CN117938688A

CN117938688A - Optimized training method, device and related equipment for generating countermeasure network

Info

Publication number: CN117938688A
Application number: CN202211249657.2A
Authority: CN
Inventors: 于梦晗; 李鹏宇
Original assignee: China Telecom Corp Ltd
Current assignee: China Telecom Corp Ltd
Priority date: 2022-10-12
Filing date: 2022-10-12
Publication date: 2024-04-26

Abstract

The disclosure provides an optimization training method, device and related equipment for generating an countermeasure network, and relates to the technical field of communication, wherein the method comprises the following steps: acquiring a first generator discarding rate, a first discriminant discarding rate, a first generator network splitting point and a first network uploading time of each edge node; generating an initial generator network and an initial discriminator network of each edge node according to the first generator discard rate, the first discriminator discard rate and the first generator network split point of each edge node; and transmitting the initial generator network, the initial arbiter network, the first generator network split point and the first network upload time of each edge node to each edge node for federal training. According to the method and the device, the federal generation countermeasure network is combined with the discarding method and the network splitting, so that the communication data volume between the center node and the edge node can be effectively reduced, bandwidth resources are saved, and the training efficiency of the federal generation countermeasure network is remarkably improved.

Description

Optimized training method, device and related equipment for generating countermeasure network

Technical Field

The present disclosure relates to the field of communications technologies, and in particular, to an optimization training method and apparatus for generating an countermeasure network, a computer readable storage medium, and an electronic device.

Background

The generation countermeasure Network (GAN) can generate analog data by labeling data, and can enhance the existing data, so that the method is widely applied to scenes such as image recognition, malicious attack detection, three-dimensional structure generation and the like.

The federal generation countermeasure network (FedGAN) combines federal learning and GAN, can complete updating of the generator and the arbiter on the premise of guaranteeing user data privacy, and has wide application prospect in the 6G network.

However, the existing federal generation countermeasure network needs to transmit the identifier network and the generator network at the same time, so that the data volume transmitted by each round of communication is too large, which results in the technical problems of excessive bandwidth resource occupation and low training efficiency of a network model.

It should be noted that the information disclosed in the above background section is only for enhancing understanding of the background of the present disclosure and thus may include information that does not constitute prior art known to those of ordinary skill in the art.

Disclosure of Invention

The disclosure aims to provide an optimized training method, an apparatus, a computer readable storage medium and an electronic device for generating an countermeasure network, so as to at least solve the technical problems of excessive bandwidth resources occupation and low training efficiency of a network model caused by excessive data volume of communication transmission in the related technology.

Other features and advantages of the present disclosure will be apparent from the following detailed description, or may be learned in part by the practice of the disclosure.

The technical scheme of the present disclosure is as follows:

According to one aspect of the present disclosure, there is provided an optimized training method of generating an countermeasure network, the method comprising: acquiring a first generator discarding rate, a first discriminant discarding rate, a first generator network splitting point and a first network uploading time of each edge node; generating an initial generator network and an initial discriminator network of each edge node according to the first generator discard rate, the first discriminator discard rate and the first generator network split point of each edge node; and transmitting the initial generator network, the initial arbiter network, the first generator network split point and the first network upload time of each edge node to each edge node for federal training.

In some embodiments of the present disclosure, the step of obtaining a first generator discard rate, a first arbiter discard rate, a first generator network split point, a first network upload time for each edge node comprises: and sending an analysis request to the analysis node so that the analysis node obtains related data from the source node according to the analysis request and calculates to obtain a first generator discarding rate, a first discriminator discarding rate, a first generator network splitting point and a first network uploading time.

In some embodiments of the present disclosure, the federal training includes: acquiring a local generator network, a local discriminator network and training information which are obtained by each edge node through local training according to the initial generator network and the initial discriminator network; and adjusting the contribution weight of each edge node according to the training information, and respectively aggregating the local generator network and the local discriminator network according to the contribution weights to obtain an aggregation generator network and an aggregation discriminator network.

In some embodiments of the present disclosure, after the step of sending the initial generator network, the initial arbiter network, the first generator network split point, and the first network upload time for each edge node to each edge node for federal training, the method further comprises: testing the training result of the federal training to obtain an actual testing result, and sending training feedback information to the analysis node if the actual testing result meets the feedback condition; acquiring a second generator discarding rate, a second discriminant discarding rate, a second generator network splitting point and a second network uploading time, which are calculated by the analysis node for each edge node according to training feedback information; generating a fine tuning generator network and a fine tuning discriminator network of each edge node according to the second generator discard rate, the second discriminator discard rate and the second generator network split point; and transmitting the fine tuning generator network and the fine tuning arbiter network of each edge node, the second generator network split point and the second network upload time to each edge node for federal training.

