WO2021008017A1 - 联邦学习方法、系统、终端设备及存储介质 - Google Patents

联邦学习方法、系统、终端设备及存储介质 Download PDF

Info

Publication number
WO2021008017A1
WO2021008017A1 PCT/CN2019/116464 CN2019116464W WO2021008017A1 WO 2021008017 A1 WO2021008017 A1 WO 2021008017A1 CN 2019116464 W CN2019116464 W CN 2019116464W WO 2021008017 A1 WO2021008017 A1 WO 2021008017A1
Authority
WO
WIPO (PCT)
Prior art keywords
federated learning
alliance
enterprise
model
request
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/116464
Other languages
English (en)
French (fr)
Inventor
程勇
刘洋
陈天健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
WeBank Co Ltd
Original Assignee
WeBank Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by WeBank Co Ltd filed Critical WeBank Co Ltd
Publication of WO2021008017A1 publication Critical patent/WO2021008017A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning

Definitions

  • This application relates to the technical field of financial technology (Fintech), in particular to a federated learning method, system, terminal device and storage medium.
  • Federated learning Learning refers to the combination of different participants (participant, or party, also known as data owner (data owner), or client (client)) for machine learning modeling methods.
  • participants do not need to ask other participants and coordinators (coordinator, also called server), parameter server (parameter server), or aggregation server (aggregation server)) expose the data you own, so federated learning can protect user privacy and ensure data security, and can solve the problem of data islands.
  • federated learning alliance always has an initiator And the participants who keep joining. In this way, a lot of manpower, material resources, and time resources need to be consumed in the implementation process of the enterprise's field construction of federated learning.
  • the main purpose of this application is to provide a federated learning method, system, terminal equipment and storage medium, aiming to save the enterprises wishing to conduct federated learning, and reduce the cost of manpower, material resources and time resources in the process of constructing federated learning model.
  • the cost of learning model construction is to provide a federated learning method, system, terminal equipment and storage medium, aiming to save the enterprises wishing to conduct federated learning, and reduce the cost of manpower, material resources and time resources in the process of constructing federated learning model.
  • this application provides a federated learning method, which includes the following steps:
  • the method further includes:
  • the step of forming a federal learning alliance according to the federal learning request includes:
  • the target enterprise is matched from the preset enterprise information database as alliance members to form a federated learning alliance.
  • the step of extracting member information of each alliance member in the federated learning alliance includes:
  • the step of constructing a federated learning model based on the member information includes:
  • the step of managing the federated learning model includes:
  • the steps of model training and model update of the federated learning model constructed by the management include:
  • the acquired federated learning request is a management request, and model training and model update management are performed on the federated learning model according to the management request; or,
  • the preset management strategy periodically perform model training and model update management on the federated learning model.
  • the federated learning method further includes:
  • the step of serving the alliance members of the federated learning alliance includes:
  • this application also provides a federal learning system, which includes:
  • the acquisition module is set to acquire the federated learning request received by the input interface
  • the alliance formation module is set to form a federal learning alliance according to the federal learning request
  • the detection module is configured to extract member information of each alliance member in the federated learning alliance
  • the model management module is configured to construct a federated learning model based on the member information and manage the federated learning model.
  • the federated learning system further includes:
  • the analysis module is configured to extract the requested items carried in the federated learning request and analyze whether the requested items can be executed.
  • the federated learning system further includes:
  • the service module is configured to provide services to the alliance members of the federated learning alliance according to the request items of the federated learning request.
  • this application also provides a terminal device.
  • the terminal device includes a memory, a processor, and computer-readable instructions that are stored on the memory and can run on the processor.
  • the computer-readable instructions are The processor implements the steps of the federated learning method as described above when executed.
  • the present application also provides a storage medium applied to a computer, and computer-readable instructions are stored on the storage medium, and when the computer-readable instructions are executed by a processor, the steps of the federated learning method described above are implemented.
  • This application obtains the federated learning request received by the input interface; forms a federated learning alliance according to the federated learning request; extracts member information of each alliance member in the federated learning alliance; constructs a federated learning model based on the member information, and Manage the federated learning model.
  • the suitable company is selected as the partner of the enterprise that submitted the federated learning request to form a federated learning alliance.
  • the member information carried by each member of the learning alliance is constructed into a federated learning model suitable for the federated learning of each alliance member, and the completed federated learning model is managed.
  • Figure 1 is a schematic structural diagram of a hardware operating environment involved in a solution of an embodiment of the present application
  • Figure 2 is a schematic flow chart of the first embodiment of the federated learning method of this application.
  • step S300 in FIG. 2 is a schematic diagram of detailed steps of step S300 in FIG. 2;
  • FIG. 4 is a schematic diagram of the modules of the Federal Learning System of this application.
  • Fig. 1 is a schematic structural diagram of a hardware operating environment involved in a solution of an embodiment of the present application.
  • Fig. 1 can be a structural diagram of the hardware operating environment of the federated learning method device.
  • the device of the federated learning method in the embodiment of this application may be a terminal device such as a PC and a portable computer.
  • the federated learning method device may include a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection and communication between these components.
  • the user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 can be a high-speed RAM memory or a stable memory (non-volatile memory), such as disk storage.
  • the memory 1005 may also be a storage device independent of the foregoing processor 1001.
  • the device structure of the federated learning method shown in FIG. 1 does not constitute a limitation on the device of the federated learning method, and may include more or fewer components than shown in the figure, or a combination of certain components, or different The layout of the components.
  • a memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and computer readable instructions.
  • the operating system is a program that manages and controls the hardware and software resources of the sample federated learning method equipment, and supports the operation of computer readable instructions and other software or programs.
  • the user interface 1003 is mainly used for data communication with various terminals;
  • the network interface 1004 is mainly used for connecting to a background server and performing data communication with the background server;
  • the processor 1001 can be used for calling Computer readable instructions stored in the memory 1005, and perform the following operations:
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005, and after executing the step of obtaining the federated learning request received by the input interface, execute the following steps:
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • the target enterprise is matched from the preset enterprise information database as alliance members to form a federated learning alliance.
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • the acquired federated learning request is a management request, and model training and model update management are performed on the federated learning model according to the management request; or,
  • the preset management strategy periodically perform model training and model update management on the federated learning model.
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • processor 1001 may also be used to call computer-readable instructions stored in the memory 1005 and execute the following steps:
  • FIG. 2 is a schematic flowchart of the first embodiment of the federal learning method of this application.
  • the embodiment of this application provides an embodiment of the federated learning method. It should be noted that although the logical sequence is shown in the flowchart, in some cases, the sequence shown or described may be executed in a different order than here. A step of.
  • the federated learning method of the embodiment of this application is applied to the device of the federated learning method.
  • the device of the federated learning method of the embodiment of this application may be a terminal device such as a PC and a portable computer, which is not specifically limited here.
  • Step S100 Obtain the federated learning request received by the input interface.
  • the federated learning request input by the enterprise user is obtained from the input interface set by the federated learning method device.
  • an enterprise user who wants to participate in federated learning can input the service request of the enterprise user who wants to participate in federated learning based on an input interface set on a PC, a portable computer or other federated learning method equipment.
  • a business user who wants to participate in federated learning is based on the input interface (specifically, the input window for service request) set up on the PC terminal (one of the federated learning method devices) for the corporate user to enter the service request, enter the input that he wants to participate in federated learning
  • the input interface specifically, the input window for service request
  • the PC terminal one of the federated learning method devices
  • step S100 after obtaining the federated learning request received by the input interface, the federated learning method of this application further includes:
  • Step S500 Extract request items carried in the federated learning request, and determine that the request items can be executed.
  • the federated learning service request input by an enterprise user may include information about the data owned by the enterprise user (for example, the amount of data and data characteristics of the user of the enterprise user), as well as information about the current enterprise user
  • the requested items of the federated learning model requested to participate for example, the problems that need to be solved through federated learning, the performance indicators of the requested federated learning model and other service requirements).
