Gatekeeper communication efficient one-time federated ensemble learning method and system

Abstract

The invention discloses a gatekeeper communication efficient one-time federated ensemble learning method and system, and belongs to the technical field of data security, and the method comprises the steps: carrying out the anonymization synthesis of local original data of a participant based on a generative adversarial network, and generating a synthetic data set; transmitting the synthesized data set to a collector at one time through a security isolation gatekeeper; and combining the synthesized data set with local data to form a complete data set, performing preprocessing and training set division on the complete data set, performing centralized model training based on the complete data set, adopting an iterative optimization method until the model is converged, and issuing the trained model to each participant. The method is based on the generative adversarial network, communication delay is reduced through a multi-body cross-gatekeeper efficient one-time transmission technology, centralized safe cross-domain modeling of a single participant is achieved, the method fits the current situation of existing network equipment, the requirement for data interaction is low, and high efficiency and usability of the model are achieved on the premise that user information safety is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of data security technology, and in particular relates to a one-time federated integrated learning method and system with efficient gatekeeper communication. Background Art

[0002] Federated learning, as a privacy-preserving technology, enables certain sensitive industries with high security requirements to implement joint modeling without requiring data to leave their local machine. This not only protects privacy but also effectively alleviates the data silo problem. However, traditional federated learning completes model training through multiple rounds of parameter exchange, but this requires frequent communication between participating parties, resulting in high computational and communication overhead. Especially in cross-domain collaborative scenarios, network isolation requirements necessitate data exchange via physically isolated devices. Traditionally, each data transfer must undergo multiple data flows. When participating parties conduct cross-domain collaborative modeling, data ferrying across network gatekeepers can take minutes. The physical disconnection of these gatekeepers results in high latency in cross-domain data exchange, limiting the efficiency of end-to-end cross-domain interactions in privacy-conscious computing and making it difficult to meet the real-world demand for high-speed modeling. Therefore, how to minimize communication costs while ensuring high model accuracy while reducing the number of interactions to just one has become a key challenge in this field.

[0003] While existing technologies can improve federated learning efficiency to a certain extent by compressing communication data or reducing the number of iterations, they still cannot circumvent the inherent latency bottleneck of gatekeeper transmission. While lightweight model parameter reduction can reduce the amount of data transmitted per pass, the cumulative time required for multiple rounds of interaction remains significant. Single-pass federated learning, while reducing the number of communications to just one, can easily lead to low-quality synthesized data and insufficient privacy protection. Therefore, in summary, current federated learning data processing cannot achieve efficient federated modeling through a single cross-gatekeeper communication while ensuring data privacy and security, while also ensuring consistency and accuracy between the synthesized data and the original data. Summary of the Invention

[0004] The present invention provides a one-time federated integrated learning method and system with efficient network-gateway communication, aiming to solve the problem in current federated learning data processing that it is impossible to achieve efficient federated modeling through a single cross-network-gateway communication while ensuring data privacy and security, while ensuring the consistency and accuracy of the synthesized data with the original data.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: The present invention provides a one-step federated ensemble learning method for efficient gatekeeper communication, comprising the following steps: S1. Based on a generative adversarial network consisting of a generator and a discriminator, the local raw data of several participants are anonymized and synthesized to generate a synthetic dataset that is consistent with the structure and statistical characteristics of the original data; S2, transmit the synthetic data sets of each participant to the only collection party at one time through the secure isolation network; The secure isolation gateway includes an intranet host system, an extranet host system, and an isolation switch matrix. The isolation switch matrix implements secure caching and exchange of data through an intranet isolation switch module and an extranet dual isolation switch module. S3. The synthetic datasets of the collector and each participant are merged with the local data to form a complete dataset. After preprocessing and dividing the complete dataset into training sets, centralized model training is performed based on the complete dataset. An iterative optimization method is used until the model converges. The trained model is distributed to each participant to complete a federated integrated learning.

[0006] In some embodiments, in S1, each participant preprocesses the local original data, divides the preprocessed data into multiple batches and marks them as real data to input into the discriminator; at the same time, the generator generates false data based on the noise vector and the conditional vector and marks them as false data, and inputs them into the discriminator.

[0007] In some embodiments, in S1, the training loss of the generator includes: the cross entropy loss between the conditional vector and the generated data label, the discrimination loss of the discriminator output, the classification loss of the classifier output, and the sum of the statistical feature difference loss between the real data and the fake data.

