CN117648981A - Reasoning method and related device - Google Patents

Abstract

The application discloses an inference method and a related device. The method is applied to an inference system, the inference system comprises a server side and a plurality of clients, and the method comprises the following steps: the first federal learning client collects first network data of a first time slot, which is related to a target network resource, extracts first local features of the first network data and sends the first local features to the federal learning server, the federal learning server collects the local features uploaded by a plurality of clients to calculate global priori features and sends the global priori features to each client, the first federal learning client can infer second network data of a second time slot according to the global priori features, and in the reasoning process, information of data of other clients except the local data is utilized to improve the accuracy of reasoning results.

Description

A reasoning method and related device

Technical Field

The present application relates to the field of computer technology, and in particular to a reasoning method and related devices.

Background Art

Federated learning is a distributed machine learning paradigm that allows multiple parties to collaboratively train artificial intelligence models using their own data without aggregating the original data of multiple parties (different organizations or users). The traditional machine learning paradigm requires the aggregation of a large amount of original data for model training, and the original data used for training is likely to come from multiple different organizations or users. Aggregating the original data of multiple different organizations or users is very likely to cause the risk of data leakage, which will expose information assets for organizations and may leak personal privacy for individual users. The existence of the above problems poses a severe challenge to the training of artificial intelligence models. In order to solve the above problems, federated learning technology came into being. Federated learning allows the original data of multiple parties to be retained locally without multi-party data aggregation, and multiple parties can jointly train artificial intelligence models by collaboratively computing (securely) interacting with intermediate calculation results. Through federated learning technology, the data of multiple users is protected, and the data of multiple parties can be fully utilized to collaboratively train models, thereby obtaining a more powerful model.

Typical federated learning can be divided into three paradigms according to the scenario: horizontal federation, vertical federation, and federated migration (federated domain adaptation). The above three paradigms solve three typical scenarios respectively.

In a horizontal federated learning architecture, there is usually a server and different clients participating in the horizontal federation. During the training phase, the client uses local data for training and uploads the trained model to the server; the server performs a weighted average of the models uploaded by all clients to obtain a global model. The global model is sent to the client for reasoning.

During the inference phase, the client performs inference based on the global model sent by the server and manages the resources of the device based on the inference results. However, since only local data is used for inference, the client's inference results are not accurate enough, which in turn causes confusion in the device's resource management.

Summary of the invention

The present application provides an inference method and related devices, which utilize information of data of other clients in addition to local data during the inference process to improve the accuracy of the inference results.

In a first aspect, the present application provides an inference method, which includes: a first federated learning client sends a first local feature to the federated learning server, where the first local feature is extracted from first network data, where the first network data is data related to a target network resource obtained by the first federated learning client in a first time slot, and the target network resource is a network resource managed by the first federated learning client; the first federated learning client receives a global prior feature from the federated learning server, where the global prior feature is obtained based on the first local feature and a second local feature, and the second local feature is provided by a second federated learning client; the first federated learning client performs inference based on the global prior feature and the second network data to obtain an inference result, where the second network data is data related to the target network resource obtained by the first federated learning client in a second time slot, and the inference result is used to manage the target network resource, wherein the second time slot is the same as the first time slot or after the first time slot.

In the above aspect, the first federated learning client collects the first network data related to the target network resource in the first time slot, extracts the first local feature of the first network data and sends it to the federated learning server. The federated learning server collects the local features uploaded by multiple clients to calculate the global prior feature, and sends it to each client. The first federated learning client can infer the second network data in the second time slot based on the global prior feature. In the inference process, the information of the data of other clients in addition to the local data is used to improve the accuracy of the inference results.

In one possible implementation, the first network data is a sampled value of data related to the target network resources in the first time slot or a statistical value from the third time slot to the first time slot, and the second network data is a sampled value of data related to the target network resources in the second time slot, and the third time slot is before the first time slot.

In a possible implementation, the global prior feature is a feature vector or a first machine learning model, and the first machine learning model is used for inference of the second network data.

In one possible implementation, when the global prior feature is a feature vector, the first federated learning client performs reasoning based on the global prior feature and the second network data to obtain an inference result, including: the first federated learning client performs reasoning based on the global prior feature, the second network data, and a local second machine learning model to obtain an inference result, and the second machine learning model is used for reasoning about the second network data.

In the above possible implementation manner, the first federated learning client needs to input the global prior features and the second network data into the local second machine learning model for reasoning, and use the result output by the second machine learning model as the reasoning result. Using the trainable second machine learning model for reasoning can improve the accuracy of the solution.

In one possible implementation, the first federated learning client performs reasoning based on the global prior features, the second network data, and the local second machine learning model to obtain the reasoning result, including: the first federated learning client inputs the second network data into the third learning model to obtain multiple features of the second network data output by the third learning model; the first federated learning client inputs the global prior features into the second machine learning model to obtain the respective weights of the multiple features of the second network data; the first federated learning client determines the reasoning result based on the multiple features of the second network data and the respective weights of the multiple features of the second network data.

In the above possible implementation manner, before inputting the second network data into the second machine learning model, the first federated learning client may also input the second network data into the third machine learning model to obtain multiple features that can reflect the characteristics of the second network data, which can save computing resources. The first federated learning client inputs the global prior features and the second network data into the local second machine learning model for reasoning, which can be that the first federated learning client inputs the global prior features into the second machine learning model, and can obtain the weights corresponding to the multiple features of the second network data output by the second machine learning model, that is, each feature corresponds to a weight, and then the reasoning result can be determined according to the multiple features of the second network data and the weights of the multiple features of the second network data. Different features correspond to different weights, which improves the accuracy of reasoning.

In one possible implementation, the second machine learning model includes multiple first task models; the first federated learning client performs reasoning based on global prior features, second network data, and a local second machine learning model to obtain reasoning results including: the first federated learning client calculates the weights of the multiple first task models based on the global prior features; the first federated learning client inputs features of the second network data into the multiple first task models to obtain reasoning features output by the multiple first task models; the first federated learning client obtains the reasoning result based on the weights of the multiple first task models and the reasoning features output by the multiple first task models.

In the above possible implementation, the second machine learning model includes multiple first task models, which can process multiple features of the second network data separately. Specifically, after the first federated learning client obtains the global prior features, the weights of the multiple first task models can be determined from the global prior features. The first federated learning client can input the features of the second network data into the multiple first task models. The input features of the multiple first task models can be features of the default type, that is, the features of the second network data are classified by category and input into the first task models of the corresponding category. The first federated learning client can combine the inference features output by the multiple first task models with the weights corresponding to the multiple first task models to perform weighted averaging to obtain the inference results. Training different types of features in different first task models and then weighted averaging based on weights can improve the accuracy of inference.

In a possible implementation, the first federated learning client calculates the weights of each of the plurality of first task models according to the global prior features, including: the first federated learning client calculates the weights of each of the plurality of first task models according to the global prior features and the second network data.

In the above possible implementation manner, the first federated learning client calculates the weight of the first task model by combining the client local data and the global prior features, and uses the real-time second network data to improve the accuracy of the weight.

In a possible implementation, the method further includes: the first federated learning client extracting features of the second network data through a third machine learning model.

In the above possible implementation manner, the features of the second network data are used for reasoning to reduce the amount of reasoning calculation.

In one possible implementation, the third machine learning model includes multiple second task models; the first federated learning client extracts features of the second network data through the third machine learning model, including: the first federated learning client determines the weights of each of the multiple second task models based on the second network data; the first federated learning client inputs the second network data into the multiple second task models to obtain sub-features of the second network data output by the multiple second task models; and obtains features of the second network data based on the weights of each of the multiple second task models and the sub-features of the second network data output by the multiple second task models.

In the above possible implementation manner, the first federated learning client first determines the weights corresponding to multiple second task models based on local data or second network data, and then inputs the second network data into the second task model, so as to obtain the sub-features of the second network data output by the second task model. The first federated learning client can then perform weighted averaging based on the weights corresponding to the multiple second task models and the sub-features output by the second task model to obtain the features of the second network data. Different types of features can be trained in different first task models, and then weighted averaged based on the weights, so as to improve the accuracy of reasoning.

In one possible implementation, each second task model is a layer of autoencoder, and the reconstruction target of the rth task model among the multiple second task models is the residual of the r-1th task model, where r is an integer greater than 1 and represents the number of second task models.

In one possible implementation, when the global prior feature is a first machine learning model, the first federated learning client performs reasoning based on the global prior feature and the second network data to obtain an inference result including: the first federated learning client extracts features of the second network data; the first federated learning client inputs the features of the second network data into the first machine learning model to obtain an inference result output by the first machine learning model.

In the above possible implementations, the federated learning server can directly send the first machine learning model as the inference model to improve the flexibility of the solution.

In one possible implementation, when the global prior feature is a first machine learning model, the first federated learning client performs reasoning based on the global prior feature and the second network data to obtain an inference result, including: the first federated learning client trains the first machine learning model using sample data; the first federated learning client extracts features of the second network data; the first federated learning client inputs the features of the second network data into the trained first machine learning model to obtain an inference result output by the trained first machine learning model.

In the above possible implementations, when the federated learning server directly sends the first machine learning model as the inference model, the client can also further train based on local sample data to improve the accuracy of the inference model.

In one possible implementation, the method further includes: the first federated learning client sends grouping information to the federated learning server, the grouping information indicating the group where the first local feature is located, so that the federated learning server obtains the global prior feature based on the first local feature, the group where the first local feature is located, the second local feature, and the group where the second local feature is located.

In the above possible implementation manner, the first federated learning client may also send grouping information to the server, where the grouping information indicates the group where the local features are located, so that the server obtains the global prior features based on the local features from multiple clients and the groups where the local features are located, thereby avoiding the situation where the global prior features determined by the data of all clients together will affect the output between groups.

In a possible implementation, the method further includes: the first federated learning client receives task synchronization information from the federated learning server, where the task synchronization information is used to indicate the first time slot; and the first federated learning client selects the first network data from the local data according to the task synchronization information.

In the above possible implementation manner, the federated learning server indicates the time slot to which the local features uploaded by the client belong, thereby improving the synchronization of the data uploaded by the client.

A second aspect of the present application provides an inference method, the method comprising: a federated learning server receives a first local feature from a first federated learning client, the first local feature is extracted from first network data, the first network data is data related to a target network resource acquired by the first federated learning client in a first time slot, and the target network resource is a network resource managed by the first federated learning client; the federated learning server obtains a global prior feature based on the first local feature and a second local feature, and the second local feature is provided by the second federated learning client; the federated learning server sends the global prior feature to the first federated learning client, so that the first federated learning client performs inference based on the global prior feature and the second network data to obtain an inference result, the second network data is data related to the target network resource acquired by the first federated learning client in a second time slot, and the inference result is used to manage the target network resource, wherein the second time slot is the same as the first time slot or after the first time slot.