  • the federated learning service request Contains information about the data owned by the enterprise users themselves, as well as the service requirements that the enterprise users need to solve for the federated learning model requested to participate, and the performance indicators of the requested federated learning model, and further analyze the requirements of the enterprise users to pass
  • the problems solved by federated learning can be solved by constructing a federated learning model, and whether the constructed federated learning model can meet the performance indicators of the learning model requested by business users and so on.
  • Step S200 forming a federated learning alliance according to the federated learning request.
  • the cooperative enterprise of the corporate user who is currently initiating the federated learning request is searched for, thereby forming a federated learning alliance.
  • step S200 includes:
  • step S201 it is determined that the requested item can be executed, and then the enterprise label information carried in the federated learning request is detected.
  • the federated learning service request input by the enterprise user may also include the enterprise label information of the enterprise user who entered the federated learning service request (for example, enterprise attribute parameters such as the enterprise type and data type used to identify the enterprise user) ).
  • the analysis determines that the problems that enterprise users need to solve through federated learning can be solved by constructing a federated learning model, and the federated learning model constructed can also meet the performance indicators of the learning model requested by the corporate user, start to detect the corporate user
  • the enterprise label information in the input federated learning service request such as enterprise attribute parameters such as the enterprise type and data type of the enterprise user.
  • Step S202 According to the enterprise tag information, the target enterprise is matched from the preset enterprise information database as alliance members to form a federated learning alliance.
  • the corporate tag information of the corporate users carried in the detected federal learning request from all corporate users stored in the preset corporate information database, find and determine the target company suitable for establishing a cooperative relationship, so as to be an alliance member with the current Enterprise users who input the federal study request form a federal study alliance.
  • the preset enterprise information database is a database used to store the attribute parameters of the enterprise users who are willing to participate in federated learning (willing to participate in the federated learning) collected in advance, wherein the preset enterprise information database stores Each wants to participate in the federation to learn one or more attribute parameters of enterprise users, such as: the type of enterprise, the type of data owned by the enterprise, and the type of enterprise business, as well as the amount of data owned by the enterprise user, the characteristics of the data, the label of the data, and the ability of the enterprise data Completed machine learning tasks (for example, detection, prediction, or classification); the performance indicators of the federated learning model expected by each enterprise user, and the federated learning overhead that can be undertaken (for example, communication overhead, time overhead, computing resources, power overhead), hope
  • the federated learning benefits obtained for example, model performance improvement, or compensation).
  • the enterprise type, data type and other enterprise attribute parameters of the enterprise user who currently wants to participate in the federated learning are saved from the saved In the preset enterprise information database of one or more attribute parameters, one by one finds out target enterprises that are the same as or complementary to the enterprise type, data type and other enterprise attribute parameters of the enterprise user currently input to federated learning as the alliance member, and the current input Enterprise users of Federated Learning jointly formed the Federal Learning Alliance.
  • Step S300 Extract member information of each alliance member in the federated learning alliance.
  • the detected member information of each alliance member in the federated learning alliance includes but is not limited to: data information (for example: the amount of data owned by the enterprise, the characteristics of the data owned, and the label of the data), and the information of each enterprise user Hope based on the task information completed by federal learning (for example: face recognition, loan risk control, product recommendation, etc.).
  • data information for example: the amount of data owned by the enterprise, the characteristics of the data owned, and the label of the data
  • federal learning for example: face recognition, loan risk control, product recommendation, etc.
  • step S300 extracting member information of each alliance member in the federated learning alliance includes:
  • Step S301 detecting all the enterprise parameters stored in the preset enterprise information database.
  • all the enterprise parameters of all enterprise users who wish to participate in federated learning stored in the preset enterprise information database are detected, including: The attribute parameters of enterprise users, such as the type of enterprise, the type of data owned by the enterprise, and the type of enterprise business, as well as the amount of data owned by the enterprise user, the characteristics of the data, the label of the data, and the machine learning tasks that the enterprise data can complete (for example, detection , Prediction or classification); the performance indicators of the federated learning model expected by each enterprise user, the federated learning overhead that can be undertaken (for example, communication overhead, time overhead, computing resources, power overhead), and the expected federated learning benefits (for example, model Performance improvement rate, or cash reward).
  • the attribute parameters of enterprise users such as the type of enterprise, the type of data owned by the enterprise, and the type of enterprise business, as well as the amount of data owned by the enterprise user, the characteristics of the data, the label of the data, and the machine learning tasks that the enterprise data can complete (for example, detection , Prediction or classification); the performance indicators of
  • Step S302 Index the enterprise label information of the target enterprise corresponding to each alliance member from all the enterprise parameters.
  • each alliance member when it is detected that the current federated learning alliance formed based on the corporate label information (enterprise attribute parameters such as corporate type, data type, etc.) of corporate users carried in the federated learning request, each alliance member also saves corporate label information, All enterprise parameters of each enterprise user stored in the detected preset enterprise information database (each wish to participate in the federation to learn one or more attribute parameters of enterprise users, such as: enterprise type, enterprise data type and enterprise business type, and The amount of data owned by enterprise users, the characteristics of the data, the labels of the data, the machine learning tasks that enterprise data can complete, such as prediction or classification; the performance indicators of the federated learning model expected by each enterprise user, and the federated learning overhead that they can bear, For example, communication costs, time costs, computing resources, power costs, desired federated learning benefits, such as model performance improvement), traverse the enterprise parameters of enterprise users corresponding to each alliance member to index the current federated learning alliance , Enterprise label information of each alliance member (that is, one or more attribute parameters of enterprise users that
  • step S303 the member information of each federated member required to form the federated learning alliance is extracted from the enterprise label information one by one.
  • the detected label information of each alliance member that is, one or more attribute parameters and data information of each alliance member, For example: the amount of data owned by the enterprise, the characteristics of the data owned, the label of the data; and, the task information that each enterprise user hopes to complete based on federal learning, such as: face recognition, loan risk control, product recommendation, etc., extract data information (The amount of data owned by the enterprise, the data characteristics it owns, the label of the data) and the task information (face recognition, loan risk control, product recommendation, etc.) that each enterprise user wants to complete based on federal learning, as the current federal learning alliance Member information of each alliance member.
  • federal learning such as: face recognition, loan risk control, product recommendation, etc.
  • Step S400 Construct a federated learning model based on the member information, and manage the federated learning model.
  • the member information of each alliance member the data information and task information of each alliance member, construct a federated learning model suitable for the federated learning of each alliance member in the current federal learning alliance , And manage the constructed federated learning model after the construction of the federated learning model is completed.
  • the enterprise users want to be based on the federation Information on the tasks completed by learning-face recognition, loan risk control, product recommendation and other member information, as well as the expenses that each alliance member can bear-communication, time, computing resource expenses and hopes to obtain benefits based on current federal learning- -Model performance requirements, model performance improvements, etc., determine the cost and benefit distribution of alliance members (federated learning participants) in the current federated learning alliance (for example, according to the amount of data of each participant and the contribution to the federated learning model Determine the reward distribution, based on the regression model, neural network prediction model, neural network classification model, etc., determine the distribution strategy of the expenses and benefits of each alliance member), so as to construct a federated learning suitable for each alliance member in the current federated learning alliance
  • the federated learning model after completing the construction of the federated learning model
  • step S400 includes:
  • Step S401 Design the model structure of the federated learning model based on the extracted member information.
  • the federated learning method of the present application can also design the model structure of the federated learning model based on an existing deep learning model that has been proven in practice.
  • Step S402 After completing the model structure design of the federated learning model, initialize the federated learning model.
  • the model parameters of the current federated learning model that has been designed are initialized through random initialization to ensure that the currently constructed federated learning model can It is suitable for all members of the current federal learning alliance to conduct federal learning.