[0008] In some embodiments, in S1, the generative adversarial network is a trained generative adversarial network, and the generative adversarial network training adopts a differentially private stochastic gradient descent method, which includes clipping the gradient and adding Gaussian noise.

[0009] In some implementations, in S2, the external network isolation switching module encapsulates the synthesized data into a private protocol data packet through the security isolation chip and caches it in the switching subsystem.

[0010] Furthermore, in S2, the cache and switching of the isolation switching matrix specifically include: After the external network isolation switching module is disconnected from the external network, the cached data is migrated to the switching subsystem of the internal network isolation switching module.

[0011] Furthermore, in S2, the intranet isolation switching module reads and parses the private protocol data packet through the security isolation chip, and restores the private protocol data packet to the original data block.

[0012] In some embodiments, in S3, the complete data set is obtained by splicing the synthetic data of each participant and the local data of the collector.

[0013] In some embodiments, in S3, preprocessing of the complete dataset includes data cleaning and feature normalization, and the iterative optimization method adopts a stochastic gradient descent algorithm.

[0014] The present invention also provides a one-time federated integrated learning system with efficient network gate communication, the system including a data synthesis module, a data exchange and transmission module, and a model training and distribution module, wherein: Data synthesis module: used to anonymize and synthesize the local raw data of several participants based on a generative adversarial network consisting of a generator and a discriminator, and generate a synthetic dataset that is consistent with the structure and statistical characteristics of the original data; Data exchange and transmission module: used to transmit the synthetic data sets of each participant to the only collection party at one time through a secure isolation network; The secure isolation gateway includes an intranet host system, an extranet host system, and an isolation switch matrix. The isolation switch matrix implements secure caching and exchange of data through an intranet isolation switch module and an extranet dual isolation switch module. Model training and distribution module: used to merge the synthetic data sets of the collector and each participant with the local data to form a complete data set. After preprocessing and dividing the complete data set into training sets, centralized model training is performed based on the complete data set. An iterative optimization method is used until the model converges. The trained model is distributed to each participant to complete a federated integrated learning.

[0015] Compared with the prior art, the present invention provides a method and system for efficient one-time federated integrated learning of gateway communication, which has the following beneficial effects: The present invention provides a one-time federated integrated learning method with efficient network gate communication, which realizes one-time data transmission through a secure isolation network gate, avoiding the high latency problem of multiple iterative communications in traditional federated learning. Furthermore, the isolation exchange module and private protocol encapsulation technology of the network gate optimize the data ferry process, so that the communication time consumption of cross-domain collaborative modeling is reduced to a minimum. Each participant uses GAN to generate false data similar to the original data structure and statistical characteristics to ensure that the original data does not need to be directly shared. Furthermore, combined with the differential privacy stochastic gradient descent technology, noise and gradient clipping are added during the model training process, doubly guaranteeing data privacy security. The present invention fits the status quo of existing network equipment and supports sensitive industries with high security to achieve joint modeling without data leaving the local area. Furthermore, the present invention meets the strict requirements of data compliance through the single transmission mechanism of the network gate, while breaking the data island problem. The present invention ensures the accuracy of the model by integrating virtual data synthesized by multiple parties and the collector's own data to centrally train the model. Furthermore, conditional vectors are used to handle the problem of unbalanced data distribution, which improves the adaptability and generalization ability of the model in complex scenarios. In addition, the present invention supports the dynamic increase or decrease of the number of participants, adapts to the needs of federated learning of different scales, optimizes the training process through differential privacy and gradient clipping technology, significantly reduces the consumption of computing resources, and is suitable for business scenarios with high-frequency data updates. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings in the specification are used to provide further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0017] Figure 1 Schematic diagram of a flow chart of a federated ensemble learning method for efficient gatekeeper communication according to the present invention; Figure 2 Schematic diagram of the flow of the GAN data synthesis algorithm in the one-step federated ensemble learning method with efficient gatekeeper communication of the present invention; Figure 3 The present invention provides a flow chart of a unilateral centralized security model training algorithm in a one-time federated ensemble learning method for efficient gatekeeper communication. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0019] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0020] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0021] In the description of the embodiments of the present invention, it should be noted that if the terms "upper," "lower," "horizontal," "inner," etc. appear, the orientation or positional relationship indicated is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the inventive product is typically placed when in use. These terms are merely for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first," "second," etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0023] In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0024] How to propose a one-shot federated learning (Federated Learning) collaborative optimization for the transmission characteristics of network gates, reducing the number of interactions to one to minimize communication costs while ensuring high model accuracy.