In some embodiments of the present disclosure, the method further comprises: predefining test thresholds for the generator network and the arbiter network; obtaining test values of a generator network and a discriminator network in an actual test result, and judging whether the test values respectively reach a test threshold value or not; if the test threshold values are respectively reached, judging that the feedback condition is met; and if the test thresholds are not respectively reached, judging that the feedback conditions are not met, and continuing to execute the federal training.

According to one aspect of the present disclosure, there is provided an optimized training method of generating an countermeasure network, the method comprising: receiving an initial generator network, an initial discriminator network, a first generator network splitting point and a first network uploading time which are sent by a central node; performing local training according to the initial generator network, the initial discriminator network and the first generator network splitting point to obtain a local generator network, a local discriminator network and training information; and transmitting the local generator network, the local arbiter network, and the training information to the central node at the first network upload time.

According to yet another aspect of the present disclosure, there is provided an optimized training device for generating an countermeasure network, the device comprising: the first information acquisition module is used for acquiring a first generator discarding rate, a first discriminator discarding rate, a first generator network splitting point and a first network uploading time of each edge node; the initial network generation module is used for generating an initial generator network and an initial discriminator network of each edge node according to the first generator discarding rate, the first discriminator discarding rate and the first generator network splitting point of each edge node; and the initial network sending module is used for sending the initial generator network, the initial discriminator network, the first generator network splitting point and the first network uploading time of each edge node to each edge node for federal training.

According to yet another aspect of the present disclosure, there is provided an optimized training device for generating an countermeasure network, the device comprising: the initial network receiving module is used for receiving an initial generator network, an initial discriminator network, a first generator network splitting point and a first network uploading time which are sent by the central node; the initial network training module is used for carrying out local training according to the initial generator network, the initial discriminator network and the first generator network splitting point to obtain a local generator network, a local discriminator network and training information; and the local network sending module is used for sending the local generator network, the local discriminator network and the training information to the central node at the first network uploading time.

According to still another aspect of the present disclosure, there is provided an electronic apparatus including: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the above-described optimized training method of generating an countermeasure network via execution of the executable instructions.

According to yet another aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the above-described method of generating optimized training for an countermeasure network.

According to the method and the device, the federal generation countermeasure network is combined with the discarding method and the network splitting, so that the communication data volume between the center node and the edge node can be effectively reduced, bandwidth resources are saved, and the training efficiency of the federal generation countermeasure network is remarkably improved.

Further, by discarding and splitting the network model, the risk of overfitting and the difficulty of deployment are reduced, and the performance of the network model is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure. It will be apparent to those of ordinary skill in the art that the drawings in the following description are merely examples of the disclosure and that other drawings may be derived from them without undue effort.

Fig. 1 shows a schematic structural diagram of a communication system to which the technical solution provided in the embodiments of the present disclosure is applicable.

FIG. 2 illustrates a flow diagram of an optimized training method for generating an countermeasure network in an embodiment of the present disclosure.

Fig. 3 shows a flow diagram of an adjustment training method in an optimization training method for generating an countermeasure network in an embodiment of the present disclosure.

Fig. 4 is a flow chart illustrating a trigger method for adjusting training in an optimized training method for generating an countermeasure network in an embodiment of the present disclosure.

Fig. 5 shows a flow diagram of an optimized training method performed by an edge node to generate an countermeasure network in an embodiment of the disclosure.

FIG. 6 illustrates an interactive schematic diagram of an optimized training method to generate an countermeasure network in an embodiment of the present disclosure.

Fig. 7 shows a schematic structural diagram of an optimized training device for generating an countermeasure network in an embodiment of the present disclosure.

Fig. 8 illustrates a schematic structure of yet another optimized training device for generating an countermeasure network in an embodiment of the present disclosure.

Fig. 9 illustrates a schematic block diagram of an electronic device generating optimized training for an countermeasure network in an embodiment of the disclosure.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments may be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Furthermore, the drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor devices and/or microcontroller devices.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present disclosure, the meaning of "a plurality" is at least two, such as two, three, etc., unless explicitly specified otherwise.

In view of the technical problems in the related art, embodiments of the present disclosure provide an encrypted traffic detection method, which is used to at least solve one or all of the technical problems.

It should be noted that, the terms or terms related to the embodiments of the present application may be referred to each other, and are not repeated.

Hereinafter, each step of the optimized training method for generating an countermeasure network in the present exemplary embodiment will be described in more detail with reference to the accompanying drawings and examples.

Fig. 1 shows a schematic structural diagram of a communication system to which the technical solution provided in the embodiments of the present disclosure is applicable. The communication system as shown in fig. 1 includes: server cluster 130, application server 120, and user device 110.

The server cluster 130 is a server, or is composed of a plurality of servers, or is a virtualization platform, or is a cloud computing service center. The server cluster 130 is a server that provides Network data, and may be NWDAF (Network DATA ANALYTICS Function) Network elements that monitor Network performance.