In one possible implementation, the first network data is a sampled value of data related to the target network resources in the first time slot or a statistical value from the third time slot to the first time slot, and the second network data is a sampled value of data related to the target network resources in the second time slot, and the third time slot is before the first time slot.

In a possible implementation, the global prior feature is a feature vector or a first machine learning model, and the first machine learning model is used for inference of the second network data.

In one possible implementation, the method further includes: the federated learning server receives grouping information from the first federated learning client, the grouping information from the first federated learning client indicates the group where the first local feature is located; the federated learning server obtains the global prior feature based on the first local feature and the second local feature, including: the federated learning server obtains the global prior feature based on the first local feature, the group where the first local feature is located, the second local feature, and the group where the second local feature is located, the group where the second local feature is located is indicated by the grouping information from the second federated learning client.

In one possible implementation, the first local feature includes a first sub-feature and a second sub-feature, the second local feature includes a third sub-feature and a fourth sub-feature, the grouping information from the first federated learning client indicates the group where the first sub-feature is located and the group where the second sub-feature is located, the grouping information from the second federated learning client indicates the group where the third sub-feature is located and the group where the fourth sub-feature is located, and the group where the first sub-feature is located is the same as the group where the third sub-feature is located; the federated learning server obtains the global prior feature based on the first local feature, the group where the first local feature is located, the second local feature, and the group where the second local feature is located, including: based on the fact that the group where the first sub-feature is located and the group where the third sub-feature is located are the same, the federated learning server processes the first sub-feature and the third sub-feature to obtain an intermediate feature; the federated learning server obtains the global prior feature based on the intermediate feature, the second sub-feature, the fourth sub-feature, the group where the second sub-feature is located, and the group where the fourth sub-feature is located.

In the above possible implementation, the first local feature uploaded by the first federated learning client includes the first sub-feature and the second sub-feature, the second local feature uploaded by the second federated learning client includes the third sub-feature and the fourth sub-feature, the grouping information from the first federated learning client can indicate the group where the first sub-feature is located and the group where the second sub-feature is located, and the grouping information from the second federated learning client indicates the group where the third sub-feature is located and the group where the fourth sub-feature is located. Assuming that the group where the first sub-feature is located is the same as the group where the third sub-feature is located, the federated learning server can first process the first sub-feature and the third sub-feature to obtain the intermediate feature, and then process the intermediate feature and the second sub-feature according to the grouping, and the processing result obtained after processing the fourth sub-feature to obtain the global prior feature. Features uploaded by different types of clients can be processed separately to reduce the impact between different types of clients.

In one possible implementation, the federated learning server obtains the global prior feature based on the first local feature and the second local feature, including: the federated learning server obtains the global prior feature based on the first local feature, the second local feature, the historical local feature from the first federated learning client, and the historical local feature from the second federated learning client.

In the above possible implementation manner, in addition to the first local feature and the second local feature, the federated learning server may also determine the global prior feature in combination with the historical local features from the first federated learning client and the historical local features from the second federated learning client to improve the accuracy of the global prior feature.

In one possible implementation, the federated learning server obtains the global prior feature based on the first local feature, the second local feature, the historical local feature from the first federated learning client, and the historical local feature from the second federated learning client, including: the federated learning server calculates the similarity between the local features of the current reasoning process and multiple groups of historical local features, the local features of the current reasoning process include the first local feature and the second local feature, each group of historical local features includes the historical local features from the first federated learning client and the historical local features from the second federated learning client in a historical reasoning process; the federated learning server performs weighted summation of the historical prior features corresponding to the multiple groups of historical local features based on the similarity between the local features of the current reasoning process and the multiple groups of historical local features to obtain the global prior feature.

In the above possible implementation manner, the federated learning server can calculate the similarity between the local features of the current reasoning process and multiple groups of historical local features, that is, determine the similarity between the first local feature and the historical local features from the first federated learning client in each group of historical local features, and determine the similarity between the second local feature and the historical local features from the second federated learning client in each group of historical local features, and then perform weighted summation of the historical prior features corresponding to the multiple groups of historical local features based on the similarity to obtain the required global prior features and improve the accuracy of the global prior features.

In one possible implementation, multiple groups of historical local features all have labels, and the labels are actual results of each group of historical local features manually marked; the method also includes: a federated learning server receives an inference result from a first federated learning client; the federated learning server determines a target inference result based on the inference result of the first federated learning client and the inference result of the second federated learning client; when the similarity between the local features of the current inference process and the historical local features of the target group is greater than or equal to a threshold, the federated learning server updates the historical local features of the target group based on the similarity between the local features of the current inference process and the historical local features of the target group, and the historical local features of the target group are among the multiple groups of historical local features, and the labels are the target inference results; when the similarity between the local features of the current inference process and the historical local features of the target group is less than a threshold, the federated learning server adds a group of historical local features based on the multiple groups of historical local features, and the added group of historical local features are the local features of the current inference process.

In the above possible implementation, multiple groups of historical local features may also have labels, which are the actual results of each group of historical local features manually marked, and can be used to compare with the reasoning results to determine the accuracy of the reasoning results. After obtaining the reasoning results, the client can upload them to the federated learning server. The federated learning server can determine the target reasoning result corresponding to the current global prior feature based on the reasoning results of multiple clients, and then select the target group historical local features with the label as the target reasoning result from the multiple groups of historical local features, and update the target group historical local features based on the similarity between the local features of the current reasoning process and the historical local features of the target group if the similarity is greater than or equal to the threshold, and use the target group historical local features less than the threshold as a new set of historical local features. The sample library can be updated and supplemented at any time to improve the effectiveness of the sample library.

In a possible implementation, the method further includes: the federated learning server sends task synchronization information to the first federated learning client, where the task synchronization information is used to indicate the first time slot, so that the first federated learning client selects the first network data from the local data according to the task synchronization information.

The third aspect of the present application provides an inference device that can implement the method in the first aspect or any possible implementation of the first aspect. The device includes corresponding units or modules for executing the above method. The units or modules included in the device can be implemented by software and/or hardware. The device can be, for example, a network device, or a chip, a chip system, or a processor that supports the network device to implement the above method, or a logic module or software that can implement all or part of the network device functions.

In a fourth aspect, the present application provides an inference device that can implement the method in the second aspect or any possible implementation of the second aspect. The device includes corresponding units or modules for executing the above method. The units or modules included in the device can be implemented by software and/or hardware. The device can be, for example, a network device, or a chip, a chip system, or a processor that supports the network device to implement the above method, or a logic module or software that can implement all or part of the network device functions.

In a fifth aspect, the present application provides a computer device, including: a processor, the processor is coupled to a memory, the memory is used to store instructions, and when the instructions are executed by the processor, the computer device implements the method in the first aspect or any possible implementation of the first aspect. The computer device may be, for example, a network device, or a chip or chip system that supports the network device to implement the above method.

In a sixth aspect, the present application provides a computer device, including: a processor, the processor is coupled to a memory, the memory is used to store instructions, and when the instructions are executed by the processor, the computer device implements the method in the second aspect or any possible implementation of the second aspect. The computer device may be, for example, a network device, or a chip or chip system that supports the network device to implement the above method.

The seventh aspect of the present application provides a computer-readable storage medium, which stores instructions. When the instructions are executed by a processor, the method provided by the first aspect or any possible implementation method of the first aspect, the second aspect or any possible implementation method of the second aspect is implemented.

In an eighth aspect of the present application, a chip system is provided, which includes at least one processor, and the processor is used to execute a computer program or instruction stored in a memory. When the computer program or instruction is executed on at least one processor, the method provided by the first aspect or any possible implementation method of the first aspect, the second aspect or any possible implementation method of the second aspect is implemented.

In the ninth aspect of the present application, a computer program product is provided, which includes a computer program code. When the computer program code is executed on a computer, it implements the method provided by the first aspect or any possible implementation method of the first aspect, the second aspect or any possible implementation method of the second aspect.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of the structure of a computer system provided in an embodiment of the present application;

FIG2 is a flow chart of a reasoning method provided in an embodiment of the present application;

FIG3 is a schematic diagram of an inference structure provided in an embodiment of the present application;

FIG4 is a schematic diagram of another reasoning structure provided in an embodiment of the present application;

FIG5 is a schematic diagram of another reasoning structure provided in an embodiment of the present application;

FIG6 is a schematic diagram of a calculation graph provided in an embodiment of the present application;

FIG7 is a schematic diagram of a knowledge graph provided in an embodiment of the present application;

FIG8 is a schematic diagram of another reasoning structure provided in an embodiment of the present application;

FIG9 is a schematic diagram of a method for acquiring local data of a federated learning server provided in an embodiment of the present application;

FIG10 is a schematic diagram of a flow chart of generating a fixed parameter task selector provided in an embodiment of the present application;

FIG11 is a schematic diagram of a process flow of selecting a task model provided in an embodiment of the present application;

FIG12 is a schematic diagram of a model pre-training process provided in an embodiment of the present application;

FIG13 is a schematic diagram of a flow chart of a model training process provided in an embodiment of the present application;

FIG14 is a schematic diagram of a flow chart of a model training according to a strategy provided in an embodiment of the present application;

FIG15 is a schematic diagram of the structure of an inference device provided in an embodiment of the present application;

FIG16 is a schematic diagram of the structure of another reasoning device provided in an embodiment of the present application;

FIG17 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application;

FIG. 18 is a schematic diagram of the structure of another computer device provided in an embodiment of the present application.

DETAILED DESCRIPTION

The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

The terms "first", "second" and corresponding terminology numbers in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances. This is merely a way of distinguishing objects with the same properties when describing the embodiments of the present application. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, so that a process, method, system, product or device that includes a series of units is not necessarily limited to those units, but may include other units that are not explicitly listed or inherent to these processes, methods, products or devices.

In the description of this application, unless otherwise specified, "/" means or, for example, A/B can mean A or B; "and/or" in this application is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and/or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of this application, "at least one" refers to one or more items, and "multiple items" refers to two or more items. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, c can be single or multiple.

FIG1 is a schematic diagram of the structure of a computer system applicable to an embodiment of the present application. As shown in FIG1 , the computer system includes a federated learning server and multiple clients. The federated learning server and multiple clients cooperate for model training and model-based reasoning. The number of clients can be adjusted according to actual needs.

It should be understood that during the inference phase, a client only uses local data for inference, and does not use data from other clients for inference. Although this can protect the privacy of other clients, it may result in inaccurate inference results.

To this end, an embodiment of the present application provides an inference method, in which a server participates in the inference process. Specifically, the federated learning server obtains a global prior feature based on the local features of multiple clients, and sends the global prior feature to each client, so that each client uses the global prior feature and local data for inference. Since the global prior feature is obtained based on the local features of multiple clients, and the local feature is extracted from the first network data, for each client, the information of the data of other clients is used in the inference process, which can improve the accuracy of the inference result.