[0025] like Figure 1 As shown, the present invention provides a federated ensemble learning method with high efficiency for gatekeeper communication, comprising the following steps: S1. Based on a generative adversarial network consisting of a generator and a discriminator, the local raw data of several participants are anonymized and synthesized to generate a synthetic dataset that is consistent with the structure and statistical characteristics of the original data; S2, transmit the synthetic data sets of each participant to the only collection party at one time through the secure isolation network; The secure isolation gateway includes an intranet host system, an extranet host system, and an isolation switch matrix. The isolation switch matrix implements secure caching and exchange of data through an intranet isolation switch module and an extranet dual isolation switch module. S3. The synthetic datasets of the collector and each participant are merged with the local data to form a complete dataset. After preprocessing and dividing the complete dataset into training sets, centralized model training is performed based on the complete dataset. An iterative optimization method is used until the model converges. The trained model is distributed to each participant to complete a federated integrated learning.

[0026] The present invention provides an efficient one-shot federated ensemble learning method for gatekeeper communication. Each participant synthesizes virtual data with a consistent structure of the original data based on a data synthesis algorithm, achieving data anonymization. All participants then transmit the synthesized virtual data to a collector via a gatekeeper, completing a one-shot transmission across multiple domains. Finally, the collector merges all collected virtual datasets with its own data into a complete dataset and performs centralized security model training based on this dataset. The present invention aims to provide an efficient one-shot federated ensemble learning method for gatekeeper communication while meeting data security requirements. This method addresses the issues of traditional gatekeepers consuming long data transfer times, high cross-domain data exchange latency due to their physical disconnection, and low end-to-end cross-domain interaction efficiency for privacy-focused computing, which hinders the real-world demand for high-real-time modeling. The present invention completes federated modeling tasks with a single data exchange, supports anonymization of original data using a Generative Adversarial Network (GAN), and reduces communication latency through efficient one-shot transmission across multiple agents across the gatekeeper, ultimately enabling centralized, secure cross-domain modeling for a single participant. The present invention is in line with the status quo of existing network equipment, and the requirements for data interaction can be reduced to a relatively low level. It can achieve the efficiency and usability of the model while ensuring the security of user information.

[0027] In some embodiments, the present invention employs a highly efficient federated ensemble learning method for gatekeeper communication. Through adversarial training of a generator and a discriminator, a synthetic dataset with consistent statistical characteristics of the original data is generated, minimizing leakage of the original data. Gradient clipping and Gaussian noise addition during training provide dual protection for data privacy, preventing the inference of model parameters from the original data. The present invention utilizes proprietary protocol encapsulation, an internal and external network isolation switching module, and secure chip parsing to ensure that data cannot be intercepted or tampered with during transmission.

[0028] Furthermore, this invention achieves a one-time data migration for all participants, avoiding the multiple rounds of iterative communication required by traditional federated learning and significantly reducing the cumulative latency of cross-domain transmission. The external network module caches data, disconnects, and then migrates it to the internal network module, improving transmission efficiency.

[0029] In some embodiments, the generator of the present invention's efficient one-time federated ensemble learning method for network gateway communication integrates cross entropy loss, discrimination loss, classification loss, and statistical difference loss to ensure that the synthesized data is highly consistent with the original data; the collector merges multi-party synthesized data with local data, and performs preprocessing such as data cleaning and feature normalization to improve the quality of the training set and ensure model accuracy and generalization ability.

[0030] Furthermore, this invention does not limit the number of participants, adapting to federated learning scenarios of varying scales. The secure isolation gateway's modular components, such as proprietary protocol encapsulation and a switching subsystem, are compatible with existing network equipment, reducing deployment costs. Its data anonymization and single-transmission mechanisms meet the compliance and confidentiality requirements of security-sensitive industries, demonstrating its adaptability.