The server cluster 130 may serve as an analysis node to distribute network performance data to at least one application server 120. In some embodiments of the present disclosure, the server cluster 130 may further include a source node for providing relevant network data to the analysis node, such as UPF (User Plane Function ) network elements, OAM (Operation Administration AND MAINTENANCE ) network elements, AMF network elements (ACCESS AND Mobility Management Function, access and mobility management functions), etc.

The application server 120 and the user equipment 110 perform cooperative computing by using a distributed computing architecture. Each application server 120 may serve as a central node to assign training tasks to different edge nodes, e.g., user devices 110, and generate a global network model based on training results of the local network models of the respective user devices 110.

Optionally, the application server 120 is connected to the server cluster 130 through a communication network. Optionally, the communication network is a wired network or a wireless network.

The user device 110 may be a mobile terminal such as a mobile phone, a game console, a tablet computer, an electronic book reader, smart glasses, an MP4 (MovingPicture Experts Group Audio Layer IV, dynamic image expert compression standard audio plane 4) player, a smart home device, an AR (Augmented Reality ) device, a VR (Virtual Reality) device, or the user device 110 may be a personal computer (Personal Computer, PC) such as a laptop portable computer and a desktop computer, etc.

The user equipment 110 is connected to the application server 120 via a communication network. Optionally, the communication network is a wired network or a wireless network.

Those skilled in the art will appreciate that the number of user devices 110 may be greater or lesser. For example, the number of the user devices 110 may be only one, or the number of the user devices 110 may be several tens or hundreds, or more. The number and device types of the user devices 110 are not limited in the embodiment of the present application.

Alternatively, the wireless network or wired network described above uses standard communication techniques and/or protocols. The network is typically the Internet, but may be any network including, but not limited to, a local area network (Local Area Network, LAN), metropolitan area network (Metropolitan Area Network, MAN), wide area network (Wide Area Network, WAN), mobile, wired or wireless network, private network, or any combination of virtual private networks. In some embodiments, data exchanged over the network is represented using techniques and/or formats including HyperText Mark-up Language (HTML), extensible markup Language (Extensible MarkupLanguage, XML), and the like. All or some of the links may also be encrypted using conventional encryption techniques such as secure sockets layer (Secure Socket Layer, SSL), transport layer security (Transport Layer Security, TLS), virtual private network (Virtual Private Network, VPN), internet protocol security (Internet ProtocolSecurity, IPsec), etc. In other embodiments, custom and/or dedicated data communication techniques may also be used in place of or in addition to the data communication techniques described above.

It should be understood that the technical solution of the embodiment of the present invention may be applied to various distributed networks, for example: global system for mobile communications (GSM, global System of Mobile Communication) systems, code division multiple access (CDMA, code Division Multiple Access) systems, wideband code division multiple access (WCDMA, wide Code Division Multiple Access) systems, universal packet Radio Service (GPRS, general Packet Radio Service), long term evolution (LTE, long Term Evolution) systems, LTE frequency division duplex (FDD, frequency Division Duplex) systems, LTE time division duplex (TDD, time Division Duplex) systems, universal mobile telecommunications system (UMTS, universal Mobile Telecommunication System) or worldwide interoperability for microwave access (Wimax, worldwide Interoperability for Microwave Access) communication systems, and the like.

FIG. 2 illustrates a flow diagram of an optimized training method for generating an countermeasure network in an embodiment of the present disclosure. The method provided by the embodiments of the present disclosure may be applied to an application server (central node) shown in fig. 1.

As shown in fig. 2, the method 200 may include the steps of:

In step S210, a first generator discard rate, a first arbiter discard rate, a first generator network split point, and a first network upload time of each edge node are acquired.

The generator discarding rate is the proportion of the generator network parameters which are randomly discarded by the edge node to all the generator network parameters.

The identifier discarding rate is the proportion of the identifier network parameters which are randomly discarded by the edge node to all the identifier network parameters.

Wherein the generator network split point is the optimal split point of the edge node to the generator network.

Wherein the network upload time is the best time period for the edge node to upload the network model.

In step S220, an initial generator network and an initial discriminator network for each edge node are generated based on the first generator discard rate, the first discriminator discard rate, and the first generator network split point for each edge node.

In step S230, the initial generator network, the initial arbiter network, the first generator network split point, and the first network upload time for each edge node are sent to each edge node for federal training.

The central node and the edge nodes can perform one or more rounds of federation training, the edge nodes return a trained network model to the central node after each round of training, and then the central node can re-determine the edge nodes which need to participate in federation training and distribute training parameters to the participating edge nodes.

In some embodiments of the present disclosure, step S210 may further include: and sending an analysis request to the analysis node so that the analysis node obtains related data from the source node according to the analysis request and calculates to obtain a first generator discarding rate, a first discriminator discarding rate, a first generator network splitting point and a first network uploading time. The network performance data of a plurality of different functional nodes are collected, comprehensive training is performed to generate an countermeasure network, accuracy of the network is improved, and the network is suitable for complex and changeable network environments.