  • This application obtains the federated learning request input by the enterprise user from the input interface set by the federated learning method device when it is detected that the enterprise user inputs the federated learning request based on the above-mentioned federated learning method device; On the input interface of, after obtaining the federated learning request input by the enterprise user, from the obtained federated learning request, extract the requested items of the federated learning model that the enterprise user requests to participate in the federated learning request, and Further analyze whether the requested items can be executed by the constructed federated learning model; when the analysis determines that the federated learning model included in the federated learning request includes the requested items of the federated learning model requested by the enterprise user, the federated learning model can be constructed When executed, it detects the enterprise label information of the enterprise user who entered the federated learning request carried in the federated learning request, and from the preset enterprise information database according to the enterprise label information of the enterprise user carried in the detected federated learning request Among all the saved enterprise users, find and determine the target enterprise
  • the suitable enterprise is selected as the partner of the enterprise that submitted the federated learning request, so as to form a federated learning alliance.
  • the member information carried by the members of the federated learning alliance is constructed into a federated learning model suitable for the federated learning of the federated members. Therefore, companies that do not need to submit a federal learning request communicate with other companies one by one on the spot to persuade other companies to build a federal learning model as their partners, and save the manpower and material resources in the process of building a federal learning model for companies wishing to conduct federal learning. And time resources, reducing the cost of building a federated learning model.
  • managing the federated learning model includes:
  • Step S403 Manage the model training and model update of the constructed federated learning model.
  • the federated learning model is trained and modeled according to the management request of each federated member based on the federated learning method and equipment. Update management, or, based on a preset management strategy, model training and model update management of the currently constructed federated learning model.
  • step S403 includes:
  • step S4031 it is detected that the acquired federated learning request is a management request, and the federated learning model is subjected to model training and model update management according to the management request.
  • the federated learning request input by the enterprise user obtained from the input interface set by the federated learning method device is a request to manage the federated learning model serving the current enterprise user, update according to the input of the enterprise user
  • the requested content, model training and model update of the federated learning model serving current enterprise users is a request to manage the federated learning model serving the current enterprise user.
  • the federated learning service request entered in the input window of the set service request is specifically for the federated learning that has been constructed to serve the current enterprise user
  • the model is updated for the management request, it is based on the content of the management request input by the enterprise user.
  • the content of the management request input by the user is: model training for the current federated learning model, then according to the management request, manage federated learning Retrain the model, continue training, or manage the federated learning model to redesign and train the model, or the content of the management request entered by the user is: update the current federated learning alliance members (federated learning participants) , According to the current update request, from the preset enterprise information database, re-search and confirm the target enterprise as the new alliance member, that is, re-select the federal learning participant.
  • Step S4032 periodically perform model training and model update management on the federated learning model.
  • the preset model management strategy is a model management strategy that is set in advance based on the stability and other characteristics of the constructed federated learning model to perform periodic update processing on the current federated learning model.
  • the federated learning model is automatically trained and updated according to a predetermined time period (for example, one week), wherein the management of the federated learning model's model training includes at least: management Retraining and continuing training of the federated learning model, or managing the redesign and training of the federated learning model; the update processing of the federated learning model includes: updating the model parameters, that is, retraining the current federated learning model, for example, A neural network structure retrains the parameters of the neural network to obtain new model parameters of the current federated learning model; the update of the model structure means updating the model structure of the current federated learning model and retraining, for example, changing the structure of the neural network and re-training. Train the parameters of the neural network to obtain a new federated learning model.
  • the management of the federated learning model's model training includes at least: management Retraining and continuing training of the federated learning model, or managing the redesign and training of the federated learning model
  • the update processing of the federated learning model includes: updating
  • the federated learning model will be trained according to the management request of each federated member based on the federated learning method and equipment. And model update management, or, based on the preset model management strategy, the currently constructed federated learning model training and model update management.
  • the management is performed according to the input of the enterprise user
  • the requested content manages the model training and model update of the federated learning model serving current enterprise users; and, from the time the federated learning model is completed, the federated learning model is automatically trained and updated according to the predetermined time period.
  • the management of the model training of the federated learning model includes: managing the retraining and continuing training of the federated learning model, or managing the redesign and training of the federated learning model
  • the update processing of the federated learning model includes: updating of model parameters And model structure updates.
  • the model training and model of the federated learning model Perform management more, or automatically and periodically manage the model training and model update of the constructed federated learning model according to the pre-established model management strategy, thereby ensuring that the constructed federated learning model is in the process of serving enterprise users
  • the high efficiency saves the cost of updating the model independently by requesters participating in federated learning, and further improves the efficiency of creating federated learning models.
  • the federated learning method of this application further includes:
  • Step A serving the alliance members of the federated learning alliance according to the request items of the federated learning request.
  • the enterprise user wants to provide services to the current enterprise user or other alliance members of the federated learning alliance where the current enterprise user is located.
  • the federated learning service request entered in the input window of the set service request is specifically for the federated learning that has been constructed to serve the current enterprise user
  • the model is updated for the update request
  • the currently constructed federated learning model is updated; further, based on the current enterprise user’s input hope to perform federated learning training to complete the service request of loan risk control prediction operation, the current Enterprise users provide the prediction results obtained from the federated learning model training; further, when the enterprise users who wish to participate in the federated learning carry a local training model, based on the federated learning model parameter request input by the enterprise user, provide the current enterprise user
  • the model parameters and the federated learning code of the currently constructed federated learning model are available for current enterprise users to perform local machine learning based on their own data.
  • the analysis detects that the obtained enterprise user is based on the input window set on the PC terminal for the enterprise user to input the service request
  • the federated learning service request inputted by the enterprise user
  • the problem solved by federated learning cannot be solved by constructing a federated learning model, and/or the built federated learning model cannot meet the requirements such as the performance indicators of the learning model requested by enterprise users, use the current federated learning method on the device .
  • the set output interface (specifically, the feedback window for inputting service requests for enterprise users) outputs to the current enterprise users a prompt message that the "federated learning service request" is rejected.
  • This application provides services to the current enterprise user or other members of the federated learning alliance in which the enterprise user is located, based on the request items that the enterprise user enters in the federal learning request entered by the enterprise user obtained from the input interface, that is, to detect
  • the obtained federated learning service request input by the enterprise user is specifically an update request for updating the federated learning model that has been constructed to serve the current enterprise user
  • the federated learning method device can automatically find a federated learning partner for current enterprise users, design the model structure and initialize the federated learning model according to the request items contained in the service request, and then send the initial federated learning model to the service request.
  • Enterprise users, as well as the training and management of the completed federated learning model do not need to participate in federated learning for corporate users to independently carry out federated learning methods, training operations, management, and maintenance, which saves the manpower, material resources and resources of federated learning participants. Time cost improves the efficiency of federated learning methods.
  • an embodiment of the present application also proposes a federated learning system
  • the federated learning system includes:
  • the acquisition module is set to acquire the federated learning request received by the input interface
  • the alliance formation module is set to form a federal learning alliance according to the federal learning request
  • the detection module is configured to detect the member information of each alliance member in the federated learning alliance.
  • the model management module is configured to construct a federated learning model based on the member information and manage the federated learning model.
  • the federated learning system further includes:
  • the analysis module is configured to extract the requested items carried in the federated learning request and determine that the requested items can be executed.
  • the alliance building module includes:
  • the second detection unit is configured to detect the enterprise label information carried in the federal learning request after determining that the requested item can be executed;
  • the alliance formation subunit is set to match the target enterprise as the alliance member from the preset enterprise information database according to the enterprise label information to form a federal learning alliance.
  • the detection module includes:
  • the first detection unit is configured to detect all enterprise parameters stored in the preset enterprise information database
  • An indexing unit configured to index the enterprise label information of the target enterprise corresponding to each member of the alliance from all the enterprise parameters
  • the extraction unit is configured to extract the member information of each federated member required to form the federated learning alliance from the enterprise label information one by one.
  • model building module includes:
  • the design unit is set to design the model structure of the federated learning model based on the extracted member information
  • the initialization unit is set to initialize the federation learning model after completing the model structure design of the federation learning model.
  • model management module includes:
  • the management unit is configured to manage the model training and model update of the constructed federated learning model.