The reasoning method provided in the embodiments of the present application can be applied in various scenarios, for example, it can be applied to speech recognition scenarios and image recognition scenarios.

Taking the voice recognition scenario as an example, at a concert, a person uses a language assistant to perform voice-to-text operations. A single device cannot determine whether the received voice is from the background (concert lyrics) or from the person who is recording the voice. At this time, the voice information collected by the devices of other users of the same concert is needed to determine whether the voice received by the device is from the background (concert lyrics) or from the voice being recorded. In this scenario, the reasoning method provided in the embodiment of the present application can be used to reason based on the voice information collected by the devices of multiple users to infer the source of the voice (background or user recording).

Taking the image recognition scenario as an example, there are many cars driving on a road, and the camera on each car captures the image of the road. Using the reasoning method provided in the embodiment of the present application, the images captured by the cameras of multiple cars can be used for reasoning to infer the current traffic conditions of the road (no congestion, normal congestion, or congestion caused by a traffic accident).

The data used in the reasoning method provided in the embodiment of the present application (including local data, first network data, second network data, etc. mentioned below) can be data in various forms such as images, voices, and texts.

It should be noted that a part of the model is needed in the reasoning process, and this part of the model can be obtained through pre-training. For this reason, the following will first introduce the reasoning method provided in the embodiment of the present application, and then introduce the training method of the model used in the reasoning method.

The following first introduces the reasoning method provided by the embodiment of the present application.

FIG2 is a flow chart of an inference method provided in an embodiment of the present application, which is applied to an inference system including a server and multiple clients, and specifically may be the inference system shown in FIG1 .

In this embodiment, taking one of the multiple clients as an example, the embodiment specifically includes:

Step 201. The server sends task synchronization information to multiple clients, so that the multiple clients select first network data from local data according to the task synchronization information, and the task synchronization information indicates a time requirement for the first network data.

The role of task synchronization information is to enable multiple clients to select the first network data related to the target network resource from the local data to perform reasoning under the premise of meeting user needs. Among them, the target network resource can be the current state of the target network, for example, the target network resource can be the fault state of the vehicle, then the first network data can include at least one of the following: the first vehicle's positioning data, driving speed data, driving mileage data, battery power data, battery voltage data, brake data, throttle data, motor voltage and current data, insulation resistance data and temperature data, etc., the embodiment of the present application does not limit the place.

The embodiments of the present application do not specifically limit the content of the task synchronization information. For example, the task synchronization information may be a specific moment, so that the moment of the first network data selected from the local data by the task synchronization information of multiple clients is the same, or in other words, the moment of the first network data selected from the local data by the task synchronization information of multiple clients is relatively close, or in other words, the data after dimensionality reduction of the number of data items for a period of time before the moment of the first network data selected from the local data by the task synchronization information of multiple clients, such as the average value, median value of multiple data items, or the principal component value obtained by dimensionality reduction such as principal component analysis, the cluster center value of the cluster, etc.

The first federated learning client is one of the multiple clients. The first federated learning client can receive task synchronization information from the server. The task synchronization information indicates a time requirement for first network data in local data of the first federated learning client.

Step 202: The first federated learning client selects first network data from local data according to the task synchronization information.

In this embodiment, the task synchronization information indicates a time point, and the first federated learning client can select data at the time point from local data as the first network data.

The first network data is a sampled value of data related to the target network resource in the first time slot or a statistical value from the third time slot to the first time slot, and the third time slot is before the first time slot.

For example, if the task synchronization information includes the time of 10:10, the first federated learning client can select the local data at 10:10 as the first network data according to the task synchronization information, or the first federated learning client can select the local data at a time close to 10:10 or the data after dimensionality reduction along the data dimension as the first network data according to the task synchronization information. For example, the first federated learning client can select the local data between 10:09 and 10:11 as the first network data.

Step 203: The first federated learning client extracts features of the first network data from the first network data, where the features of the first network data include first local features.

The features of the first network data may include one feature or multiple features; when the features of the first network data include multiple features, the local feature may be one or more of the multiple features.

The features of the first network data are generally represented by feature vectors. Taking the first network data including three features as an example, the features of the first network data can be represented as follows: Among them, the local feature can be one or more of these three features. Which or which features to transmit can be agreed in advance, or it can be determined based on the current first network data characteristics, the characteristic characteristics of the first network data, the client computing resources, and the client and/or server network resources. The decision strategy can be a manually written logic or a model obtained by machine learning. The embodiment of the present application is based on For example, n is the identifier of the first federated learning client, which can be 1, 2, 3, etc. Different n refers to different clients. The characteristics of the first network data are also referred to as implicit characteristics below. The characteristics of the first network data refer to the characteristics of the original data. For example, if the amount of information in the original data is small, one feature is directly transmitted. If the amount of information in the original data is large, multiple features are transmitted. The amount of information in the original data can be calculated using entropy, or it can be calculated using a neural network or machine learning model.

The first federated learning client can extract features of the first network data through a local feature extraction model, wherein there are many types of local feature extraction models, which are not specifically limited in the embodiments of the present application.

As an implementable manner, when the first network data features are of multiple types, the local feature extraction model may include multiple second task models. Accordingly, step 203 may specifically be: the first federated learning client inputs the first network data into multiple second task models respectively to obtain multiple features of the first network data output by the multiple second task models. That is, the first federated learning client inputs the first network data into different second task models according to the type, and each second task model outputs the features corresponding to the type respectively.

The embodiment of the present application does not specifically limit the type of the second task model.

As an implementable approach, each second task model is a layer of autoencoder, and the reconstruction target of the rth task model is the residual of the r-1th task model, where r is an integer greater than 1 and represents the number of second task models.

Specifically, an autoencoder is a neural network with the same input and learning target, and its structure is divided into two parts: encoder and decoder. After inputting the first network data, the implicit features output by the encoder, namely the "encoded features", can be regarded as the representation of the first network data.

Among them, the autoencoder can be implemented using a deep neural network (DNN), a convolutional neural network (CNN) or a Transformer neural network.

Specifically, take the three second task models as an example, that is, the local feature extraction model includes three layers of autoencoders. The first layer of the autoencoder input is the original feature _Xn , which is passed through the encoder Get hidden features Encoder It can be implemented using deep neural networks. Hidden features The dimension of is lower than that of the original feature _Xn . Hidden features By using the encoder The corresponding decoder Get the reconstructed features of the original features _Xn Decoder A deep neural network can be used.

The second layer of the autoencoder input is the original feature _Xn , which is passed through the encoder Get hidden features Encoder It can be implemented using deep neural networks. Hidden features The dimension of is lower than that of the original feature _Xn . Hidden features By using the encoder The corresponding decoder Get the original feature _Xn and the first layer model reconstruction The Difference Reconstruction features

The third layer of the autoencoder input is the original feature _Xn , which is passed through the encoder Get hidden features Encoder It can be implemented using a deep neural network. The dimension of the latent feature h3 is lower than that of the original feature _Xn . The latent feature h3 is obtained by the decoder D3 corresponding to the encoder E3 to obtain the original feature _Xn and the first layer model reconstruction. and layer 2 model reconstruction The difference of the sum Reconstruction features

Among them, the pre-training process of the autoencoder can be:

(1) The federated learning server sends the autoencoder model of the nth layer; (2) The client receives the autoencoder model of the nth layer, trains the model with local data for several rounds, and then uploads the autoencoder model of the nth layer to the federated learning server; (3) Repeat (1) to (2) until the autoencoder of the nth layer converges; (4) Repeat (1) to (3) to train the autoencoder of each layer layer by layer.

Step 204. The first federated learning client sends a first local feature to the server, where the first local feature is extracted from the first network data, and the first federated learning client is any one of the multiple clients. Accordingly, the server receives local features from the multiple clients, where the local features are extracted from the first network data.

In this embodiment, the first federated learning client can send the first local feature to the server after selecting it from the features of the first network data. Other clients can also extract local features from the local first network data and send them to the server. For example, the second federated learning client selects the second local feature from the features of the local first network data and sends it to the server.

Step 205: The server obtains global prior features based on local features from multiple clients.

In this embodiment, the federated learning server collects local features reported by multiple clients, and then processes these local features to determine global prior features based on the features of multiple clients. The global prior features are obtained by multi-client information fusion, which can improve the accuracy of the global prior features.

Among them, the federated learning server receives local features uploaded by multiple clients Local features uploaded by multiple clients The input features of the global prior feature extraction model are spliced together and input into the global prior feature extraction model to obtain the global prior features corresponding to each client as output.

The global prior feature extraction model uses a DNN, CNN, Transformer or RNN model. The input of the model includes the concatenated local features and the previous output value. The global prior feature extraction model can also be implemented using a lookup table, for example, using the concatenated local features as the key and the corresponding global prior features as the value.

Concatenation refers to concatenating data along one or more dimensions of a feature. For example is a 3D tensor of D×W×H. If the dimensions of W are concatenated, the concatenated dimensions are D×NW×H. It is a 3D tensor of D×W×H. If the dimensions of W and H are concatenated, the concatenated 3D tensor is of dimension D×N ₁ W×N ₂ H, where N ₁ N ₂ =N.

Since the client data only has an impact within the group and has no impact between groups, if the data of all clients are used together to determine the global prior features, it will affect the output between groups, so group input is required. As an implementable method, the first federated learning client can also send group information to the server, and the group information indicates the group where the local features are located, so that the server obtains the global prior features based on the local features from multiple clients and the groups where the local features are located.

Among them, the first local feature uploaded by the first federated learning client includes the first sub-feature and the second sub-feature, and the second local feature uploaded by the second federated learning client includes the third sub-feature and the fourth sub-feature. The grouping information from the first federated learning client can indicate the group where the first sub-feature is located and the group where the second sub-feature is located. The grouping information from the second federated learning client indicates the group where the third sub-feature is located and the group where the fourth sub-feature is located. Assuming that the group where the first sub-feature is located is the same as the group where the third sub-feature is located, the federated learning server can first process the first sub-feature and the third sub-feature to obtain the intermediate feature, and then process the intermediate feature and the second sub-feature according to the grouping, and the processing result obtained by processing the fourth sub-feature to obtain the global prior feature. Features uploaded by different types of clients can be processed separately to reduce the impact between different types of clients.

Specifically, the grouping information is information about the grouping between clients. The client within the group corresponds to the client with the same land nature (farmland, road, town), and the client between the groups corresponds to the client with different land nature. Different land natures have different effects on signal transmission, so grouping is performed to reduce the impact between clients of different lands. For example, on a highway, the data of vehicles (1, 2, 3) in the same direction or the same lane are the same group, and the data of vehicles (4, 5, 6) in different directions are another group. Then the grouping information uploaded by vehicles (1, 2, 3) indicates the same group, and the grouping information uploaded by vehicles (4, 5, 6) indicates the same group.