The analysis node may be NWDAF network elements, and is configured to monitor network performance, and obtain the network parameters according to the network performance.

The source node may include a UPF (User Plane Function ) network element, an OAM (Operation Administration AND MAINTENANCE ) network element, an AMF network element (ACCESS AND Mobility Management Function, access and mobility management functions), and the like.

Wherein the analysis request may include request information and edge node information.

For example, the request information may include at least one of a network congestion level, a data network performance, a wireless local area network performance, a quality of service sustainability.

For example, the edge node information may include at least one of area information of the edge node, time window information of the edge node (including federal learning start time per round, shortest and longest durations per round), producer network information of the edge node (e.g., size of producer network), arbiter network information of the edge node (e.g., size of arbiter network), local data volume of the edge node, computing power of the edge node.

The analysis node can directly calculate the first generator discarding rate, the first discriminator discarding rate, the first generator network splitting point and the first network uploading time of each edge node according to the service characteristic pre-configuration related rule. Alternatively, the analysis node may derive the parameters based on pre-training a meta-learning model or reinforcement learning model.

In some embodiments of the present disclosure, the federal training method in step S230 may further include: acquiring a local generator network, a local discriminator network and training information which are obtained by carrying out local training on each edge node according to an initial generator network, an initial discriminator network, a first generator network splitting point and first network uploading time; and adjusting the contribution weight of each edge node according to the training information, and respectively aggregating the local generator network and the local discriminator network according to the contribution weights to obtain an aggregation generator network and an aggregation discriminator network. The local network trained by the plurality of edge nodes is aggregated to train the global aggregation network, so that the feature representation for carrying out global information is enhanced, and the accuracy of the network is improved.

Wherein the training information is data generated during the training process. For example, it may include: training time of the generator network and the arbiter network of the edge node, uploading time of the second network, local test results of the generator network and the arbiter network, and data amount change.

For example, if the longer the training time of the edge node's generator network and the arbiter network, the longer the network upload time, the smaller the contribution weight assigned to that edge node.

In some embodiments of the present disclosure, the central node further feeds back the results of the federal training to the analysis node after step S230, which readjusts the generator discard rate, the arbiter discard rate, the generator network split point, and the network upload time based on the collected data. For example, fig. 3 shows a flow diagram of an adjustment training method in an optimization training method for generating an countermeasure network in an embodiment of the present disclosure. As shown in fig. 3, the method 300 may include the steps of:

in step S310, the training result of the federal training is tested to obtain an actual test result, and if the actual test result meets the feedback condition, training feedback information is sent to the analysis node.

Wherein the feedback conditions may be set manually by the user.

The training feedback information may include, among other things, analysis requests to adjust federal training results and actual generator discard rates for each edge node, arbiter discard rates for each edge node, generator network split points for each edge node, training delays for each edge node, communication delays for each edge node, local test results for generator and arbiter networks for each edge node, contribution weights for generator and arbiter networks for each edge node, and test results for aggregate generators and aggregate arbiters.

The training time delay of each edge node is a generator training time delay and a discriminant training time delay calculated according to the starting time and the ending time of the generator and discriminant training.

Wherein the communication delay of each edge node is the difference between the reception completion time and the start upload time.

The local test results of the generator network and the arbiter network of each edge node comprise the accuracy, generalization error, deviation and the like of the generator network and the arbiter network.

In step S320, the second generator discard rate, the second discriminator discard rate, the second generator network split point, and the second network upload time calculated by the analysis node for each edge node in response to the training feedback information are acquired.

For example, if the training delay and the communication delay of an edge node are reflected in the training feedback information, the discarding rate of the generator and the discarding rate of the discriminator are properly increased; if the edge node training and communication delay is short in the training feedback information, the local test result is close to the actual test result, and the discarding rate can be properly reduced.

In step S330, a fine-tuning generator network and a fine-tuning arbiter network for each edge node are generated based on the second generator discard rate, the second arbiter discard rate, and the second generator network split point.

In step S340, the fine tuning generator network and the fine tuning arbiter network, the second generator network split point, and the second network upload time for each edge node are sent to each edge node for federal training.

The method of generating the trim generator network and the trim arbiter network for each edge node in step S330 and step S340 is similar to that in step S220 and step S230, and thus will not be described herein.

According to the embodiment of the disclosure, a closed-loop mechanism can be formed by the method for performing federal training by adjusting the center node and the edge node through the analysis node, so that the self-monitoring and self-evolution of the network are realized, and the network risk is reduced.

In some embodiments of the present disclosure, step S310 further includes a method for determining whether the actual test result meets the feedback condition, for example, a method 400 shown in fig. 4, and as shown in fig. 4, the method 400 may include the following steps:

in step S410, test thresholds for the generator network and the arbiter network are predefined.