  • the management unit includes:
  • the first management subunit is set to detect that the acquired federated learning request is a management request, and perform model training and model update management on the federated learning model according to the management request;
  • the second management subunit is configured to periodically perform model training and model update management on the federated learning model according to a preset management strategy.
  • the federated learning system further includes:
  • the service module is configured to provide services to the alliance members of the federated learning alliance according to the request items of the federated learning request.
  • the service module includes:
  • the model distribution unit is configured to distribute the completed federated learning model to the alliance members of the federated learning alliance; or,
  • the parameter supply unit is configured to provide model training parameters to the alliance members for the alliance members to independently perform model training.
  • the embodiment of the present application also proposes a storage medium that is applied to a computer, that is, the storage medium is a computer-readable storage medium, and the computer-readable storage medium may be a non-volatile readable storage medium.
  • Computer readable instructions are stored on the medium, and when the computer readable instructions are executed by a processor, the steps of the federated learning method as described above are realized.
  • the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. ⁇
  • the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, The optical disc) includes a number of instructions to enable a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the method described in each embodiment of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种联邦学习方法、系统、终端设备及存储介质,涉及金融科技(Fintech)技术领域。所述联邦学习方法包括:获取输入接口所接收到的联邦学习请求(S100);根据所述联邦学习请求组建联邦学习联盟(S200);提取所述联邦学习联盟中各联盟成员的成员信息(S300);基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理(S400)。

Description

联邦学习方法、系统、终端设备及存储介质
本申请要求于2019年7月17日提交中国专利局、申请号为201910647743.0、发明名称为“联邦学习方法、系统、终端设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及金融科技(Fintech)技术领域,尤其涉及一种联邦学习方法、系统、终端设备及存储介质。
背景技术
伴随着金融科技(Fitech),尤其是互联网金融科技的快速发展,已经有越来越多的技术应用于金融领域,其中,联邦学习技术基于对用户隐私和数据的安全保障,正逐渐受到越来越多的重视。
联邦学习(federated learning)是指,通过联合不同的参与者(participant,或者party,也称为数据拥有者(data owner)、或者客户(client))进行机器学习建模的方法。在联邦学习中,参与者不需要向其它参与者和协调者(coordinator,也称为服务器(server),参数服务器(parameter server),或者聚合服务器(aggregation server))暴露自己所拥有的数据,因而联邦学习可以很好的保护用户隐私和保障数据安全,并可以解决数据孤岛问题。
然而,现有的希望参与联邦学习的企业需要进行实地协商,以组建联邦学习合作关系,并由其中某一个企业负责设计模型结构和提供初始模型参数,并且,一个联邦学习联盟总是有发起者和不断加入的参与者。如此,在企业进行实地构建联邦学习的实施过程中,需要消耗大量的人力、物力以及时间资源。
发明内容
本申请的主要目的在于提供一种联邦学习方法、系统、终端设备及存储介质,旨在省去希望进行联邦学习的企业,在构建联邦学习模型过程中的人力、物力和时间资源开销,降低联邦学习模型构建的成本。
为实现上述目的,本申请提供一种联邦学习方法,所述联邦学习方法包括以下步骤:
获取输入接口所接收到的联邦学习请求;
根据所述联邦学习请求组建联邦学习联盟;
提取所述联邦学习联盟中各联盟成员的成员信息;以及,
基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
可选地,在所述获取输入接口所接收到的联邦学习请求的步骤之后,还包括:
提取所述联邦学习请求中携带的请求事项,并分析所述请求事项是否能够被执行;
所述根据所述联邦学习请求组建联邦学习联盟的步骤包括:
确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
可选地,所述提取所述联邦学习联盟中各联盟成员的成员信息的步骤包括:
检测所述预设企业信息库保存的全部企业参数;
从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息;
逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
可选地,所述基于所述成员信息构建联邦学习模型的步骤包括:
依据提取出的所述成员信息,设计联邦学习模型的模型结构;
在完成所述联邦学习模型的模型结构设计之后,初始化联邦学习模型。
可选地,所述对所述联邦学习模型进行管理的步骤,包括:
管理构建的所述联邦学习模型的模型训练和模型更新;
所述管理构建的所述联邦学习模型的模型训练和模型更新的步骤包括:
检测到获取的联邦学习请求为管理请求,依据所述管理请求对所述联邦学习模型进行模型训练和模型更新管理;或者,
按照预设管理策略,周期性的对所述联邦学习模型进行模型训练和模型更新管理。
可选地,所述联邦学习方法,还包括:
根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
可选地,所述对所述联邦学习联盟的联盟成员进行服务的步骤包括:
分发构建完成的联邦学习模型至所述联邦学习联盟的联盟成员;或者,
向所述联盟成员提供模型训练参数,以供所述联盟成员自主进行模型训练。
此外,本申请还提供一种联邦学习系统,所述联邦学习系统包括:
获取模块,设置为获取输入接口所接收到的联邦学习请求;
联盟组建模块,设置为根据所述联邦学习请求组建联邦学习联盟;
检测模块,设置为提取所述联邦学习联盟中各联盟成员的成员信息;
模型管理模块,设置为基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
可选地,所述联邦学习系统,还包括:
分析模块,设置为提取所述联邦学习请求中携带的请求事项,并分析所述请求事项是否能够被执行。
可选地,所述联邦学习系统,还包括:
服务模块,设置为根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
此外,本申请还提供一种终端设备,所述终端设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,所述计算机可读指令被所述处理器执行时实现如上所述的联邦学习方法的步骤。