The federated learning server receives local features uploaded by multiple clients And the corresponding group information The grouping information can be obtained by Clustering is performed separately.

Will Same Splice together and use different models according to the global prior feature extraction model, such as DNN, CNN, RNN, Transformer, corresponding to different splicing methods:

For models such as DNN and CNN, where input data needs to be input simultaneously, The global positions are stitched together, and the positions not selected by the group are replaced by default values, as shown in the following table. The global position is shown in Table 1 below:

Table 1

if and Belong to the same group; A grouping; If there is one group, the data is concatenated as shown in Table 2, Table 3 and Table 4:

Table 2

Table 3

Table 4

If RNN and Transformer models are used, the input data can be fed into the model one by one. Add the global position code and then input it into the model. The global position code is shown in the following table.

The sequence of inputs to the model is

The spliced local features are input into the first layer model of the global prior feature extraction model to obtain the implicit features of the first layer. Represents the implicit features of the mth group in the first layer.

Extract the implicit features of the first layer model output by the global prior feature extraction model With local features Perform splicing as shown in Table 5 below:

Table 5

The spliced local features are grouped in the first layer and input into the second layer model of the global prior feature extraction model to obtain the second layer of each group The second layer of grouping can be regrouping based on the first layer of grouping, or all local features can be regrouped.

Extract the implicit features of the second layer model output by the global prior feature extraction model With local features Perform splicing, and input the splicing result into the third layer model of the global prior feature extraction model to obtain the global prior features corresponding to each client.

Step 206: The federated learning server sends the global prior features to the multiple clients respectively, and correspondingly, the first federated learning client receives the global prior features.

In this embodiment, after determining the global prior feature, the federated learning server can send the global prior feature to multiple connected clients.

Step 207: The first federated learning client performs reasoning based on the global prior features and the second network data to obtain a reasoning result.

In this embodiment, the first federated learning client can infer the local second network data based on the global prior feature to obtain the inference result of the second network data. Among them, the second time slot is the same as the first time slot or after the first time slot, that is, the second network data can be at the same time as the first network data, or it can be data after the time of the first network data. For example, assuming that the first network data is 9:00 data, the second network data can be data at any time between 9:00 and 9:15, and this embodiment of the application is not limited to this.

During the inference process, the second network data may be directly processed, or the implicit features extracted from the second network data may be inferred, or one or more features may be selected from the implicit features for inference, which is not limited in the embodiments of the present application.

The global prior feature is a feature vector or an inference model, and the inference model is used to output an inference result of the second network data according to the second network data.

For the scenario where the global prior feature is a feature vector, the first federated learning client performs reasoning based on the global prior feature, the second network data, and the local second machine learning model to obtain the reasoning result, and the second machine learning model is used for the reasoning of the second network data. Specifically, the first federated learning client needs to input the global prior feature and the second network data into the local second machine learning model for reasoning, and use the result output by the second machine learning model as the reasoning result.

Among them, before inputting the second network data into the second machine learning model, the first federated learning client can also input the second network data into a third machine learning model (such as a local feature extraction model) to obtain multiple features that can reflect the characteristics of the second network data, which can save computing resources. The first federated learning client inputs the global prior features and the second network data into the local second machine learning model for reasoning. The first federated learning client inputs the global prior features into the second machine learning model, and can obtain the weights corresponding to the multiple features of the second network data output by the second machine learning model, that is, each feature corresponds to a weight, and then the reasoning result can be determined according to the multiple features of the second network data and the weights of the multiple features of the second network data.

FIG3 is a schematic diagram of an inference structure provided in an embodiment of the present application. As shown in FIG3 , the client calculates weight values corresponding to different implicit features based on global prior features. Specifically, the local feature extraction model adopts a multi-layer autoencoder.

Client _Cn receives the global prior features sent by the federated learning server The inference model (the second machine learning model) uses the DNN model. The client C _n converts the global prior features Input the inference model, and the inference model outputs multiple implicit features The corresponding weight value And save the weight value

Client _Cn obtains the local data to be inferred X(t) _n at time t through the collector, and extracts multiple implicit features output by the local feature extraction model from the data to be inferred X(t) _n Extract multiple implicit features from the local feature extraction model output The weight value output by the inference model and local anchor datasets Calculate the distance r(t) _n,m from the classification to each anchor point, where

Select y n, _{m corresponding to the smallest value of r(t) n, m} corresponding to all anchor point data sets as the output of X(t) _n .

Calculate multiple latent features output by the local feature extraction model The corresponding classification results are used as the evaluation indicators of output stability, namely

Select all anchor point datasets corresponding to The smallest value corresponding to y _n,m is used as the hidden feature of X(t) _n The categories of all anchor datasets are selected. The smallest value corresponding to y _n,m is used as the hidden feature of X(t) _n The categories of all anchor datasets are selected The smallest value corresponding to y _n,m is used as the hidden feature of X(t) _n Category.

The evaluation index of output stability = maximum number of the same category/total number of hidden feature vectors.

If the three categories are consistent, the stability evaluation index = 1, if two categories are consistent, the stability evaluation index = 2/3, and if none of the categories are consistent, the stability evaluation index = 1/3.

The second machine learning model described above is a task model, which does not need to process the second network data. The following describes the way in which the task model needs to process the second network data, wherein the second machine learning model includes multiple first task models, which can process multiple features of the second network data separately. Specifically, after the first federated learning client obtains the global prior features, it can determine the weights of the multiple first task models from the global prior features. The first federated learning client can input the features of the second network data into the multiple first task models. The input features of the multiple first task models can be features of the default type, that is, the features of the second network data are classified according to categories and input into the first task models of the corresponding categories. The first federated learning client can perform weighted averaging on the inference features output by the multiple first task models in combination with the weights corresponding to the multiple first task models to obtain the inference result.

Among them, the way in which the first federated learning client determines the respective weights of multiple first task models based on the global prior features can be that the first federated learning client needs to combine the global prior features and the second network data to calculate the respective weights of the multiple first task models, that is, to calculate the weights in combination with the features of the data to be inferred, thereby improving the accuracy of the weights.

Among them, before inputting the second network data into the second machine learning model, the first federated learning client can also input the second network data into a third learning model (such as a local feature extraction model) to obtain multiple features that can reflect the characteristics of the second network data, thereby saving computing resources.

FIG4 is a schematic diagram of another reasoning structure provided in an embodiment of the present application. Referring to FIG4 , the client C _n receives the global prior feature information sent by the federated learning server. And save the global prior features The reasoning model adopts a moderated expert model, and the hybrid expert model includes a model selector and a plurality of first task models, such as task model 1 to task model N in the figure.

Client _Cn obtains the local data to be inferred X(t) _n at time t through the collector. The inference model (the second machine learning model) adopts a hybrid expert model to combine the data to be inferred X(t) _n and the global prior features. Input the hybrid expert model to calculate the inference results.

Among them, the hybrid expert model is composed of a model selector, a task model group composed of multiple task models and a classifier: the model selector input is the global prior feature (the second network data can also be added), and the output is the weight of multiple task models in the "task model group" (the selection can be a 0-1 weight); the "task model group" is composed of N "task models", and each "task model" outputs an implicit feature vector based on the input data; the classifier input is the implicit feature vector after the weighted summation of the weights output by the model selector, and the output is the classification result.

The specific steps are as follows:

Step 1: The global prior features (the second network data can also be added) calculate the weight values of multiple task models through the "model selector".

The global prior features are first input into the model selector model (the model can be implemented by DNN, CNN, Transformer), and the model parameters are The input-output relationship of the model selector model is expressed as: or

Then calculate the distance from g(t)′ to each anchor point The distance can use Euclidean distance, cosine distance or custom distance. The anchor point set of the model selector model is:

in Represents the kth anchor point corresponding to the input mth task model. Represents the anchor point set corresponding to the mth task model.

Then select the best distance from each anchor point to g(t)′ in the set of anchor points corresponding to each task model. The optimal distance is related to the specific method of calculating the distance. If Euclidean distance is used, the optimal distance is the reciprocal of the minimum Euclidean distance; if cosine distance is used, the optimal distance is the maximum cosine distance.

Finally, the weight vector g = [g ₁ g ₂ …g _n,m …g _m,M ] output by the “model selector” is calculated, where g _n,m represents the weight of the mth task network.

Step 2: Calculate the implicit feature extraction value of the task model selected by the "model selector" after the local original features. The parameters of the mth task network are Indicates that the input-output relationship of the mth task network is expressed as:

Step 3: The "weighted summer" performs weighted summation on the M implicit feature vectors of the "task model group" using the M weights output by the "model selector" (the implicit features of the unselected task models do not participate in the calculation).

Step 4: The classifier is implemented using DNN, and the model parameters are The input-output relationship is:

In this embodiment, the global prior features sent by the federated learning server can also be directly the model selector in the hybrid expert model, which is not limited here, and the local feature extraction model can also be a hybrid expert model. The local feature extraction model (third machine learning model) can include multiple second task models, and the first federated learning client can extract the features of the second network data through the third machine learning model in the following way: the first federated learning client first determines the weights corresponding to the multiple second task models based on the local data or the second network data, and then inputs the second network data into the second task model, and the sub-features of the second network data output by the second task model can be obtained. The first federated learning client can perform weighted averaging based on the weights corresponding to the multiple second task models and the sub-features output by the second task model to obtain the features of the second network data.

FIG5 is another schematic diagram of a reasoning structure provided in an embodiment of the present application. Please refer to FIG5 . The federated learning server generates task synchronization information and sends the task synchronization information to multiple clients.

The client _Cn receives the task synchronization information sent by the federated learning server, and finds data _Xn that meets the task synchronization information from the local data. Among them, meeting the task synchronization information means that the collection time of data _Xn is closest to the time of the task synchronization information.

Client _Cn obtains implicit features through the local feature extraction model in As the local feature of client C _n . The local feature extraction model is implemented by a hybrid expert model, which consists of multiple task models i and a model selector. The data is extracted by selecting the task model through the model selector to obtain the local feature

Task model i is trained using decoupled representation learning, that is, the local features output by the task model It can be divided into K groups, the features within the group are similar, and the features between the groups are specific. Different neural network structures get The types are different, so It can be a scalar, a vector, a distance or a tensor, as follows:

The local feature extractor uses a fully connected neural network or RNN, such as but is a scalar, where W1 and W2 are model parameters and is a matrix; if exist is a vector, where and W1 are model parameters, which are matrices. Introducing the expression of tensor calculation W2 is a three-dimensional tensor, ° is a tensor calculation; W2 can also be a higher tensor, in this case What you get Can be a matrix or a tensor.