[0031] The present invention also provides a one-time federated integrated learning system with efficient network gate communication, the system including a data synthesis module, a data exchange and transmission module, and a model training and distribution module, wherein: Data synthesis module: used to anonymize and synthesize the local raw data of several participants based on a generative adversarial network consisting of a generator and a discriminator, and generate a synthetic dataset that is consistent with the structure and statistical characteristics of the original data; Data exchange and transmission module: used to transmit the synthetic data sets of each participant to the only collection party at one time through a secure isolation network; The secure isolation gateway includes an intranet host system, an extranet host system, and an isolation switch matrix. The isolation switch matrix implements secure caching and exchange of data through an intranet isolation switch module and an extranet dual isolation switch module. Model training and distribution module: used to merge the synthetic data sets of the collector and each participant with the local data to form a complete data set. After preprocessing and dividing the complete data set into training sets, centralized model training is performed based on the complete data set. An iterative optimization method is used until the model converges. The trained model is distributed to each participant to complete a federated integrated learning.

[0032] The following further describes in detail a method and system for efficient one-time federated integrated learning of gateway communication of the present invention through specific embodiments.

[0033] like Figure 1-Figure 3 As shown in FIG, the present invention uses a federated ensemble learning method. First, each participant synthesizes a virtual data structure consistent with the original data based on the data synthesis algorithm of the generative adversarial network to achieve data anonymization. Second, all participants transmit the synthesized virtual data to the collector through the gateway to complete the one-time transmission of multiple domains. Finally, the collector merges all the collected virtual data sets and the data it owns into a complete data set, and performs corresponding centralized security model training based on this data set. The process of the generative adversarial network data synthesis algorithm is as follows: Figure 2 As shown, the process of the unilateral centralized security model training algorithm is as follows Figure 3 The method of the present invention specifically comprises the following steps: Step 1: Execute the anonymization synthesis phase of the original data of the participants. Participants, denoted as , and a collector Each party has the same features, a total of m, and the number of samples for each party is Each participant based on all data sets Train the GAN network to generate fake data that is similar to its data structure and statistical characteristics ; Step 2: Each participant will generate the dataset Each of them is transmitted to the collection party through a security isolation network at one time. The security isolation network gate is composed of an intranet host system, an extranet host system and an isolation switching matrix. The isolation switching matrix is ​​based on dual isolation switching modules for the intranet and extranet. The isolation switching module contains a security isolation chip and a switching chip. The security isolation chip converts data blocks into data packets in its own protocol format. The switching subsystem and switch control subsystem in the switching chip realize temporary caching and secure exchange of data. Step 3: By integrating data from all parties as all data , and based on this, a model is trained locally and securely. After the training is completed, the model results are sent to all participants.

[0034] Specifically, in the above step 1, the present invention is for any participant First, the data is preprocessed, and then Randomly divided into batches , the identification labels are all 1 (real data), and input into the discriminator middle.

[0035] In step 1, corresponding to batches, uniformly sampled from the hypercube Noise vectors, and conditional vectors are added to deal with the imbalanced distribution problem as generators Input. Finally, a batch of false data is obtained , whose identification labels are all 0 (false data), and are also input into The cross entropy between the conditional vector and the generated output class is the loss of the generator, denoted as .

[0036] In step 1, Get two types of positive and negative samples to train D, where the discrimination loss is recorded as False data Also input to the classifier The predicted dataset labels are recorded as .

[0037] Furthermore, in step 1, based on , , and the difference loss of statistical features between true and false data The model is trained by summing the four types of losses. The model is trained using differentially private stochastic gradient descent for parameter optimization. This means that after calculating the gradient, the gradient is clipped according to the clipping parameter c and Gaussian noise is added.

[0038] In addition, when the loss sum reaches convergence, the model stops iterating and outputs the final synthetic dataset .

[0039] In step 2 of this embodiment, the switch control subsystem of the external network (each participant's network domain) isolation switch module first connects to its security isolation chip. Each participant's host encapsulates the synthesized data into a data packet according to a proprietary protocol format through the security isolation chip and transfers it to the switching subsystem cache. Furthermore, in step 2, the switch control subsystem disconnects from the external network, connects the internal and external network isolation switching modules, and transfers the data of the switching subsystem in the external network isolation switching module to the switching subsystem in the internal network isolation switching module; The switch control subsystem of the intranet (collector's domain) isolation switching module disconnects from the external network isolation switching module, then connects to the local security isolation chip, reads out the cached data in the switching subsystem, and unpacks the private protocol format into data blocks.

[0040] In step 3 of this embodiment, data from all parties are merged. The synthetic data collected from k participants and unrestrained data Splice together to form a complete data set .right Perform data preprocessing and divide the training set into training set and training set.

[0041] Furthermore, for a certain model, the above data is used for training until the model converges. The trained model is sent to .