The test threshold may include, but is not limited to: an accuracy threshold, a generalization error threshold, and a bias threshold for the generator network and the arbiter network.

In step S420, test values of the generator network and the arbiter network in the actual test result are obtained, and it is determined whether the test thresholds are reached respectively.

In some embodiments of the present disclosure, it may be determined whether the actual test accuracy, the actual test error, the actual test bias of the generator network and the arbiter network, respectively, reach respective accuracy thresholds, generalization error thresholds, and bias thresholds, respectively.

In step S430, if the test threshold values have been reached, it is determined that the feedback condition is satisfied.

In step S440, if the test thresholds are not reached, it is determined that the feedback conditions are not satisfied, and the federal training is continued.

By setting feedback conditions, the closed-loop control system of the analysis node, the application node and the edge node can be adaptively adjusted, so that the control operation based on the test threshold can adapt to the requirements of rapidly-changing network environment, and the communication quality is improved.

Fig. 5 shows a flow diagram of an optimized training method for generating an countermeasure network in an embodiment of the disclosure. The method provided by the embodiment of the present disclosure may be applied to the user equipment (edge node) shown in fig. 1. As shown in fig. 5, the method 500 may include the steps of:

In step S510, the initial generator network and the initial arbiter network, the first generator network split point, and the first network upload time transmitted by the center node are received.

In step S520, local training is performed according to the initial generator network, the initial arbiter network, and the first generator network split point to obtain a local generator network, a local arbiter network, and training information.

In step S530, the local generator network, the local arbiter network, and the training information are transmitted to the central node at the first network upload time.

Step S510 to step S530 in the method 500 correspond to step S230 in the method 200, and are not described herein.

The method combines the federal generation countermeasure network with the discarding method and the model splitting, can effectively reduce the calculation and communication pressure of the edge nodes, saves bandwidth resources, and remarkably improves the training efficiency of the federal generation countermeasure network.

In some embodiments of the present disclosure, one or more rounds of federal training may also be performed between the edge node and the central node. For example, the central node, upon receiving the local generator network, the local arbiter network, and the training information sent by the edge nodes, may re-determine the edge nodes that need to participate in federal learning, i.e., the participating edge nodes, and assign parameters to the participating edge nodes.

In some embodiments of the present disclosure, the step of transmitting, by the edge node, the local producer network, the local arbiter network, and the training information to the central node at a first network upload time further comprises an adjustment training of the producer network and the arbiter network. Specifically, acquiring, by the edge node, a fine-tuning generator network and a fine-tuning arbiter network for each edge node generated by the center node according to the second generator discard rate, the second arbiter discard rate, and the second generator network split point; performing local training according to the fine tuning generator network, the fine tuning initial discriminant network and the second generator network split point to obtain a local generator network, a local discriminant network and training information; and transmitting the local generator network, the local arbiter network, and the training information to the central node at a second network upload time. Specifically, the method corresponds to the method 300 in fig. 3, and thus is not described herein. The method for adjusting training in the embodiment of the disclosure can form a closed-loop mechanism to realize self-monitoring and self-evolution of the network, thereby reducing network risks.

FIG. 6 illustrates an interactive schematic diagram of an optimized training method to generate an countermeasure network in an embodiment of the present disclosure. As shown in fig. 6, the method 600 may include the steps of:

in step S602, the central node needs to train the federally generated countermeasure network to send an analysis request to an analysis node (e.g., NWDAF network elements).

In step S604, the analysis node collects relevant data from the data source (e.g., UPF, OAM, AMF, etc.) according to the received request information, where the relevant data may include, for example, uplink and downlink throughput, uplink and downlink data transmission rate, quality of service, performance data of the edge node device, etc. of each edge node.

In step S606, the analysis node performs analysis based on the collected data, mainly to predict the network performance of each edge node. The analysis node deduces the optimal parameters such as the generator discarding rate, the discriminator discarding rate, the generator network splitting point, the network uploading time and the like of each edge node according to the information received in the step S602 and the data collected in the step S604. Inference modes include, but are not limited to: 1) The analysis node pre-configures related rules according to service characteristics, and directly calculates to obtain optimal parameters; 2) The analysis node pre-trains a meta learning model or reinforcement learning model.

In step S608, the analysis node infers the relevant parameter response of each edge node in step S606 to the central node.

In step S610, the central node decides the final parameters such as the generator discard rate, the arbiter discard rate, the generator network split point, and the network upload time of each edge node according to the response of the analysis node and the internal logic configuration. The central node generates an initial generator network and an initial arbiter network for each edge node locally based on the drop rate and the network split point.

In step S612, the central node transmits the parameters of the initial generator network, the initial arbiter network, the generator network split point, the network split related parameters, the network upload time, and the like of each edge node to each edge node.