The local feature extractor uses a convolutional neural network or a Transformer neural network. is a matrix. Taking an image as an example, the image is obtained after passing through the convolution layer of CNN. at this time It represents the feature map of the mth channel corresponding to the convolution kernel m, which is a matrix. If the data in a channel is averaged, then It's a scalar.

The characteristics within a group are similar, and the characteristics between groups are specific. Similarity and specificity can be measured by and The distance between them is obtained. If m1 and m2 belong to the same feature group, their distance value is small. If they belong to different feature groups, their distance value is large.

Client C _n upload To the federated learning server, the federated learning server receives local features uploaded by multiple clients. The federated learning server can store local features uploaded by multiple clients. The input features of the global prior feature extraction model are spliced and input into the global prior feature extraction model, and the global prior features are output. The global prior features are the task selectors of the inference models of each client. The global prior features (task selectors of the inference model) corresponding to each client are sent to the corresponding client C _n .

The client _Cn receives the global prior features (task selector of the inference model) sent by the federated learning server, obtains the local data to be inferred X(t) _n at time t through the collector, obtains the local features h(t) _{n from the data to be inferred X(t)n ,} and inputs the local features h(t) _n into the inference model to obtain the inference result.

The reasoning model adopts a hybrid expert model, which is composed of a model selector and a task model. The model selector is a global prior feature (task selector of the reasoning model). Each task model corresponds to a feature group of local features h(t) _n , and the features in each feature group have similarity. That is, the input of task model 1 is the first group of features of h(t) _n , the input of task model 2 is the second group of features of h(t) _n , and the input of task model 2 is the nth group of features of h(t) _n .

The global prior feature can also be directly an inference model, such as a classifier or regressor with model parameters. Where z represents the input of the classifier or regressor, Represents the model parameters of a classifier or regressor. The classifiers or regressors of each client can be the same or different. Can be

FIG6 is a schematic diagram of a calculation graph provided in an embodiment of the present application. As shown in FIG6, the global prior feature can be a calculation graph with a set of feature values and a calculation logic relationship. For example, represents the kth feature vector of the nth client. The feature vector can be learned through data during the training process as part of the model parameters, or it can be a manually set value; _Cn represents the computational flow graph of the nth client. The computational graphs of each client can be the same or different. The computational graph can be obtained through data training or through knowledge graphs.

The left figure shows the distance between the input feature vector v and the feature value in the calculation graph, that is, v and Calculate the cosine distance for each value in

Then calculate the following logical operation

The right picture shows the cosine distance after calculation. The value greater than the threshold is 1, and the value less than the threshold is 0.

Then calculate the logical operation. The addition here corresponds to the OR operation in logic, and the multiplication corresponds to the AND operation in logic:

The global prior feature can be a knowledge graph, taking image recognition as an example. The local feature extractor is trained by decoupled representation learning, that is, the implicit features obtained by the local feature extractor can be grouped into different self-concept learning, such as wheels, doors, windows, lights, human eyes, human noses, human mouths, etc., that is, the implicit features obtained by the local feature extractor are h = {h(1) = wheels, h(2) = doors, h(3) = windows, h(4) = lights, h(5) = human eyes, h(6) = human noses, h(7) = human mouths}.

FIG7 is a schematic diagram of a knowledge graph provided by an embodiment of the present application. As shown in FIG7, at this time, the local feature extractor not only calculates the value of the feature but also calculates the position of the feature when participating in reasoning. Activation indicates that the sum of the values of a certain channel of the local feature extractor or the combination of channels or a pixel value in a channel is greater than a threshold.

When the client directly uses the trained local feature extraction model and inference model:

Client _Cn receives the global prior features sent by the federated learning server and calculates the graph or classifier/regressor Or knowledge graph, save the calculation graph locally or classifier/regressor Or a knowledge graph.

Client _Cn obtains the local data to be inferred X(t) _n at time t through the collector, and extracts multiple implicit features output by the local feature extraction model from the data to be inferred X(t) _n. Then Input computation graph or classifier/regressor Or the knowledge graph gets the output result.

When the client still needs to retrain the inference model: the first federated learning client uses sample data to train the first machine learning model, extracts the features of the second network data, and then inputs the features of the second network data into the trained first machine learning model to obtain the inference result output by the trained first machine learning model.

Specifically, the client _Cn receives the global prior features sent by the federated learning server and calculates the graph or classifier/regressor Or knowledge graph, initialize local inference local feature extraction model parameters, use or classifier/regressor Or the knowledge graph is used as a classifier or regressor to form an inference model, and then the inference model is trained using local data. And save the inference model after training The local data here refers to historical data and corresponding historical reasoning results.

Client _Cn obtains the local data to be inferred X(t) _n at time t through the collector. The data to be inferred X(t) _n is calculated through the inference model Get the output result.

In addition to the first local feature and the second local feature, the federated learning server can also determine the global prior feature by combining the historical local features from the first federated learning client and the historical local features from the second federated learning client. The federated learning server can calculate the similarity between the local features of the current reasoning process and multiple groups of historical local features, that is, determine the similarity between the first local feature and the historical local features from the first federated learning client in each group of historical local features, and determine the similarity between the second local feature and the historical local features from the second federated learning client in each group of historical local features, and then perform weighted summation of the historical prior features corresponding to the multiple groups of historical local features based on the similarity to obtain the required global prior features.

In a feasible implementation, multiple groups of historical local features may also have labels, which are the actual results of each group of historical local features manually marked, and can be used to compare with the reasoning results to determine the accuracy of the reasoning results. After obtaining the reasoning results, the client can upload them to the federated learning server. The federated learning server can determine the target reasoning result corresponding to the current global prior feature based on the reasoning results of multiple clients, and then select the target group historical local features with the label as the target reasoning result from the multiple groups of historical local features. If the similarity between the local features of the current reasoning process and the historical local features of the target group is greater than or equal to a threshold, the target group historical local features can be updated according to the similarity between the local features of the current reasoning process and the historical local features of the target group. If the similarity between the local features of the current reasoning process and the historical local features of the target group is less than the threshold, the local features of the current reasoning process can be saved locally as a new group of historical local features.

FIG8 is a schematic diagram of another reasoning structure provided in an embodiment of the present application. Referring to FIG8 , the federated learning server receives local features uploaded by multiple clients. Compute local features through global prior models With the sample library The similarity of The corresponding hg(m) is weighted summed as The global prior features.

The sample library of the federated learning server is represents the historical local feature set corresponding to the m-th sample data, and hg(m) represents the historical global prior feature corresponding to the m-th sample data.

Similarity can be measured using Euclidean distance, cosine distance, or trained neural network. Take cosine distance as an example:

The weighted sum is:

The federated learning server sends the global prior feature hg to each client. Each client obtains the data to be inferred and combines the global prior feature hg to obtain the local inference result. Then the inference result is uploaded to the federated learning server. If the client has the post-observation annotation function, the post-annotated label is uploaded to the federated learning server.

The federated learning server receives the inference results or labeled labels uploaded by the client and counts them. The corresponding class y has the largest number of categories, from Find all samples with the same category, that is, y(m) is the same as y. Calculate The similarity of samples of the same class as the category, such as Euclidean distance, cosine distance or neural network. For sample data with similarity greater than the threshold, the corresponding hg(m) is updated according to the similarity. If all selected samples do not meet the threshold, then As a new example.

Take the cosine distance as an example:

Similarity:

Update of hg(m): hg(m)=r(m)*hg+(1-r(m))*hg(m).

After obtaining the inference results and manually annotated labels, the client can also train the global prior model, local feature extraction model, and inference model in Figure 3. Specifically, each client calculates the error of the inference result and the label, uses error back propagation to calculate the error of the global prior feature, and uploads the error of the global prior feature to the federated learning server. The federated learning server receives the error of the global prior feature uploaded by each client, uses error back propagation to calculate the error of the local feature, and sends it to each client. Each client and the federated learning server use their own error calculation model gradients and update the model. Among them, the client updates the inference model according to the error of the inference result and the label, updates the local feature extraction model according to the error of the local feature, and the federated learning server updates the global prior local feature extraction model according to the error of the global prior feature.

The federated learning server may also save partial data of multiple clients, use the partial data as local data of the federated learning server, and train an initial inference model based on the local data of the federated learning server, which is not specifically limited here.

FIG9 is a schematic diagram of a method for obtaining local data of a federated learning server provided in an embodiment of the present application. As shown in FIG9 , the federated learning server uses local data to train a "difference judgment model" and sends it to the client side. The client side uses the "difference judgment model" sent by the federated learning server to calculate the difference index, and samples the difference index, and uploads the sampled difference index to the federated learning server. The federated learning server collects the difference index for clustering to obtain the global difference index class center, and then sends the global difference index class center to the client.

The client side receives the difference indicator type center sent by the cloud side, classifies the local data on the client side into various centers and counts the classification results, and counts the number of data in each center, as well as the number of data far away from the center and the number of important data, and uploads the statistical results to the federated learning server. The local data on the client side includes ordinary data and important data. The important data can be intrusion data, wrong data, inconsistent voting data, or data far from the anchor point, which is not limited here.

The federated learning server receives the statistical results and generates a collection strategy based on the local data (which can be different for each client side), and sends it to each client.

The client receives the collection strategy sent by the federated learning server and collects data, samples the collected data, and uploads the sampled data to the federated learning server. The federated learning server uses active learning to label the uploaded data and sends it to each client.

The federated learning server uses local data to train a labeling model and sends it to the client. The client uses the labeling model to label local data, and then trains the local inference model based on the labeled local data and the labeled data sent by the federated learning server.

The federated learning server can also compress the local data of the federated learning server, such as using a training data generator to compress the data. The federated learning server can also train an inference model based on local data as the initial inference model of the client, which will not be described in detail here.

For the training of the hybrid expert model in Figure 4, it is necessary to perform training in the manner of fixed parameter task selector generation, selective task model delivery, model pre-training, model training, and model strategy training.

FIG10 is a schematic diagram of a flow chart of generating a fixed parameter task selector provided in an embodiment of the present application. For details, please refer to FIG10 :

Step 1001. The federated learning server generates K1 cluster center information according to categories.

Step 1002: The federated learning server sends K1 class center information to the client.

Step 1003. The client receives K1 cluster center information, classifies K types of data into K1 cluster center information according to the cluster center information, counts the mean and number of data belonging to each cluster center, and uploads them to the federated learning server.

Step 1004: The client uploads the mean and number of data belonging to each cluster center to the federated learning server.

Step 1005: The federated learning server calculates the average value of the data belonging to each global cluster center to obtain new cluster center information.

