentiate between different models, effectively reducing the impact of non-independent and identically distributed (IMSD) data on the final model's accuracy and performance, enabling faster convergence within a limited number of training epochs. This addresses the problem of traditional federated learning simply averaging the trained model parameters without considering the data type and accuracy of each terminal. This invention improves the accuracy of the trained model with minimal impact on time complexity, performance, and efficiency. This invention is applicable to various data types, and compared to ordinary arithmetic averaging federated learning algorithms, it achieves higher accuracy for both IMSD and non-IMSD data. Attached Figure Description
[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 This is a schematic diagram of a federated learning method based on the RSSI principle provided by the present invention. Detailed Implementation
[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0028] The purpose of this invention is to provide a federated learning method and system based on the RSSI principle, which can improve the accuracy of training models with almost no impact on model time complexity, performance and efficiency, thereby improving the accuracy of data recognition.
[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0030] like Figure 1 As shown, the federated learning method based on the RSSI principle provided by this invention includes:
[0031] S101, The server determines the initial recognition model and training parameters; the training parameters include: the number of training rounds and the target accuracy of the model after training.
[0032] S101 specifically includes:
[0033] The server determines the target accuracy of the trained model and the parameters of the initial recognition model based on the test dataset. The test dataset includes either the MNIST image recognition dataset or the CIFAR-10 image recognition dataset.
[0034] S102, each terminal obtains the initial model from the server, trains the initial model based on the corresponding data, updates the model parameters, and uploads the updated model parameters to the server.
[0035] S103, the server uses the RSSI positioning algorithm and the centroid algorithm to aggregate the parameters of the updated model, updates the global model with the aggregated recognition model parameters, and transmits the updated global model to each terminal; and returns to S102, until training is completed and the final model is determined;
[0036] After the model is downloaded from the server to the terminal, training begins. Due to the different data characteristics of each terminal, the accuracy after training varies. The accuracy is fitted to the signal attenuation amplitude in the RSSI algorithm, and the parameter vector is fitted to the coordinates in the RSSI algorithm. The trained model parameters are fitted to the final result of the RSSI localization algorithm. Because of signal attenuation, the precise center coordinates cannot usually be obtained well through the localization algorithm. Instead, the Euclidean distance between each coordinate is calculated, and the final result, i.e., the aggregated recognition model parameters, is obtained through the centroid algorithm.
[0037] S103 specifically includes:
[0038] The parameters of the updated recognition model determined by each terminal are used as the anchor node coordinates in the RSSI algorithm, and the aggregated recognition model parameters are determined by combining the centroid algorithm.
[0039] The aggregated recognition model parameters are used to update the recognition model, and the updated recognition model is then transmitted to each terminal.
[0040] The following specific examples illustrate the training process, which requires the following parameters: number of local training iterations E, calculation threshold M, average accuracy aug, and test accuracy test. acc And accuracy acc. Here, M represents the target training accuracy; avg represents the average accuracy across all clients after each training iteration; test acc This indicates the final accuracy of the training and testing sets.
[0041] In the RSSI positioning algorithm, wireless signals typically attenuate gradually with increasing transmission distance. This method uses the training accuracy of the terminal in each round to fit the signal attenuation amplitude, and the parameter vector serves as the anchor point in the positioning algorithm. From this, the path loss formula can be derived as follows:
[0042]
[0043] Where PL(d0) represents the RSSI value when d0 = 1m, k represents the path loss coefficient, and γ represents the degree of signal loss as distance increases; γ is a variable that follows a Gaussian random distribution. Given nodes A, B, and C, and an unknown node D, the distance between nodes A and D calculated using the RSSI model is r. a The distance between nodes B and D is r. b The distance between nodes C and D is r. c With A, B, and C as centers respectively; r a r b r c Drawing a circle with a radius yields the overlapping region. The basic idea of the centroid algorithm is to calculate the coordinates of three feature points in the overlapping region of the three circles. The method for calculating the coordinates of the feature points is as follows:
[0044]
[0045] With these three points as the vertices of the triangle, the unknown point is the centroid of the triangle.
[0046] The specific aggregation process is as follows:
[0047] The server initializes the recognition model w0 and sends the recognition model to each terminal.
[0048] After receiving the recognition model from the server, terminal i trains the recognition model using its own local data.
[0049] After terminal i completes this round of training, the training results will be... The data is uploaded to the server. The server calculates the anchor point coordinates based on (1) and the distance to the final result. It then calculates the model parameters for the current round using the centroid algorithm.
[0050] The server will The updated model parameters are transmitted to each training terminal for the next round of training.
[0051] Training stops when the model accuracy no longer significantly improves and gradually converges to within the computational threshold M, and the final model is obtained.
[0052] S104, each terminal uses the final recognition model to recognize the corresponding data.
[0053] Corresponding to the above method, the present invention provides a federated learning system based on the RSSI principle, comprising:
[0054] An initial module is used by the server to determine the initial recognition model and training parameters; the training parameters include: the number of training rounds and the target accuracy of the model after training.