In step S614, each edge node performs local training according to the received parameters, and calculates a generator training delay and a arbiter training delay according to the start time and the end time of the training. The training time delay refers to the time of local training, and the network is not necessarily uploaded immediately after the training is completed.

In step S616, the edge node uploads the network to the central node. Parameters include, but are not limited to, post-training generator network, arbiter network, training delay, start upload time, generator performance local test results, arbiter performance local test results (including accuracy, generalization error, bias, etc.), local data volume changes, etc.

In step S618, the central node decides the contribution weight of each edge node and performs network aggregation, and calculates the communication delay of each member. The communication delay is calculated by subtracting the time to start uploading from the time to finish receiving. And then, the central node evaluates the aggregation generator and the aggregation discriminator by using the local test data to obtain the actual test results of the aggregation generator and the aggregation discriminator.

In step S620, after one or more rounds of federal learning training, the central node sends training feedback information to the analysis node.

In steps S622 to S626, the analysis node collects relevant data (step S604) and analyzes the relevant data according to the training feedback information received in step S620 and the collected data to obtain a new generator discard rate, a new discriminator discard rate, a new generator model split point, and a new model upload time window for each edge node, where the method includes, but is not limited to: 1) The AI analysis network element pre-configures a related formula according to the service characteristics, and directly calculates to obtain the optimal parameters; 2) The AI-analyzing network element pre-trains a meta-learning model or reinforcement learning model (step S606).

To name a few examples: if the training time delay and the communication time delay of a certain edge node are longer, the discarding rate of the generator and the discriminator is properly increased; the training and communication delay of an edge node is shorter, the local test result is close to the actual test result, and the discarding rate can be properly increased. The discard rates of the generator and the arbiter are different and should be calculated separately.

In step S628, the central node decides the final parameters such as the generator discard rate, the arbiter discard rate, the generator network split point, and the network upload time of each edge node according to the recommendation of the analysis node and the internal logic configuration. The central node generates an initial network model for each edge node based on the actual drop rate and split point for each edge node (as in step S610).

In step S630, the center node and the edge nodes continue federation learning, and each round of federation learning training should include three steps S612-S616. Steps S620-S630 should be repeated until the federal learning ends.

The method disclosed by the invention can be applied to a 5G network, wherein NWDAF network elements are used as analysis nodes, AF application function network elements or application servers are used as central nodes, user equipment is used as edge nodes, and the GAN scene is jointly federally trained. NWDAF network elements assist the AF/application server in network performance analysis and calculate relevant parameters.

The method can also be applied to other networks, such as a 6G network, and the analysis node assists the application layer to perform federal training with the edge node. Therefore, the analysis node, the center node and the edge node form a closed loop, and the self-monitoring and self-evolution of the network are realized, so that the network risk is reduced.

Fig. 7 shows a schematic structural diagram of an optimized training device for generating an countermeasure network in an embodiment of the present disclosure. As shown in fig. 7, the apparatus 700 includes:

A first information obtaining module 710, configured to obtain a first generator discard rate, a first arbiter discard rate, a first generator network split point, and a first network upload time of each edge node; an initial network generation module 720, configured to generate an initial generator network and an initial discriminator network for each edge node according to the first generator discard rate, the first discriminator discard rate, and the first generator network split point for each edge node; an initial network sending module 730, configured to send the initial generator network, the initial arbiter network, the first generator network split point, and the first network upload time of each edge node to each edge node for federal training.

In some embodiments of the present disclosure, the first information obtaining module 710 may be further configured to send an analysis request to the analysis node, so that the analysis node obtains the relevant data from the source node according to the analysis request, and calculates a first generator discard rate, a first arbiter discard rate, a first generator network split point, and a first network upload time.

In some embodiments of the present disclosure, the apparatus 700 may further include: the training feedback module is used for testing the training result of the federal training to obtain an actual test result, and sending training feedback information to the analysis node if the actual test result meets the feedback condition; the second information acquisition module is used for acquiring a second generator discarding rate, a second discriminant discarding rate, a second generator network splitting point and a second network uploading time, which are calculated by the analysis node for each edge node according to training feedback information; the fine tuning network generation module is used for generating a fine tuning generator network and a fine tuning discriminator network of each edge node according to the second generator discarding rate, the second discriminator discarding rate and the second generator network splitting point; and the fine tuning network sending module is used for sending the fine tuning generator network and the fine tuning discriminator network of each edge node, the second generator network split point and the second network uploading time to each edge node for federal training.

In some embodiments of the present disclosure, a feedback condition judgment module for predefining test thresholds for the generator network and the arbiter network; obtaining test values of a generator network and a discriminator network in an actual test result, and judging whether the test values respectively reach a test threshold value or not; if the test threshold values are respectively reached, judging that the feedback condition is met; if the test thresholds are not respectively reached, judging that the feedback conditions are not met, and continuing to execute the federal training.