Step 1006. The federated learning server determines whether the new cluster center information has converged. If not, go to step 1002. If converged, go to step 1007. The convergence condition may be that the change in the new cluster center information compared with the previous cluster center information is less than a threshold, or the number of iterations is reached.

Step 1007: The federated learning server sends new K1 cluster center information.

Step 1008: The client counts the number of K1 cluster center information corresponding to the data of the K classes.

Step 1009: The client uploads the data of K classes corresponding to the number of K1 cluster center information to the federated learning server.

Step 1010: The federated learning server receives the K classes of data uploaded by each client, and the corresponding numbers of K1 cluster center information, and generates a fixed parameter task selector rule.

Exemplary, Solution 1: For scenarios requiring privacy protection:

The data of K classes uploaded by client n correspond to the number of K1 cluster center information in the matrix A _n :

Indicates that the k-th class uploaded by the n-th client belongs to the k1-th cluster center information.

Step 1: Calculate the global K category data corresponding to K1 cluster center information A:

in a ^k =[ ^ak,1 … ^ak,k1 ] represents the global data volume of K1 cluster center information corresponding to the kth class.

Step 2: For each category, select the cluster number corresponding to the cluster center information corresponding to the topK1 with the largest number as the expert selected for that category.

Among them, there are topK1 values in b ^k = [b ^k,1 …b ^k,k1 ] that are not zero, and the non-zero positions in b ^k are the positions corresponding to the topK1 in a ^k . The values of the non-zero positions in b ^k can be 1 or the weighted value of the data volume, and the weighted value is:

On this basis, in order to meet the task balance among experts, it is set that each expert can use topK2 times at most. When the number of times an expert is selected exceeds topK2, the expert will no longer be selected when selecting topK1. The order of expert selection can be sorted by the amount of data selected by the experts, starting with the experts with the largest amount of data. For example, if the value of a ^k,k1 is the largest, then the expert is selected starting from the kth class.

B is used as a fixed parameter task selector rule.

Solution 2: For the scenario where the number of experts selected by each user is as small as possible as the optimization goal:

Step 1: Normalize the data belonging to different cluster center information in each category of _An to obtain the probability that each category of data belongs to different cluster center information:

in

Step 2: Cluster the K3 cluster center information, and the cluster center information belonging to the same cluster center information In a group.

Step 3: Same as Solution 1, except that the number of experts in each group is limited to not more than topK4. If it exceeds, only the experts with the largest amount of data among the experts corresponding to topK4 are selected as the experts for the corresponding category.

Step 4: For inconsistent categories and experts between groups, the matching relationship between categories and experts is selected in the way with the most matches. For example, if there are 10 groups, if expert 2 has the most groups, then expert 2 is selected.

FIG11 is a schematic diagram of a process flow of selecting a task model provided in an embodiment of the present application. Please refer to FIG11 :

Step 1101. The federated learning server sends down fixed parameter task selector rules.

Step 1102: The client receives the fixed parameter task selector rule and counts the task model ID number corresponding to the local data.

Step 1103. The client uploads the task model ID to the federated learning server.

Step 1104. The federated learning server matches each client task model ID and matches the corresponding task model and the task model selector with trainable parameters.

FIG12 is a schematic diagram of a model pre-training process provided in an embodiment of the present application. Please refer to FIG12 :

Step 1201. The federated learning server sends a task model and a task model selector with trainable parameters.

Step 1202. The client receives a task model selector with trainable parameters and counts the task model ID number corresponding to the local data.

Step 1203: The client uploads the task model ID to the federated learning server.

Step 1204. The federated learning server matches each client task model ID and matches the corresponding task model.

Step 1205: The federated learning server sends the task model.

Step 1206. The client uses local data and the task model of the federated learning server as the initial value of the client's task model, selects the task model through a fixed parameter task selector, and trains the client's task model for several rounds.

Step 1207. The client uses local data and takes the parameter trainable task model selector of the federated learning server as the initial value of the parameter trainable task model selector of the client, pseudo-labels the data through the fixed parameter task selector, and then trains the parameter trainable task model selector for several rounds.

Step 1208: The client uploads the updated client task model and the task model selector with trainable parameters.

Step 1209. The federated learning server receives the updated task models and parameter-trainable task model selectors of each client, and obtains the task model and parameter-trainable task model selector of the federated learning server by weighted average.

Step 1210. Repeat steps 1201 to 1209 several times.

FIG13 is a schematic diagram of a model training process provided in an embodiment of the present application. Please refer to FIG13 :

Step 1301. The federated learning server sends a parameter trainable task model selector of the federated learning server.

Step 1302: The client receives the parameter-trainable task model selector from the federated learning server and counts the task model ID number corresponding to the local data.

Step 1303: The client uploads the task model ID number to the federated learning server.

Step 1304: The federated learning server receives each client's task model ID and matches the corresponding task model of the federated learning server.

Step 1305: The federated learning server sends the corresponding task model of the federated learning server.

Step 1306. The client uses local data and the task model of the federated learning server as the initial model, selects the task model through the parameter trainable task model selector of the federated learning server, and trains the client's task model for N rounds.

Step 1307: The client uploads the updated client task model to the federated learning server.

Step 1308: The federated learning server receives the task models of the clients updated by each client, and obtains the task model of the federated learning server by weighted average.

Step 1309: The federated learning server sends the updated task model of the federated learning server to the client.

Step 1310. The client uses local data, takes the task model of the federated learning server as the task model of the client, takes the parameter trainable task model selector of the federated learning server as the initial value of the parameter trainable task model selector of the client, selects the task model through the parameter trainable task model selector of the client, and trains the parameter trainable task model selector of the client for N rounds.

Step 1311. The client uploads the updated client's parameter trainable task model selector.

Step 1312: The federated learning server receives each updated client parameter trainable task model selector, and obtains a federated learning server parameter trainable task model selector by weighted average.

Step 1313. Repeat steps 1301 to 1312 several times.

FIG14 is a schematic diagram of a flow chart of a model training according to a strategy provided in an embodiment of the present application. Please refer to FIG14 :

Step 1401. The federated learning server sends the federated learning server's parameter trainable task model selector.

Step 1402. The client receives the parameter trainable task model selector from the federated learning server, and counts the number of each type of data processed by each task model of the local data.

Step 1403. The client uploads the statistical results to the federated learning server.

Step 1404. The federated learning server receives the number of each type of data processed by each task model of each client, and counts the number of each type of data processed by each task model globally.

Step 1405. The federated learning server generates the correspondence between the task model and various types of data.

Step 1406. The federated learning server sends the correspondence between the task model and various types of data.

Step 1407. The client uses local data and the task model of the federated learning server as the initial value of the client's task model, selects the client's task model as the student through the correspondence between the task model and various types of data, and uses the task model of the federated learning server as the teacher model, and trains the client's task model for N rounds through knowledge distillation.

Step 1408. The client uses local data and takes the parameter-trainable task model selector of the federated learning server as the initial value of the parameter-trainable task model selector of the client, and trains the parameter-trainable task model selector L6 rounds by pseudo-labeling the data through the correspondence between the task model and various types of data.

Step 1409: The client uploads the updated client task model and the task model selector with trainable parameters.

Step 1410: The federated learning server receives each updated client parameter trainable task model selector, and obtains a federated learning server parameter trainable task model selector by weighted average.

Step 1411. Repeat steps 1401 to 1410 several times.

In an embodiment of the present application, a first federated learning client collects first network data related to a target network resource in a first time slot, extracts a first local feature of the first network data and sends it to a federated learning server. The federated learning server collects local features uploaded by multiple clients to calculate a global prior feature and sends it to each client. The first federated learning client can infer the second network data in a second time slot based on the global prior feature. During the inference process, information on data of other clients in addition to local data is used to improve the accuracy of the inference results.

The above describes the reasoning method, and the following describes the device for executing the method.

FIG. 15 is a schematic diagram of the structure of an inference device provided in an embodiment of the present application. Referring to FIG. 15 , the device 150 includes:

The transceiver unit 1501 is used to send a first local feature to the federated learning server, where the first local feature is extracted from the first network data, where the first network data is data related to the target network resource acquired by the first federated learning client in the first time slot, where the target network resource is a network resource managed by the first federated learning client, and receive a global prior feature from the federated learning server, where the global prior feature is obtained based on the first local feature and the second local feature, where the second local feature is provided by the second federated learning client;

Processing unit 1502 is used to perform reasoning based on the global prior features and the second network data to obtain an inference result, where the second network data is data related to the target network resources obtained by the first federated learning client in the second time slot, and the inference result is used to manage the target network resources, wherein the second time slot is the same as the first time slot or after the first time slot.

The transceiver unit 1501 is used to execute step 204 and step 206 in the method embodiment of FIG. 2 , and the processing unit 1502 is used to execute step 207 in the method embodiment of FIG. 2 .

Optionally, the first network data is a sampling value of data related to the target network resources in the first time slot or a statistical value from the third time slot to the first time slot, and the second network data is a sampling value of data related to the target network resources in the second time slot, and the third time slot is before the first time slot.

Optionally, the global prior feature is a feature vector or a first machine learning model, and the first machine learning model is used for inference of the second network data.

Optionally, when the global prior feature is a feature vector, the processing unit 1502 is specifically configured to:

Reasoning is performed based on the global prior features, the second network data, and the local second machine learning model to obtain an inference result, and the second machine learning model is used for reasoning about the second network data.

Optionally, the processing unit 1502 is specifically used to: input the second network data into the third learning model to obtain multiple features of the second network data output by the third learning model; input the global prior features into the second machine learning model to obtain the respective weights of the multiple features of the second network data; determine the inference result based on the multiple features of the second network data and the respective weights of the multiple features of the second network data.

Optionally, the second machine learning model includes multiple first task models; the processing unit 1502 is specifically used to: calculate the weights of each of the multiple first task models based on global prior features; input the features of the second network data into the multiple first task models to obtain inference features output by the multiple first task models; obtain inference results based on the weights of each of the multiple first task models and the inference features output by the multiple first task models.

Optionally, the processing unit 1502 is specifically configured to calculate the weights of the plurality of first task models according to the global prior features and the second network data.

Optionally, the processing unit is also used to: extract features of the second network data through a third machine learning model.

Optionally, the third machine learning model includes multiple second task models; the processing unit 1502 is specifically used to: determine the weights of each of the multiple second task models based on the second network data; input the second network data into the multiple second task models to obtain sub-features of the second network data output by the multiple second task models; obtain features of the second network data based on the weights of each of the multiple second task models and the sub-features of the second network data output by the multiple second task models.

Optionally, each second task model is a layer of autoencoder, and the reconstruction target of the rth task model among the multiple second task models is the residual of the r-1th task model, where r is an integer greater than 1 and represents the number of second task models.