此外,本申请还提供一种存储介质,应用于计算机,所述存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现如上所述的联邦学习方法的步骤。
本申请通过获取输入接口所接收到的联邦学习请求;根据所述联邦学习请求组建联邦学习联盟;提取所述联邦学习联盟中各联盟成员的成员信息;基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。实现了,基于接收希望通过联邦学习建模的企业通过输入接口所输入的联邦学习请求,筛选出适合的企业作为该提出联邦学习请求企业的合作伙伴,以组建成联邦学习联盟,根据组建的联邦学习联盟各联盟成员所携带的成员信息,构建成适于各联盟成员进行联邦学习的联邦学习模型,并对构建完成的联邦学习模型进行管理。从而实现了,无需提出联邦学习请求的企业实地逐一的向其他企业进行沟通,以说服其他企业作为自己的合作伙伴构建联邦学习模型,节省了希望进行联邦学习的企业构建联邦学习模型过程中的人力、物力和时间资源,降低了联邦学习模型构建的成本。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的结构示意图;
图2为本申请联邦学习方法第一实施例的流程示意图;
图3为图2中步骤S300的细化步骤示意图;
图4为本申请联邦学习系统的模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
如图1所示,图1是本申请实施例方案涉及的硬件运行环境的结构示意图。
需要说明的是,图1即可为联邦学习方法设备的硬件运行环境的结构示意图。本申请实施例联邦学习方法设备可以是PC,便携计算机等终端设备。
如图1所示,该联邦学习方法设备可以包括:处理器1001,例如CPU,网络接口1004,用户接口1003,存储器1005,通信总线1002。其中,通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard),可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如WI-FI接口)。存储器1005可以是高速RAM存储器,也可以是稳定的存储器(non-volatile memory),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解,图1中示出的联邦学习方法设备结构并不构成对联邦学习方法设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图1所示,作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及计算机可读指令。其中,操作系统是管理和控制样本联邦学习方法设备硬件和软件资源的程序,支持计算机可读指令以及其它软件或程序的运行。
在图1所示的联邦学习方法设备中,用户接口1003主要用于与各个终端进行数据通信;网络接口1004主要用于连接后台服务器,与后台服务器进行数据通信;而处理器1001可以用于调用存储器1005中存储的计算机可读指令,并执行以下操作:
获取输入接口所接收到的联邦学习请求;
根据所述联邦学习请求组建联邦学习联盟;
提取所述联邦学习联盟中各联盟成员的成员信息;以及,
基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,在执行所述获取输入接口所接收到的联邦学习请求的步骤之后,执行以下步骤:
提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
检测所述预设企业信息库保存的全部企业参数;
从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息;
逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
依据提取出的所述成员信息,设计联邦学习模型的模型结构;
在完成所述联邦学习模型的模型结构设计之后,初始化联邦学习模型。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
管理构建的所述联邦学习模型的模型训练和模型更新。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
检测到获取的联邦学习请求为管理请求,依据所述管理请求对所述联邦学习模型进行模型训练和模型更新管理;或者,
按照预设管理策略,周期性的对所述联邦学习模型进行模型训练和模型更新管理。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
进一步地,处理器1001还可以用于调用存储器1005中存储的计算机可读指令,并执行以下步骤:
分发构建完成的联邦学习模型至所述联邦学习联盟的联盟成员;或者,
向所述联盟成员提供模型训练参数,以供所述联盟成员自主进行模型训练。
基于上述的结构,提出本申请联邦学习方法的各个实施例。
请参照图2,图2为本申请联邦学习方法第一实施例的流程示意图。
本申请实施例提供了联邦学习方法的实施例,需要说明的是,虽然在流程图中示出了逻辑顺序,但是在某些情况下,可以以不同于此处的顺序执行所示出或描述的步骤。
本申请实施例联邦学习方法应用于联邦学习方法设备,本申请实施例联邦学习方法设备可以是PC,便携计算机等终端设备,在此不做具体限制。
本实施例联邦学习方法包括:
步骤S100,获取输入接口所接收到的联邦学习请求。
当检测到企业用户基于上述联邦学习方法设备输入联邦学习请求时,从该联邦学习方法设备设置的输入接口上,获取该企业用户所输入的联邦学习请求。
本实施例中,希望参与联邦学习的企业用户,可基于PC、便携计算机或者其他联邦学习方法设备上所设置的输入接口,输入该企业用户希望参与联邦学习的服务请求。
例如,当希望参与联邦学习的企业用户基于PC终端(联邦学习方法设备之一)上所设置的供企业用户输入服务请求的输入接口(具体可为服务请求的输入窗口),输入希望参与联邦学习的服务请求时,获取当前服务请求的输入窗口中,企业用户所输入的联邦学习服务请求。
进一步地,在步骤S100,获取输入接口所接收到的联邦学习请求之后,本申请联邦学习方法,还包括:
步骤S500,提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行。
在从联邦学习方法设备设置的输入接口上,获取该企业用户所输入的联邦学习请求之后,从获取的联邦学习请求中,提取出该联邦学习请求中所包含的企业用户所请求参与的联邦学习模型的各请求事项,并进一步分析该各请求事项能否被构建出的联邦学习模型所执行。
在本实施例中,企业用户所输入的联邦学习服务请求中,可以包括该企业用户自身所拥有数据的信息(例如,企业用户所用户数据量和数据特征等信息),以及包括当前企业用户对所请求参与的联邦学习模型的请求事项(例如,需要通过联邦学习所解决的问题、所请求联邦学习模型的性能指标等服务要求)。
例如,在获取到企业用户基于PC终端上所设置的供企业用户输入服务请求的输入窗口中,获取到希望参与联邦学习的企业用户所输入的联邦学习服务请求之后,提取出联邦学习服务请求中所包含的企业用户自身所拥有数据的信息,以及企业用户对所请求参与的联邦学习模型,所需要解决的问题、所请求联邦学习模型的性能指标等服务要求,并进一步分析企业用户所需要通过联邦学习解决的问题,是否能够通过构建联邦学习模型进行解决、以及构建出的联邦学习模型是否能够满足企业用户所请求学习模型的性能指标等等。
步骤S200,根据所述联邦学习请求组建联邦学习联盟。
根据获取到的企业用户所输入联邦学习请求中所包含的,企业用户自身所拥有数据的信息,查找确定当前发起联邦学习请求的企业用户的合作企业,从而组建联邦学习联盟。
进一步地,步骤S200,包括:
步骤S201,确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息。
当分析确定联邦学习请求中所包含的企业用户所请求参与的联邦学习模型的各请求事项,能够被构建出的联邦学习模型所执行时,检测联邦学习请求中所携带的输入该联邦学习请求的企业用户的企业标签信息。
在本实施例中,企业用户所输入的联邦学习服务请求中,还可以包括输入联邦学习服务请求的企业用户的企业标签信息(例如,用于标识企业用户的企业类型、数据类型等企业属性参数)。
例如,当分析确定企业用户所需要通过联邦学习解决的问题,能够通过构建联邦学习模型进行解决、以及构建出的联邦学习模型也能够满足企业用户所请求学习模型的性能指标之后,开始检测企业用户所输入的联邦学习服务请求中的企业标签信息,诸如企业用户的企业类型、数据类型等企业属性参数。
步骤S202,根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
根据检测到的联邦学习请求中所携带的企业用户的企业标签信息,从预设企业信息库中所保存的全部企业用户中,查找确定适于建立合作关系的目标企业,以作为联盟成员与当前输入联邦学习请求的企业用户组建联邦学习联盟。
在本实施例中,预设企业信息库为用于存储预先收集的愿意(在有报酬的情况下愿意)参与联邦学习的企业用户各属性参数的数据库,其中,该预设企业信息库中保存各希望参与联邦学习企业用户的一个或者多个属性参数,例如:企业类型、企业所拥有数据类型和企业业务类型,以及企业用户拥有的数据量,拥有的数据特征,数据的标签,企业数据能够完成的机器学习任务(例如,检测、预测或者分类);各企业用户所期望的联邦学习模型性能指标,能够承担的联邦学习开销(例如,通信开销,时间开销,计算资源,电量开销),希望获得的联邦学习收益(例如,模型性能提升幅度,或者是报酬)。