Fig. 8 illustrates a schematic structure of yet another optimized training device for generating an countermeasure network in an embodiment of the present disclosure. As shown in fig. 8, the apparatus 800 includes:

An initial network receiving module 810, configured to receive an initial generator network and an initial arbiter network, a first generator network split point, and a first network upload time sent by a central node; an initial network training module 820 for performing local training according to the initial generator network, the initial arbiter network, and the first generator network split point to obtain a local generator network, a local arbiter network, and training information; and a local network sending module 830, configured to send the local generator network, the local arbiter network, and training information to the central node at the first network upload time.

In some embodiments of the present disclosure, the apparatus 800 may further include: a fine tuning network acquisition module for acquiring a fine tuning generator network and a fine tuning arbiter network for each edge node generated by the center node according to the second generator discard rate, the second arbiter discard rate, and the second generator network split point; the adjustment training module is used for carrying out local training according to the fine adjustment generator network, the fine adjustment initial discriminant network and the second generator network split point to obtain a local generator network, a local discriminant network and training information; and a second local network sending module for sending the local generator network, the local arbiter network, and the training information to the central node at a second network upload time.

With respect to the optimized training device 700 for generating an countermeasure network and the optimized training device 800 for generating an countermeasure network in the above embodiments, a specific manner in which the respective modules perform operations has been described in detail in the embodiments related to the method, and will not be described in detail herein.

Those skilled in the art will appreciate that the various aspects of the present disclosure may be implemented as a system, method, or program product. Accordingly, various aspects of the disclosure may be embodied in the following forms, namely: an entirely hardware embodiment, an entirely software embodiment (including firmware, micro-code, etc.) or an embodiment combining hardware and software aspects may be referred to herein as a "circuit," module "or" system.

An electronic device 900 according to such an embodiment of the present disclosure is described below with reference to fig. 9. The electronic device 900 shown in fig. 9 is merely an example and should not be construed to limit the functionality and scope of use of embodiments of the present disclosure in any way.

As shown in fig. 9, the electronic device 900 is embodied in the form of a general purpose computing device. Components of electronic device 900 may include, but are not limited to: the at least one processing unit 910, the at least one storage unit 920, and a bus 930 connecting the different system components (including the storage unit 920 and the processing unit 910).

Wherein the storage unit stores program code that is executable by the processing unit 910 such that the processing unit 910 performs steps according to various exemplary embodiments of the present disclosure described in the above-described "exemplary methods" section of the present specification. For example, the processing unit 910 may perform step S210 shown in fig. 2, obtain a first generator discard rate, a first arbiter discard rate, a first generator network split point, and a first network upload time of each edge node; step S220, generating an initial generator network and an initial discriminator network of each edge node according to the first generator discarding rate, the first discriminator discarding rate and the first generator network splitting point of each edge node; step S230, the initial generator network, the initial arbiter network, the first generator network split point and the first network upload time of each edge node are sent to each edge node for federal training.

The storage unit 920 may include readable media in the form of volatile storage units, such as Random Access Memory (RAM) 9201 and/or cache memory 9202, and may further include Read Only Memory (ROM) 9203.

The storage unit 920 may also include a program/utility 924 having a set (at least one) of program modules 9205, such program modules 9205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment.

The bus 930 may be one or more of several types of bus structures including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The electronic device 900 may also communicate with one or more external devices (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 900, and/or with any device (e.g., router, modem, etc.) that enables the electronic device 900 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 950. Also, electronic device 900 may communicate with one or more networks such as a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet, through network adapter 960. As shown, the network adapter 960 communicates with other modules of the electronic device 900 over the bus 930. It should be appreciated that although not shown, other hardware and/or software modules may be used in connection with electronic device 900, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

In an exemplary embodiment of the present disclosure, a computer-readable storage medium having stored thereon a program product capable of implementing the method described above in the present specification is also provided. In some possible implementations, various aspects of the disclosure may also be implemented in the form of a program product comprising program code for causing a terminal device to carry out the steps according to the various exemplary embodiments of the disclosure as described in the "exemplary methods" section of this specification, when the program product is run on the terminal device.

A program product for implementing the above-described method according to an embodiment of the present disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, server, terminal, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, server, terminal, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The computer readable signal medium may include a data signal propagated in baseband or as part of a carrier wave with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, server, terminal, or device.

Program code embodied on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).

According to one aspect of the present disclosure, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the methods provided in the various alternative implementations of the above-described embodiments.

It should be noted that although in the above detailed description several modules or units of a device for action execution are mentioned, such a division is not mandatory. Indeed, the features and functionality of two or more modules or units described above may be embodied in one module or unit in accordance with embodiments of the present disclosure. Conversely, the features and functions of one module or unit described above may be further divided into a plurality of modules or units to be embodied.