Optionally, when the global prior feature is the first machine learning model, the processing unit 1502 is specifically used to: extract the features of the second network data; input the features of the second network data into the first machine learning model to obtain the inference result output by the first machine learning model.

Optionally, when the global prior feature is the first machine learning model, the processing unit 1502 is specifically used to: train the first machine learning model using sample data; extract features of the second network data; input the features of the second network data into the trained first machine learning model to obtain the inference result output by the trained first machine learning model.

Optionally, the transceiver unit 1501 is also used to: send grouping information to the federated learning server, the grouping information indicating the group where the first local feature is located, so that the federated learning server obtains the global prior feature based on the first local feature, the group where the first local feature is located, the second local feature, and the group where the second local feature is located.

Optionally, the transceiver unit 1501 is further used to: receive task synchronization information from the federated learning server, where the task synchronization information is used to indicate the first time slot; and select the first network data from the local data according to the task synchronization information.

FIG. 16 is a schematic diagram of the structure of another inference device provided in an embodiment of the present application. Referring to FIG. 16 , the device 160 includes:

The transceiver unit 1601 is used to receive a first local feature from a first federated learning client, where the first local feature is extracted from first network data, where the first network data is data related to a target network resource acquired by the first federated learning client in a first time slot, where the target network resource is a network resource managed by the first federated learning client;

A processing unit 1602 is configured to obtain a global prior feature according to the first local feature and the second local feature, where the second local feature is provided by the second federated learning client;

The transceiver unit 1601 is used to send the global prior feature to the first federated learning client, so that the first federated learning client performs reasoning based on the global prior feature and the second network data to obtain an inference result, where the second network data is data related to the target network resources obtained by the first federated learning client in the second time slot, and the inference result is used to manage the target network resources, wherein the second time slot is the same as the first time slot or after the first time slot.

The transceiver unit 1601 is used to execute step 204 and step 206 in the method embodiment of FIG. 2 , and the processing unit 1602 is used to execute step 205 in the method embodiment of FIG. 2 .

Optionally, the first network data is a sampling value of data related to the target network resources in the first time slot or a statistical value from the third time slot to the first time slot, and the second network data is a sampling value of data related to the target network resources in the second time slot, and the third time slot is before the first time slot.

Optionally, the global prior feature is a feature vector or a first machine learning model, and the first machine learning model is used for inference of the second network data.

Optionally, the transceiver unit 1601 is further used to: receive grouping information from the first federated learning client, where the grouping information from the first federated learning client indicates the group where the first local feature is located;

The processing unit 1602 is specifically used to obtain a global prior feature according to the first local feature, the group where the first local feature is located, the second local feature, and the group where the second local feature is located, wherein the group where the second local feature is located is indicated by the group information from the second federated learning client.

Optionally, the first local feature includes a first sub-feature and a second sub-feature, the second local feature includes a third sub-feature and a fourth sub-feature, the grouping information from the first federated learning client indicates a group where the first sub-feature is located and a group where the second sub-feature is located, the grouping information from the second federated learning client indicates a group where the third sub-feature is located and a group where the fourth sub-feature is located, and the group where the first sub-feature is located is the same as the group where the third sub-feature is located;

The processing unit 1602 is specifically used for: based on the fact that the group where the first sub-feature is located is the same as the group where the third sub-feature is located, the federated learning server processes the first sub-feature and the third sub-feature to obtain an intermediate feature; and obtains a global prior feature according to the intermediate feature, the second sub-feature, the fourth sub-feature, the group where the second sub-feature is located, and the group where the fourth sub-feature is located.

Optionally, the processing unit 1602 is specifically used to obtain a global prior feature according to the first local feature, the second local feature, the historical local feature from the first federated learning client, and the historical local feature from the second federated learning client.

Optionally, the processing unit 1602 is specifically used to: calculate the similarity between the local features of the current reasoning process and multiple groups of historical local features, the local features of the current reasoning process include first local features and second local features, each group of historical local features includes historical local features from the first federated learning client and historical local features from the second federated learning client in a historical reasoning process; according to the similarity between the local features of the current reasoning process and multiple groups of historical local features, perform weighted summation of historical prior features corresponding to the multiple groups of historical local features to obtain global prior features.

Optionally, multiple groups of historical local features all have labels, and the labels are actual results of each group of historical local features manually marked; the processing unit 1602 is also used to: receive inference results from the first federated learning client; determine the target inference results based on the inference results of the first federated learning client and the inference results of the second federated learning client; when the similarity between the local features of the current inference process and the historical local features of the target group is greater than or equal to a threshold, update the historical local features of the target group according to the similarity between the local features of the current inference process and the historical local features of the target group, and the historical local features of the target group are among the multiple groups of historical local features, and the labels are the target inference results; when the similarity between the local features of the current inference process and the historical local features of the target group is less than a threshold, add a group of historical local features on the basis of the multiple groups of historical local features, and the added group of historical local features are the local features of the current inference process.

Optionally, the transceiver unit 1601 is further used to: send task synchronization information to the first federated learning client, where the task synchronization information is used to indicate the first time slot, so that the first federated learning client selects the first network data from the local data according to the task synchronization information.

FIG17 is a possible logical structure diagram of a computer device 170 provided in an embodiment of the present application. The computer device 170 includes: a processor 1701, a communication interface 1702, a storage system 1703, and a bus 1704. The processor 1701, the communication interface 1702, and the storage system 1703 are interconnected via the bus 1704. In an embodiment of the present application, the processor 1701 is used to control and manage the actions of the computer device 170. For example, the processor 1701 is used to execute the steps performed by the sending end in the method embodiment of FIG2. The communication interface 1702 is used to support the computer device 170 to communicate. The storage system 1703 is used to store program codes and data of the computer device 170.

Among them, the processor 1701 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. The processor 1701 can also be a combination that implements a computing function, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, and the like. The bus 1704 can be a peripheral component interconnect standard (Peripheral Component Interconnect, PCI) bus or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in Figure 17, but it does not mean that there is only one bus or one type of bus.

The transceiver unit 1501 in the apparatus 150 is equivalent to the communication interface 1702 in the computer device 170 , and the processing unit 1502 in the apparatus 150 is equivalent to the processor 1701 in the computer device 170 .

The computer device 170 of this embodiment may correspond to the first federated learning client in the method embodiment of FIG. 2 . The communication interface 1702 in the computer device 170 may implement the functions and/or various steps of the first federated learning client in the method embodiment of FIG. 2 , which will not be described in detail for the sake of brevity.

FIG18 is another possible logical structure diagram of a computer device 180 provided in an embodiment of the present application. The computer device 180 includes: a processor 1801, a communication interface 1802, a storage system 1803, and a bus 1804. The processor 1801, the communication interface 1802, and the storage system 1803 are interconnected via the bus 1804. In an embodiment of the present application, the processor 1801 is used to control and manage the actions of the computer device 180. For example, the processor 1801 is used to execute the steps performed by the receiving end in the method embodiment of FIG2. The communication interface 1802 is used to support the computer device 180 to communicate. The storage system 1803 is used to store the program code and data of the computer device 180.

Among them, the processor 1801 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. The processor 1801 can also be a combination that implements a computing function, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, and the like. The bus 1804 can be a peripheral component interconnect standard (Peripheral Component Interconnect, PCI) bus or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in Figure 18, but it does not mean that there is only one bus or one type of bus.

The transceiver unit 1601 in the apparatus 160 is equivalent to the communication interface 1802 in the computer device 180 , and the processing unit 1602 in the apparatus 160 is equivalent to the processor 1801 in the computer device 180 .

The computer device 180 of this embodiment may correspond to the receiving end in the method embodiment of FIG. 2 . The communication interface 1802 in the computer device 180 may implement the functions and/or various steps of the receiving end in the method embodiment of FIG. 2 . For the sake of brevity, they will not be described in detail here.

It should be understood that the division of the units in the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And the units in the device can all be implemented in the form of software calling through processing elements; they can also be all implemented in the form of hardware; some units can also be implemented in the form of software calling through processing elements, and some units can be implemented in the form of hardware. For example, each unit can be a separately established processing element, or it can be integrated in a certain chip of the device. In addition, it can also be stored in the memory in the form of a program, and called and executed by a certain processing element of the device. The function of the unit. In addition, all or part of these units can be integrated together, or they can be implemented independently. The processing element described here can also be a processor, which can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each unit above can be implemented by an integrated logic circuit of hardware in the processor element or in the form of software calling through a processing element.

In one example, the unit in any of the above devices may be one or more integrated circuits configured to implement the above method, such as: one or more application specific integrated circuits (ASIC), or one or more digital singnal processors (DSP), or one or more field programmable gate arrays (FPGA), or a combination of at least two of these integrated circuit forms. For another example, when the unit in the device can be implemented in the form of a processing element scheduler, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call a program. For another example, these units can be integrated together and implemented in the form of a system-on-a-chip (SOC).

In another embodiment of the present application, a computer-readable storage medium is further provided, in which computer-executable instructions are stored. When the processor of the device executes the computer-executable instructions, the device executes the method executed by the first federated learning client in the above method embodiment.

In another embodiment of the present application, a computer-readable storage medium is further provided, in which computer-executable instructions are stored. When the processor of the device executes the computer-executable instructions, the device executes the method executed by the federated learning server in the above method embodiment.

In another embodiment of the present application, a computer program product is further provided, the computer program product comprising computer-executable instructions stored in a computer-readable storage medium. When the processor of the device executes the computer-executable instructions, the device executes the method executed by the first federated learning client in the above method embodiment.

In another embodiment of the present application, a computer program product is further provided, the computer program product comprising computer executable instructions, the computer executable instructions being stored in a computer readable storage medium. When the processor of the device executes the computer executable instructions, the device executes the method executed by the federated learning server in the above method embodiment.

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

In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk.