例如,根据检测到的当前企业用户所输入联邦学习请求中携带的,当前希望参与联邦学习的企业用户的企业用户的企业类型、数据类型等企业属性参数,从保存各希望参与联邦学习企业用户的一个或者多个属性参数的预设企业信息库中,逐一查找确定出与当前输入联邦学习的企业用户的企业类型、数据类型等企业属性参数相同或者互补的目标企业以作为联盟成员,与当前输入联邦学习的企业用户共同组建成联邦学习联盟。
步骤S300,提取所述联邦学习联盟中各联盟成员的成员信息。
检测并提取出依据联邦学习请求中所携带的企业用户的企业标签信息组建成的联邦学习联盟中,各联盟成员的成员信息。
在本实施例中,检测的联邦学习联盟中各联盟成员的成员信息包括但不限于:数据信息(例如:企业拥有的数据量,拥有的数据特征,数据的标签),以及,各企业用户所希望基于联邦学习完成的任务信息(例如:人脸识别,贷款风控,商品推荐等)。
进一步地,请参照图3,图3为本申请联邦学习方法一实施例中步骤S300的细化流程示意图,步骤S300,提取所述联邦学习联盟中各联盟成员的成员信息,包括:
步骤S301,检测所述预设企业信息库保存的全部企业参数。
例如,在存储有预先收集的希望参与联邦学习的企业用户各属性参数的预设企业信息库中,检测该预设企业信息库中保存的各希望参与联邦学习企业用户的全部企业参数,包括:企业用户的属性参数,例如:企业类型、企业所拥有数据类型和企业业务类型,以及企业用户拥有的数据量,拥有的数据特征,数据的标签,企业数据能够完成的机器学习任务(例如,检测、预测或者分类);各企业用户所期望的联邦学习模型性能指标,能够承担的联邦学习开销(例如,通信开销,时间开销,计算资源,电量开销),希望获得的联邦学习收益(例如,模型性能提升幅度,或者是现金报酬)。
步骤S302,从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息。
在检测到当前组建的联邦学习联盟各联盟成员的成员信息中,当前联盟成员保存有标签信息时,从检测到的预设企业信息库中保存的各企业用户全部企业参数中,遍历各联盟成员所对应企业用户的企业参数,从而索引出当前联邦学习联盟中,各联盟成员的企业标签信息。
例如,在检测到当前依据联邦学习请求中所携带的企业用户的企业标签信息(企业类型、数据类型等企业属性参数)组建成的联邦学习联盟中,各联盟成员同样保存有企业标签信息时,从检测到的预设企业信息库中保存的各企业用户全部企业参数(各希望参与联邦学习企业用户的一个或者多个属性参数,例如:企业类型、企业所拥有数据类型和企业业务类型,以及企业用户拥有的数据量,拥有的数据特征,数据的标签,企业数据能够完成的机器学习任务,例如,预测或者分类;各企业用户所期望的联邦学习模型性能指标,能够承担的联邦学习开销,例如,通信开销,时间开销,计算资源,电量开销,希望获得的联邦学习收益,例如,模型性能提升幅度)中,遍历各联盟成员所对应企业用户的企业参数,从而索引出当前联邦学习联盟中,各联盟成员的企业标签信息(即,各希望参与联邦学习企业用户的一个或者多个属性参数;数据信息,例如:企业拥有的数据量,拥有的数据特征,数据的标签;和,各企业用户所希望基于联邦学习完成的任务信息,例如:人脸识别,贷款风控,商品推荐等)。
步骤S303,逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
例如,在预设企业信息库所保存的各联盟成员所对应全部企业参数中,逐一从检测到的各联盟成员的标签信息,即,各联盟成员的一个或者多个属性参数,以及数据信息,例如:企业拥有的数据量,拥有的数据特征,数据的标签;和,各企业用户所希望基于联邦学习完成的任务信息,例如:人脸识别,贷款风控,商品推荐等,提取出数据信息(企业拥有的数据量,拥有的数据特征,数据的标签)以及各企业用户所希望基于联邦学习完成的任务信息(人脸识别,贷款风控,商品推荐等),作为当前各联邦学习联盟中各联盟成员的成员信息。
步骤S400,基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
根据检测到的当前组建成的联邦学习联盟中,各联盟成员的成员信息中,各联盟成员的数据信息和任务信息,构建出适于当前联邦学习联盟中各联盟成员进行联邦学习的联邦学习模型,并在联邦学习模型构建完成后,对构建的联邦学习模型进行管理。
例如,基于从预设企业信息库中检测到的当前联邦学习联盟中,各联盟成员的数据信息--企业用户拥有的数据量、拥有的数据特征、数据的标签,各企业用户所希望基于联邦学习完成的任务信息--人脸识别、贷款风控、商品推荐等成员信息,以及各联盟成员所能够承担的开销--通信、时间、计算资源开销和希望基于当前联邦学习获取到的收益--模型性能要求、模型性能提升幅度等,确定当前联邦学习联盟中的联盟成员(联邦学习参与者)的开销和利益分配(例如,根据每个参与者的数据量和对联邦学习模型的贡献来确定奖励分配,基于回归模型、神经网络预测模型、神经网络分类模型等,确定各联盟成员的开销和收益的分配策略),从而构建出适用于当前联邦学习联盟中的各联盟成员进行联邦学习的联邦学习模型,在完成联邦学习模型的构建之后,对联邦学习模型的模型训练以及模型更新等进行管理,其中,对联邦学习模型的模型训练进行管理至少包括:管理联邦学习模型的重新训练、继续训练,或者,管理联邦学习模型的重新设计和训练。
进一步地,步骤S400,包括:
步骤S401,依据提取出的所述成员信息,设计联邦学习模型的模型结构。
例如,根据各联盟成员的数据信息--企业用户拥有的数据量、拥有的数据特征、数据的标签,各企业用户所希望基于联邦学习完成的任务信息--人脸识别、贷款风控、商品推荐等成员信息来设计模型结构,或者在另一个实施例中,本申请联邦学习方法还可以根据现有的、经过实践证明的深度学习模型来设计联邦学习模型的模型结构。
步骤S402,在完成所述联邦学习模型的模型结构设计之后,初始化联邦学习模型。
例如,在根据各联盟成员的数据信息以及任务信息设计完成当前联邦学习模型的模型结构之后,通过随机初始化来初始化已经设计完成的当前联邦学习模型的模型参数,从而确保当前构建的联邦学习模型能够适用于当前联邦学习联盟的各联盟成员进行联邦学习。
本申请通过当检测到企业用户基于上述联邦学习方法设备输入联邦学习请求时,从该联邦学习方法设备设置的输入接口上,获取该企业用户所输入的联邦学习请求;在从联邦学习方法设备设置的输入接口上,获取该企业用户所输入的联邦学习请求之后,从获取的联邦学习请求中,提取出该联邦学习请求中所包含的企业用户所请求参与的联邦学习模型的各请求事项,并进一步分析该各请求事项能否被构建出的联邦学习模型所执行;当分析确定联邦学习请求中所包含的企业用户所请求参与的联邦学习模型的各请求事项,能够被构建出的联邦学习模型所执行时,检测联邦学习请求中所携带的输入该联邦学习请求的企业用户的企业标签信息,根据检测到的联邦学习请求中所携带的企业用户的企业标签信息,从预设企业信息库中所保存的全部企业用户中,查找确定适于建立合作关系的目标企业,以作为联盟成员与当前输入联邦学习请求的企业用户组建联邦学习联盟;检测依据联邦学习请求中所携带的企业用户的企业标签信息组建成的联邦学习联盟中,各联盟成员的成员信息;根据检测到的当前组建成的联邦学习联盟中,各联盟成员的成员信息中,各联盟成员的数据信息和任务信息,构建出适于当前联邦学习联盟中各联盟成员进行联邦学习的联邦学习模型,并在联邦学习模型构建完成后,对构建的联邦学习模型进行管理。
实现了,基于接收希望通过联邦学习建模的企业通过输入接口所输入的联邦学习请求,筛选出适合的企业作为该提出联邦学习请求企业的合作伙伴,以而组建成联邦学习联盟,根据组建的联邦学习联盟各联盟成员所携带的成员信息,构建成适于各联盟成员进行联邦学习的联邦学习模型。从而,无需提出联邦学习请求的企业实地逐一的向其他企业进行沟通,以说服其他企业作为自己的合作伙伴构建联邦学习模型,节省了希望进行联邦学习的企业构建联邦学习模型过程中的人力、物力和时间资源,降低了联邦学习模型构建的成本。
进一步地,提出本申请联邦学习方法的第二实施例。
基于上述联邦学习方法第一实施例,本实施例中,在上述步骤S400中,对所述联邦学习模型进行管理,包括:
步骤S403,管理构建的所述联邦学习模型的模型训练和模型更新。
在构建完成适用于当前联邦学习联盟中,各联盟成员进行联邦学习的联邦学习模型之后,依据各联盟成员基于联邦学习方法设备所提出的管理请求,对构建完成的联邦学习模型的模型训练和模型更新进行管理,或者,基于预设管理策略,对当前构建完成的联邦学习模型的模型训练和模型更新更新管理。
进一步地,步骤S403,包括:
步骤S4031,检测到获取的联邦学习请求为管理请求,依据所述管理请求对所述联邦学习模型进行模型训练和模型更新管理。
当检测到从该联邦学习方法设备设置的输入接口上,获取到的该企业用户所输入的联邦学习请求为对服务当前企业用户的联邦学习模型进行管理的请求时,根据该企业用户所输入更新请求的内容,对服务当前企业用户的联邦学习模型进行模型训练和模型更新。
例如,当检测到获取到的企业用户基于联邦学习方法设备--PC终端,所设置服务请求的输入窗口中输入的联邦学习服务请求,具体为对已经构建完成以服务于当前企业用户的联邦学习模型进行更新处理的管理请求时,根据该企业用户所输入管理请求的内容,例如,用户所输入管理请求的内容为:对当前联邦学习模型的进行模型训练,则根据该管理请求,管理联邦学习模型进行重新训练、继续训练,或者,管理联邦学习模型重新进行模型设计以及模型训练,或者,用户所输入管理请求的内容为:对当前联邦学习联盟中的联盟成员(联邦学习参与者)进行更新,则根据当前更新请求,从预设企业信息库中,重新查找确定目标企业作为新的联盟成员,即重新选择联邦学习参与者。