[0055] The model parameter update module is used by each terminal to update the parameters of the initial recognition model based on the initial recognition model obtained from the server and the corresponding data, and then upload the updated parameters of the recognition model to the server.
[0056] The model update module is used by the server to aggregate the parameters of the updated model using the RSSI localization algorithm and the centroid algorithm, update the global model using the aggregated model parameters, transmit the updated global model to each terminal, and return to the model parameter update module until training is completed and the final model is determined.
[0057] The terminal data recognition module is used by each terminal to recognize the corresponding data using the final recognition model.
[0058] In order to execute the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, the present invention provides a federated learning system based on the RSSI principle, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein when the computer program instructions are executed by the processor, the federated learning method based on the RSSI principle is implemented.
[0059] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0060] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A federated learning method based on the RSSI principle, characterized in that, include: The server determines the initial model and training parameters; The training parameters include: the number of training rounds and the target accuracy of the model after training; Each terminal uses the initial model obtained from the server and trains the initial model based on the corresponding data, updates the model's parameters, and uploads the updated model's parameters to the server. The server uses the RSSI positioning algorithm and the centroid algorithm to aggregate the parameters of the updated model, updates the global model with the aggregated model parameters, and transmits the updated global model to each terminal. It then returns to each terminal the steps of training the initial model based on the initial model obtained from the server and the corresponding data, updating the model parameters, and uploading the updated model parameters to the server, until the model converges and training is completed, and the final global model is determined. Each terminal uses the final recognition model to identify the corresponding data; The server determines the initial model and training parameters, specifically including: The server determines the target accuracy of the trained model and the parameters of the initial model based on the test dataset. The test dataset includes: the MNIST image recognition dataset or the CIFAR-10 image recognition dataset; The server uses the RSSI positioning algorithm and the centroid algorithm to aggregate the parameters of the updated model, updates the global model with the aggregated model parameters, and transmits the updated global model to each terminal. Specifically, this includes: The parameters of the updated recognition model determined by each terminal are used as the anchor node coordinates in the RSSI algorithm, and the aggregated recognition model parameters are determined by combining the centroid algorithm. The aggregated recognition model parameters are used to update the recognition model, and the updated recognition model is transmitted to each terminal. Training parameters: number of local training iterations (E), calculation threshold (M), average accuracy (avg), and test accuracy (test). acc And accuracy acc; where the calculation threshold M represents the target training accuracy; average accuracy avg represents the average accuracy of all clients after each training iteration; test accuracy test acc This represents the final accuracy on the training and testing sets. In the RSSI positioning algorithm, wireless signals typically attenuate gradually with increasing transmission distance. The signal attenuation amplitude is fitted using the training accuracy of the terminal in each round, and the parameter vector serves as the anchor point in the positioning algorithm. From this, the path loss formula can be derived as follows: Where PL(d0) represents the RSSI value when d0 = 1m, k represents the path loss coefficient, and γ represents the degree of signal loss as distance increases, which is a variable following a Gaussian random distribution; the distance between node A and the unknown node D calculated according to the RSSI model is r. a The distance between node B and the unknown node D is r. b The distance between node C and the unknown node D is r. b With nodes A, B, and C as centers respectively; r a r b r b Draw a circle with radius to obtain the overlapping region; the centroid algorithm calculates the coordinates of three feature points in the overlapping region of the three circles, takes these three feature points as the vertices of the triangle, and the unknown node as the centroid of the triangle; The specific aggregation process is as follows: The server initializes the recognition model w0 and sends the recognition model w0 to each terminal; After receiving the recognition model w0 sent by the server, terminal i trains the recognition model with its own local data; After terminal i completes this round of training, the training results will be... The data is uploaded to the server. The server calculates the anchor point coordinates based on the path loss formula and the distance to the final result. It then uses the centroid algorithm to calculate the model parameters for the current round. The server will The updated model parameters are transmitted to each training terminal for the next round of training; Training stops when the accuracy of the recognition model no longer improves significantly and gradually converges to within the calculation threshold M, and the final recognition model is obtained.
2. A federated learning system based on the RSSI principle, used to implement the federated learning method based on the RSSI principle as described in claim 1, characterized in that, include: The initialization module is used by the server to determine the initial model and training parameters; The training parameters include: the number of training rounds and the target accuracy of the model after training; The model parameter update module is used by each terminal to update the parameters of the initial model obtained from the server and trained based on the corresponding data, and then upload the updated model parameters to the server. The model update module is used by the server to aggregate the parameters of the updated model using the RSSI localization algorithm and the centroid algorithm, update the global model with the aggregated model parameters, and transmit the updated global model to each terminal; and return to the model parameter update module until the training parameters are completed and the final global model is determined. The terminal data recognition module is used by each terminal to recognize the corresponding data using the final recognition model.
3. A federated learning system based on the RSSI principle, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, wherein when the computer program instructions are executed by the processor, a federated learning method based on the RSSI principle as described in claim 1 is implemented.
Citation Information
Patent Citations
Federal learning-based human activity identification method and system
CN114386621A
Federal learning indoor positioning method and device based on convex hull area weighting
CN115344739A