Furthermore, although the steps of the methods in the present disclosure are depicted in a particular order in the drawings, this does not require or imply that the steps must be performed in that particular order, or that all illustrated steps be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform, etc.

From the above description of embodiments, those skilled in the art will readily appreciate that the example embodiments described herein may be implemented in software, or may be implemented in software in combination with the necessary hardware. Thus, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.) or on a network, including several instructions to cause a computing device (may be a personal computer, a server, a mobile terminal, or a network device, etc.) to perform the method according to the embodiments of the present disclosure.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

Claims

1. A method of optimizing training to generate an countermeasure network, the method comprising:

Acquiring a first generator discarding rate, a first discriminant discarding rate, a first generator network splitting point and a first network uploading time of each edge node;

generating an initial generator network and an initial discriminator network of each edge node according to the first generator discarding rate, the first discriminator discarding rate and the first generator network splitting point of each edge node; and

And transmitting the initial generator network, the initial arbiter network, the first generator network split point and the first network uploading time of each edge node to each edge node for federal training.

2. The method of claim 1, wherein the step of obtaining a first generator discard rate, a first arbiter discard rate, a first generator network split point, a first network upload time for each edge node comprises:

and sending an analysis request to an analysis node so that the analysis node obtains relevant data from a source node according to the analysis request and calculates to obtain a first generator discarding rate, a first discriminator discarding rate, a first generator network splitting point and a first network uploading time.

3. An optimized training method for generating an countermeasure network as claimed in claim 2, wherein said federal training includes:

Acquiring a local generator network, a local discriminator network and training information which are obtained by each edge node through local training according to the initial generator network and the initial discriminator network;

and adjusting the contribution weight of each edge node according to the training information, and respectively aggregating the local generator network and the local discriminator network according to the contribution weight to obtain an aggregation generator network and an aggregation discriminator network.

4. An optimized training method for generating an countermeasure network as claimed in claim 3, after the step of transmitting said initial generator network, said initial arbiter network, said first generator network split point, and said first network upload time for each edge node to each edge node for federal training, said method further comprising:

Testing the training result of the federal training to obtain an actual testing result, and if the actual testing result meets a feedback condition, sending training feedback information to the analysis node;

Acquiring a second generator discarding rate, a second discriminant discarding rate, a second generator network splitting point and a second network uploading time, which are calculated by the analysis node for each edge node according to the training feedback information;

Generating a fine tuning generator network and a fine tuning discriminator network of each edge node according to the second generator discard rate, the second discriminator discard rate and the second generator network split point; and

And sending the fine tuning generator network, the fine tuning discriminator network, the second generator network split point and the second network uploading time of each edge node to each edge node for federal training.

5. The method of optimizing training for generating an countermeasure network of claim 4, the method further comprising:

Predefining test thresholds for the generator network and the arbiter network;

Obtaining test values of a generator network and a discriminator network in an actual test result, and judging whether the test values respectively reach the test threshold values;

If the test threshold values are respectively reached, judging that the feedback condition is met;

And if the test thresholds are not respectively reached, judging that the feedback conditions are not met, and continuing to execute the federal training.

6. A method of optimizing training to generate an countermeasure network, the method comprising:

Receiving an initial generator network, an initial discriminator network, a first generator network splitting point and a first network uploading time which are sent by a central node;

performing local training according to the initial generator network, the initial discriminator network and the first generator network splitting point to obtain a local generator network, a local discriminator network and training information; and

And transmitting the local generator network, the local arbiter network and training information to the central node at the first network uploading time.

7. An optimized training device for generating an countermeasure network, the device comprising:

The first information acquisition module is used for acquiring a first generator discarding rate, a first discriminator discarding rate, a first generator network splitting point and a first network uploading time of each edge node;

the initial network generation module is used for generating an initial generator network and an initial discriminator network of each edge node according to the first generator discarding rate, the first discriminator discarding rate and the first generator network splitting point of each edge node; and

And the initial network sending module is used for sending the initial generator network, the initial discriminator network, the first generator network splitting point and the first network uploading time of each edge node to each edge node for federal training.

8. An optimized training device for generating an countermeasure network, the device comprising:

The initial network receiving module is used for receiving an initial generator network, an initial discriminator network, a first generator network splitting point and a first network uploading time which are sent by the central node;

The initial network training module is used for carrying out local training according to the initial generator network, the initial discriminator network and the first generator network splitting point to obtain a local generator network, a local discriminator network and training information; and

And the local network sending module is used for sending the local generator network, the local discriminator network and the training information to the central node at the first network uploading time.

9. An electronic device, comprising:

A processor; and

A memory for storing executable instructions of the processor;

wherein the processor is configured to perform the optimized training method of generating an countermeasure network of any of claims 1-6 via execution of the executable instructions.

10. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the method of generating an optimized training of an countermeasure network of any of claims 1 to 6.