步骤S4032,按照预设模型管理策略,周期性的对所述联邦学习模型进行模型训练和模型更新管理。
本实施例中,预设模型管理策略为,预先基于构建的联邦学习模型的稳定性等特征,设置的对当前联邦学习模型进行定期更新处理的模型管理策略。
例如,从联邦学习模型构建完成之时起,按照预定时间周期(例如,一星期)自动对联邦学习模型进行模型训练以及进行更新处理,其中,对联邦学习模型的模型训练进行管理至少包括:管理联邦学习模型的重新训练、继续训练,或者,管理联邦学习模型的重新设计和训练;对联邦学习模型的更新处理包括:模型参数的更新,即对于当前联邦学习模型进行再次训练,例如,对同一个神经网络结构再次训练神经网络的参数,从而获得当前联邦学习模型新的模型参数;模型结构的更新,即更新当前联邦学习模型的模型结构,并重新训练,例如,改变神经网络的结构,重新训练神经网络的参数,从而获得新的联邦学习模型。
本申请在构建完成适用于当前联邦学习联盟中,各联盟成员进行联邦学习的联邦学习模型之后,依据各联盟成员基于联邦学习方法设备所提出的管理请求,对构建完成的联邦学习模型进行模型训练和模型更新管理,或者,基于预设模型管理策略,对当前构建完成的联邦学习模型训练和模型进行更新管理。当检测到从该联邦学习方法设备设置的输入接口上,获取到的该企业用户所输入的联邦学习请求为对服务当前企业用户的联邦学习模型进行管理的请求时,根据该企业用户所输入管理请求的内容,对服务当前企业用户的联邦学习模型的模型训练以及模型更新进行管理;以及,从联邦学习模型构建完成之时起,按照预定时间周期自动对联邦学习模型进行模型训练以及模型更新处理,其中,对联邦学习模型的模型训练进行管理包括:管理联邦学习模型的重新训练、继续训练,或者,管理联邦学习模型的重新设计和训练;对联邦学习模型的更新处理包括:模型参数的更新和模型结构的更新。
实现了,在构建完成适用提出联邦学习请求的企业用户的联邦学习模型之后,根据企业用户输入的联邦学习请求中对当前联邦学习模型进行管理处理的管理请求,对联邦学习模型的模型训练以及模型更行进行管理,或者,依据预先制定的模型管理策略自动周期性的对构建完成的联邦学习模型的模型训练以及模型更行进行管理,从而,保证了构建的联邦学习模型在服务企业用户过程中的高效性,节省了参与联邦学习的请求者自主对模型进行更新处理的成本,进一步提高了联邦学习模型的创建效率。
进一步地,提出本申请联邦学习方法的第三实施例。
基于上述联邦学习方法第一实施例和第二实施例,在本实施例中,本申请联邦学习方法,还包括:
步骤A,根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
根据从输入接口获取到的企业用户所输入联邦学习请求中,该企业用户所希望的请求事项,向当前企业用户或者当前企业用户所处联邦学习联盟的其他各联盟成员提供服务。
例如,当检测到获取到的企业用户基于联邦学习方法设备--PC终端,所设置服务请求的输入窗口中输入的联邦学习服务请求,具体为对已经构建完成以服务于当前企业用户的联邦学习模型进行更新处理的更新请求时,对当前构建完成的联邦学习模型进行更新处理;进一步地,基于当前企业用户所输入的希望进行联邦学习训练以完成贷款风控预测操作的服务请求时,向当前企业用户提供进行联邦学习模型训练得出的预测结果;更进一步的,当希望参与联邦学习的企业用户自身携带有本地训练模型时,基于企业用户所输入联邦学习模型参数请求,向当前企业用户提供当前构建完成的联邦学习模型的模型参数以及联邦学习代码,以供当前企业用户自己基于自身所拥有的数据进行本地的机器学习。
进一步地,在另一个实施例中,当分析检测到获取到的企业用户基于PC终端上所设置的供企业用户输入服务请求的输入窗口中,所输入的联邦学习服务请求中,企业用户所需要通过联邦学习解决的问题,不能通过构建联邦学习模型进行解决,和/或者,构建出的联邦学习模型不能够满足企业用户所请求学习模型的性能指标等请求项时,通过当前联邦学习方法设备上,所设置的输出接口(具体可以为企业用户输入服务请求的反馈窗口)向当前企业用户输出“联邦学习服务请求”被拒绝的提示信息。
本申请根据从输入接口获取到的企业用户所输入联邦学习请求中,该企业用户所希望的请求事项,向当前企业用户或者当前企业用户所处联邦学习联盟的其他各联盟成员提供服务,即检测到获取到的企业用户输入的联邦学习服务请求,具体为对已经构建完成以服务于当前企业用户的联邦学习模型进行更新处理的更新请求时,对当前构建完成的联邦学习模型进行更新处理;基于企业用户所输入联邦学习模型参数请求,向当前企业用户提供当前构建完成的联邦学习模型的模型参数以及联邦学习代码,以供当前企业用户自己基于自身所拥有的数据进行本地的机器学习等。
实现了,企业用户仅需要将该企业自身所拥有的数据的情况、希望通过机器学习处理的任务、自身能承担的开销以及期望的收益等,通过联邦学习方法设备进行输入以生成希望进行联邦学习的服务请求,联邦学习方法设备即可自动根据该服务请求所包含的各请求事项,为当前企业用户寻找联邦学习合作伙伴、设计模型结构并初始化联邦学习模型,进而将初始联邦学习模型发送给该企业用户、以及进行构建完成的联邦学习模型的训练和管理等,无需希望参与联邦学习的企业用户自主进行联邦学习方法、训练运行、管理以及维护等,节省了联邦学习参与者的人力、物力和时间成本,提高了联邦学习方法效率。
此外,请参照图4,本申请实施例还提出一种联邦学习系统,所述联邦学习系统包括:
获取模块,设置为获取输入接口所接收到的联邦学习请求;
联盟组建模块,设置为根据所述联邦学习请求组建联邦学习联盟;
检测模块,设置为检测所述联邦学习联盟中,各联盟成员的成员信息;以及,
模型管理模块,设置为基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
可选地,所述联邦学习系统,还包括:
分析模块,设置为提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行。
可选地,联盟组建模块,包括:
第二检测单元,设置为确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
联盟组建子单元,设置为根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
可选地,检测模块,包括:
第一检测单元,设置为检测所述预设企业信息库保存的全部企业参数;
索引单元,设置为从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息;
提取单元,设置为逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
可选地,模型构建模块,包括:
设计单元,设置为依据提取出的所述成员信息,设计联邦学习模型的模型结构;
初始化单元,设置为在完成所述联邦学习模型的模型结构设计之后,初始化联邦学习模型。
可选地,模型管理模块,包括:
管理单元,设置为管理构建的所述联邦学习模型的模型训练和模型更新。
可选地,管理单元,包括:
第一管理子单元,设置为检测到获取的联邦学习请求为管理请求,依据所述管理请求对所述联邦学习模型进行模型训练和模型更新管理;
第二管理子单元,设置为按照预设管理策略,周期性的对所述联邦学习模型进行模型训练和模型更新管理。
可选地,所述联邦学习系统,还包括:
服务模块,设置为根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
可选地,服务模块,包括:
模型分发单元,设置为分发构建完成的联邦学习模型至所述联邦学习联盟的联盟成员;或者,
参数供给单元,设置为向所述联盟成员提供模型训练参数,以供所述联盟成员自主进行模型训练。
本实施例提出的联邦学习系统各个模块运行时实现如上所述的联邦学习方法的步骤,在此不再赘述。
此外,本申请实施例还提出一种存储介质,应用于计算机,即所述存储介质为计算机可读存储介质,所述计算机可读存储介质可以为非易失性可读存储介质。
所述介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时实现如上所述的联邦学习方法的步骤。
其中,在所述处理器上运行的计算机可读指令被执行时所实现的方法可参照本申请基于联邦学习方法各个实施例,此处不再赘述。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种联邦学习方法,其中,所述联邦学习方法包括以下步骤:
    获取输入接口所接收到的联邦学习请求;
    根据所述联邦学习请求组建联邦学习联盟;
    提取所述联邦学习联盟中各联盟成员的成员信息;以及,
    基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
  2. 如权利要求1所述的联邦学习方法,其中,在所述获取输入接口所接收到的联邦学习请求的步骤之后,还包括:
    提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行;
    所述根据所述联邦学习请求组建联邦学习联盟的步骤包括:
    确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
    根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
  3. 如权利要求2所述的联邦学习方法,其中,所述提取所述联邦学习联盟中各联盟成员的成员信息的步骤包括:
    检测所述预设企业信息库保存的全部企业参数;