[0042] In summary, the present invention provides an efficient one-time federated integrated learning method and system for network-gateway communication, which achieves a balance between privacy protection and data utility through GAN synthetic data and differential privacy technology; single transmission and network-gateway optimize the efficiency of data transmission; from data generation and transmission to training, it embeds security mechanisms of isolation exchange and protocol encapsulation to form a multi-level protection system; and only relies on existing network equipment, lowering the implementation threshold; supports dynamic participants and has strong adaptability.

[0043] Finally, it should be noted that the above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any ordinary technician in this industry can smoothly implement the present invention as shown in the specification and described above. Any equivalent changes, modifications and evolutions made by using the technical content disclosed above are all equivalent embodiments of the present invention. At the same time, any equivalent changes, modifications and evolutions made to the above embodiments based on the essential technology of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. A method for efficient one-step federated ensemble learning for gatekeeper communication, characterized in that: The steps include: S1. Based on a generative adversarial network consisting of a generator and a discriminator, the local raw data of several participants are anonymized and synthesized to generate a synthetic dataset that is consistent with the structure and statistical characteristics of the original data; S2, transmit the synthetic data sets of each participant to the only collection party at one time through the secure isolation network; The secure isolation gateway includes an intranet host system, an extranet host system, and an isolation switch matrix. The isolation switch matrix implements secure caching and exchange of data through an intranet isolation switch module and an extranet dual isolation switch module. S3. The synthetic datasets of the collector and each participant are merged with the local data to form a complete dataset. After preprocessing and dividing the complete dataset into training sets, centralized model training is performed based on the complete dataset. An iterative optimization method is used until the model converges. The trained model is distributed to each participant to complete a federated integrated learning.

2. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 1, characterized in that: In S1, each participant preprocesses the local original data, divides the preprocessed data into multiple batches and marks them as real data to input into the discriminator; at the same time, the generator generates false data based on the noise vector and the conditional vector and marks them as false data and inputs them into the discriminator.

3. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 1, characterized in that: In S1, the training loss of the generator includes: the cross entropy loss between the conditional vector and the generated data label, the discrimination loss of the discriminator output, the classification loss of the classifier output, and the sum of the statistical feature difference loss between the real data and the fake data.

4. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 1, characterized in that: In S1, the generative adversarial network is a trained generative adversarial network, and the generative adversarial network training adopts a differential privacy stochastic gradient descent method, which includes clipping the gradient and adding Gaussian noise.

5. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 1, characterized in that: In S2, the external network isolation switching module encapsulates the synthesized data into a private protocol data packet through the security isolation chip and caches it in the switching subsystem.

6. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 5, characterized in that: In S2, the caching and switching of the isolation switching matrix specifically include: After the external network isolation switching module is disconnected from the external network, the cached data is migrated to the switching subsystem of the internal network isolation switching module.

7. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 5, characterized in that: In S2, the intranet isolation switching module reads and parses the private protocol data packet through the security isolation chip, and restores the private protocol data packet to the original data block.

8. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 1, characterized in that: In the S3, the complete data set is obtained by splicing the synthetic data of each participant and the local data of the collector.

9. The efficient one-step federated ensemble learning method for gatekeeper communication according to claim 1, characterized in that: In the S3, the preprocessing of the complete data set includes data cleaning and feature normalization, and the iterative optimization method adopts the stochastic gradient descent algorithm.

10. The system according to any one of claims 1 to 9, wherein the method for efficient one-step federated ensemble learning for gatekeeper communication is based on, characterized in that: The system includes a data synthesis module, a data exchange and transmission module, and a model training and distribution module, wherein: Data synthesis module: used to anonymize and synthesize the local raw data of several participants based on a generative adversarial network consisting of a generator and a discriminator, and generate a synthetic dataset that is consistent with the structure and statistical characteristics of the original data; Data exchange and transmission module: used to transmit the synthetic data sets of each participant to the only collection party at one time through a secure isolation network; The secure isolation gateway includes an intranet host system, an extranet host system, and an isolation switch matrix. The isolation switch matrix implements secure caching and exchange of data through an intranet isolation switch module and an extranet dual isolation switch module. Model training and distribution module: used to merge the synthetic data sets of the collector and each participant with the local data to form a complete data set. After preprocessing and dividing the complete data set into training sets, centralized model training is performed based on the complete data set. An iterative optimization method is used until the model converges. The trained model is distributed to each participant to complete a federated integrated learning.

Images