Claims (30)

1. A reasoning method, characterized in that the method includes: The first federated learning client sends a first local feature to the federated learning server. The first local feature is extracted from the first network data. The first network data is the first federated learning client. Data related to the target network resource obtained by the client in the first time slot, and the target network resource is a network resource managed by the first federated learning client; The first federated learning client receives global prior features from the federated learning server. The global prior features are obtained based on the first local features and the second local features. The second local features are Provided by the second federated learning client; The first federated learning client performs inference based on the global prior features and second network data to obtain inference results. The second network data is obtained by the first federated learning client in the second time slot. Data related to the target network resource, the inference result is used to manage the target network resource, wherein the second time slot is the same as the first time slot or after the first time slot. 2. The method according to claim 1, characterized in that the first network data is the sampling value of the data related to the target network resource in the first time slot or from the third time slot to the third time slot. The statistical value up to one time slot, the second network data is the sampling value of the data related to the target network resource in the second time slot, and the third time slot is before the first time slot. 3. The method according to claim 1 or 2, characterized in that the global a priori feature is a feature vector or a first machine learning model, and the first machine learning model is used for inference of the second network data. . 4. The method according to claim 3, characterized in that, when the global a priori feature is the feature vector, the first federated learning client uses the global a priori feature and the second network Perform inference on data to obtain inference results including: The first federated learning client performs inference based on the global prior features, the second network data and the local second machine learning model to obtain the inference result, and the second machine learning model is used in the second network Inference from data. 5. The method according to claim 4, characterized in that the first federated learning client performs inference based on the global prior features, the second network data and the local second machine learning model to obtain the inference result. include: The first federated learning client inputs the second network data into the third learning model to obtain multiple features of the second network data output by the third learning model; The first federated learning client inputs the global prior features into the second machine learning model to obtain respective weights of multiple features of the second network data; The first federated learning client determines the inference result based on multiple features of the second network data and respective weights of the multiple features of the second network data. 6. The method of claim 4, wherein the second machine learning model includes a plurality of first task models; The first federated learning client performs inference based on the global prior features, the second network data, and the local second machine learning model to obtain inference results including: The first federated learning client calculates respective weights of the multiple first task models based on the global prior features; The first federated learning client inputs the characteristics of the second network data into the plurality of first task models to obtain the inference characteristics output by the plurality of first task models; The first federated learning client obtains an inference result based on the respective weights of the multiple first task models and the inference features output by the multiple first task models. 7. The method according to claim 6, wherein the first federated learning client calculates the respective weights of the plurality of first task models according to the global a priori features including: The first federated learning client calculates respective weights of the plurality of first task models based on the global prior features and the second network data. 8. The method according to claim 6 or 7, characterized in that the method further comprises: The first federated learning client extracts features of the second network data through a third machine learning model. 9. The method of claim 8, wherein the third machine learning model includes a plurality of second task models; The first federated learning client uses a third machine learning model to extract features of the second network data including: The first federated learning client determines respective weights of the plurality of second task models based on the second network data; The first federated learning client inputs the second network data into the plurality of second task models to obtain the sub-features of the second network data output by the plurality of second task models; The characteristics of the second network data are obtained according to the respective weights of the plurality of second task models and the sub-features of the second network data output by the plurality of second task models. 10. The method according to claim 9, characterized in that each second task model is a layer of autoencoders, and the reconstruction target of the r-th task model in the plurality of second task models is the r-th The residual of 1 task model, where r is an integer greater than 1 and represents the number of the second task model. 11. The method according to claim 3, characterized in that, when the global a priori feature is the first machine learning model, the first federated learning client uses the global a priori feature and Perform inference on the second network data to obtain inference results including: The first federated learning client extracts features of the second network data; The first federated learning client inputs the characteristics of the second network data into the first machine learning model to obtain the inference result output by the first machine learning model. 12. The method according to claim 3, characterized in that, when the global a priori feature is the first machine learning model, the first federated learning client uses the global a priori feature and Perform inference on the second network data to obtain inference results including: The first federated learning client uses sample data to train the first machine learning model; The first federated learning client extracts features of the second network data; The first federated learning client inputs the characteristics of the second network data into the trained first machine learning model to obtain the inference result output by the trained first machine learning model. 13. The method according to any one of claims 1 to 12, characterized in that the method further comprises: The first federated learning client sends grouping information to the federated learning server, and the grouping information indicates the group in which the first local feature is located, so that the federated learning server can use the first local feature according to the first local feature. The group in which the first local feature is located, the second local feature and the group in which the second local feature is located obtain global prior features. 14. The method according to any one of claims 1 to 13, characterized in that the method further comprises: The first federated learning client receives task synchronization information from the federated learning server, and the task synchronization information is used to indicate the first time slot; The first federated learning client selects the first network data from local data according to the task synchronization information. 15. A reasoning method, characterized in that the method includes: The federated learning server receives the first local feature from the first federated learning client, the first local feature is extracted from the first network data, and the first network data is the first federated learning client in Data related to the target network resource obtained in the first time slot, where the target network resource is a network resource managed by the first federated learning client; The federated learning server obtains global prior features based on the first local feature and the second local feature, and the second local feature is provided by the second federated learning client; The federated learning server sends the global prior features to the first federated learning client, so that the first federated learning client performs inference based on the global prior features and the second network data to obtain inferences. As a result, the second network data is data related to the target network resource obtained by the first federated learning client in the second time slot, and the inference result is used to manage the target network resource, where: The second time slot is the same as the first time slot or after the first time slot. 16. The method according to claim 15, characterized in that the first network data is the sampled value of the data related to the target network resource in the first time slot or from the third time slot to the third time slot. The statistical value up to one time slot, the second network data is the sampling value of the data related to the target network resource in the second time slot, and the third time slot is before the first time slot. 17. The method according to claim 15 or 16, characterized in that the global a priori feature is a feature vector or a first machine learning model, and the first machine learning model is used for inference of the second network data. . 18. The method according to any one of claims 15 to 17, characterized in that the method further comprises: The federated learning server receives grouping information from the first federated learning client, where the grouping information from the first federated learning client indicates the group in which the first local feature is located; The federated learning server obtains global prior features based on the first local feature and the second local feature, including: The federated learning server obtains global prior features based on the first local feature, the group in which the first local feature is located, the second local feature, and the group in which the second local feature is located. The second The group in which the local feature is located is indicated by the group information from the second federated learning client. 19. The method of claim 18, wherein the first local feature includes a first sub-feature and a second sub-feature, and the second local feature includes a third sub-feature and a fourth sub-feature, so The grouping information from the first federated learning client indicates the group in which the first sub-feature is located and the group in which the second sub-feature is located, and the grouping information from the second federated learning client indicates the group in which the first sub-feature is located. The group in which the third sub-feature is located is the same as the group in which the fourth sub-feature is located, and the group in which the first sub-feature is located is the same as the group in which the third sub-feature is located; The federated learning server obtains global prior features based on the first local feature, the group in which the first local feature is located, the second local feature, and the group in which the second local feature is located, including: Based on the fact that the group where the first sub-feature is located and the group where the third sub-feature is located are the same, the federated learning server processes the first sub-feature and the third sub-feature to obtain intermediate features; The federated learning server obtains a global prior based on the intermediate feature, the second sub-feature, the fourth sub-feature, the group in which the second sub-feature is located, and the group in which the fourth sub-feature is located. feature. 20. The method according to any one of claims 15 to 19, wherein the federated learning server obtains global a priori features based on the first local feature and the second local feature, including: The federated learning server obtains global prior features based on the first local features, the second local features, the historical local features from the first federated learning client, and the historical local features from the second federated learning client. . 21. The method according to claim 20, characterized in that the federated learning server uses the first local feature, the second local feature, the historical local feature from the first federated learning client and the historical local feature from the first federated learning client. The second federated learning client uses historical local features to obtain global prior features including: The federated learning server calculates the similarity between the local features of the current reasoning process and multiple groups of historical local features. The local features of the current reasoning process include the first local features and the second local features. Each group of historical local features Features include historical local features from the first federated learning client and historical local features from the second federated learning client during a historical reasoning process; The federated learning server performs a weighted summation of the historical prior features corresponding to the multiple sets of historical local features based on the similarity between the local features of the current reasoning process and multiple sets of historical local features to obtain the global prior features. test characteristics. 22. The method according to claim 21, characterized in that the plurality of groups of historical local features all have labels, and the labels are the actual results of each group of the historical local features that are manually labeled; The method also includes: The federated learning server receives the inference results from the first federated learning client; The federated learning server determines a target inference result based on the inference result of the first federated learning client and the inference result of the second federated learning client; When the similarity between the local features of the current reasoning process and the historical local features of the target group is greater than or equal to the threshold, the federated learning server determines the similarity between the local features of the current reasoning process and the historical local features of the target group. The similarity updates the historical local features of the target group, the historical local features of the target group are among the multiple groups of historical local features, and the label is the target inference result; When the similarity between the local features of the current reasoning process and the historical local features of the target group is less than the threshold, the federated learning server adds a set of historical local features based on the multiple sets of historical local features. Features, the added set of historical local features are local features of the current reasoning process. 23. The method according to any one of claims 15 to 22, characterized in that the method further comprises: The federated learning server sends task synchronization information to the first federated learning client, and the task synchronization information is used to indicate the first time slot, so that the first federated learning client synchronizes according to the task Information selects the first network data from local data. 24. An inference device, characterized in that it is applied to a client, and the device includes: A transceiver unit, configured to send a first local feature to the federated learning server. The first local feature is extracted from the first network data. The first network data is the first federated learning client. Data related to the target network resource obtained in the first time slot, where the target network resource is a network resource managed by the first federated learning client; The transceiver unit is configured to receive global prior features from the federated learning server. The global prior features are obtained based on the first local features and the second local features. The second local features are obtained by The second federated learning client is provided; A processing unit configured to perform inference based on the global prior features and second network data to obtain inference results. The second network data is obtained by the first federated learning client in the second time slot and is the same as the second network data obtained by the first federated learning client in the second time slot. Data related to the target network resource, the inference result is used to manage the target network resource, wherein the second time slot is the same as the first time slot or after the first time slot. 25. An inference device, characterized in that it is applied to the server, and the device includes: A transceiver unit configured to receive the first local feature from the first federated learning client, where the first local feature is extracted from the first network data, and the first network data is the first federated learning client. Data related to the target network resource obtained in the first time slot, where the target network resource is a network resource managed by the first federated learning client; A processing unit configured to obtain global prior features based on the first local features and the second local features, where the second local features are provided by the second federated learning client; The transceiver unit is configured to send the global a priori features to the first federated learning client, so that the first federated learning client performs inference based on the global a priori features and the second network data to obtain Inference result, the second network data is data related to the target network resource obtained by the first federated learning client in the second time slot, and the inference result is used to manage the target network resource, wherein, The second time slot is the same as or subsequent to the first time slot. 26. A computer device, characterized in that the computer device includes: a memory and a processor; The processor is configured to execute computer programs or instructions stored in the memory, so that the computer device performs the method according to any one of claims 1-14. 27. A computer device, characterized in that the computer device includes: a memory and a processor; The processor is configured to execute computer programs or instructions stored in the memory, so that the computer device performs the method according to any one of claims 15-23. 28. A computer-readable storage medium, characterized in that the computer-readable storage medium has program instructions, and when the program instructions are executed directly or indirectly, the method of any one of claims 1 to 23 is achieved. method is implemented. 29. A chip system, characterized in that the chip system includes at least one processor, the processor is used to execute a computer program or instructions stored in a memory, and when the computer program or instructions are executed in the at least one When executed in a processor, the method according to any one of claims 1 to 23 is implemented. 30. A computer program product, characterized by comprising instructions that, when run on a computer, cause the computer to perform the method of any one of claims 1 to 23.