    从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息;
    逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
  4. 如权利要求3所述的联邦学习方法,其中,所述基于所述成员信息构建联邦学习模型的步骤包括:
    依据提取出的所述成员信息,设计联邦学习模型的模型结构;
    在完成所述联邦学习模型的模型结构设计之后,初始化联邦学习模型。
  5. 如权利要求1所述的联邦学习方法,其中,所述对所述联邦学习模型进行管理的步骤,包括:
    管理构建的所述联邦学习模型的模型训练和模型更新;
    所述管理构建的所述联邦学习模型的模型训练和模型更新的步骤包括:
    检测到获取的联邦学习请求为管理请求,依据所述管理请求对所述联邦学习模型进行模型训练和模型更新管理;或者,
    按照预设管理策略,周期性的对所述联邦学习模型进行模型训练和模型更新管理。
  6. 如权利要求1所述的联邦学习方法,其中,所述联邦学习方法,还包括:
    根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
  7. 如权利要求6所述的联邦学习方法,其中,所述对所述联邦学习联盟的联盟成员进行服务的步骤包括:
    分发构建完成的联邦学习模型至所述联邦学习联盟的联盟成员;或者,
    向所述联盟成员提供模型训练参数,以供所述联盟成员自主进行模型训练。
  8. 一种联邦学习系统,其中,所述联邦学习系统包括:
    获取模块,设置为获取输入接口所接收到的联邦学习请求;
    联盟组建模块,设置为根据所述联邦学习请求组建联邦学习联盟;
    检测模块,设置为提取所述联邦学习联盟中各联盟成员的成员信息;以及,
    模型管理模块,设置为基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
  9. 如权利要求8所述的联邦学习系统,其中,所述联邦学习系统,还包括:
    分析模块,设置为提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行。
  10. 如权利要求8所述的联邦学习系统,其中,所述联盟组建模块,包括:
    第二检测单元,设置为确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
    联盟组建子单元,设置为根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
  11. 如权利要求8所述的联邦学习系统,其中,所述模型管理模块,包括:
    管理单元,设置为管理构建的所述联邦学习模型的模型训练和模型更新。
  12. 如权利要求8所述的联邦学习系统,其中,所述联邦学习系统,还包括:
    服务模块,用于根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
  13. 一种终端设备,其中,所述终端设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机可读指令,所述计算机可读指令被所述处理器执行时,实现如下步骤:
    获取输入接口所接收到的联邦学习请求;
    根据所述联邦学习请求组建联邦学习联盟;
    提取所述联邦学习联盟中各联盟成员的成员信息;以及,
    基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
  14. 如权利要求13所述的终端设备,其中,在所述获取输入接口所接收到的联邦学习请求的步骤之后,还包括:
    提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行;
    所述根据所述联邦学习请求组建联邦学习联盟的步骤包括:
    确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
    根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
  15. 如权利要求14所述的终端设备,其中,所述提取所述联邦学习联盟中各联盟成员的成员信息的步骤包括:
    检测所述预设企业信息库保存的全部企业参数;
    从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息;
    逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
  16. 如权利要求13所述的终端设备,其中,所述计算机可读指令被所述处理器执行时,还实现如下步骤:
    根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
  17. 一种存储介质,其中,应用于计算机,所述存储介质上存储有计算机可读指令,所述计算机可读指令被处理器执行时,实现如下步骤:
    获取输入接口所接收到的联邦学习请求;
    根据所述联邦学习请求组建联邦学习联盟;
    提取所述联邦学习联盟中各联盟成员的成员信息;以及,
    基于所述成员信息构建联邦学习模型,并对所述联邦学习模型进行管理。
  18. 如权利要求17所述的存储介质,其中,在所述获取输入接口所接收到的联邦学习请求的步骤之后,还包括:
    提取所述联邦学习请求中携带的请求事项,并确定所述请求事项能够被执行;
    所述根据所述联邦学习请求组建联邦学习联盟的步骤包括:
    确定到所述请求事项能够被执行,则检测所述联邦学习请求中携带的企业标签信息;
    根据所述企业标签信息从预设企业信息库中匹配目标企业作为联盟成员,以组建联邦学习联盟。
  19. 如权利要求18所述的存储介质,其中,所述提取所述联邦学习联盟中各联盟成员的成员信息的步骤包括:
    检测所述预设企业信息库保存的全部企业参数;
    从全部所述企业参数中,索引各所述联盟成员所对应目标企业的所述企业标签信息;
    逐一从所述企业标签信息中,提取出组建所述联邦学习联盟所需的各所述联邦成员的成员信息。
  20. 如权利要求17所述的存储介质,其中,所述计算机可读指令被所述处理器执行时,还实现如下步骤:
    根据所述联邦学习请求的请求事项,对所述联邦学习联盟的联盟成员进行服务。
PCT/CN2019/116464 2019-07-17 2019-11-08 联邦学习方法、系统、终端设备及存储介质 Ceased WO2021008017A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910647743.0 2019-07-17
CN201910647743.0A CN110363305B (zh) 2019-07-17 2019-07-17 联邦学习方法、系统、终端设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021008017A1 true WO2021008017A1 (zh) 2021-01-21

Family

ID=68221208

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/116464 Ceased WO2021008017A1 (zh) 2019-07-17 2019-11-08 联邦学习方法、系统、终端设备及存储介质

Country Status (2)

Country Link
CN (1) CN110363305B (zh)
WO (1) WO2021008017A1 (zh)

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112949866A (zh) * 2021-03-25 2021-06-11 深圳前海微众银行股份有限公司 泊松回归模型的训练方法、装置、电子设备及存储介质
CN113034295A (zh) * 2021-04-21 2021-06-25 中国建设银行股份有限公司 险种推荐方法、装置、电子设备及存储介质
CN113158223A (zh) * 2021-01-27 2021-07-23 深圳前海微众银行股份有限公司 基于状态转移核优化的数据处理方法、装置、设备及介质
CN113157434A (zh) * 2021-02-26 2021-07-23 西安电子科技大学 一种横向联邦学习系统用户节点的激励方法及系统
CN113268920A (zh) * 2021-05-11 2021-08-17 西安交通大学 一种基于联邦学习的无人机群感知数据安全共享方法
CN113409134A (zh) * 2021-06-30 2021-09-17 中国工商银行股份有限公司 基于联邦学习的企业融资授信方法及装置
CN113469371A (zh) * 2021-07-01 2021-10-01 建信金融科技有限责任公司 联邦学习方法和装置
CN113505520A (zh) * 2021-05-17 2021-10-15 京东科技控股股份有限公司 用于支持异构联邦学习的方法、装置和系统
CN113569151A (zh) * 2021-09-18 2021-10-29 平安科技(深圳)有限公司 基于人工智能的数据推荐方法、装置、设备及介质
CN113836809A (zh) * 2021-09-26 2021-12-24 上海万向区块链股份公司 基于区块链和联邦学习的跨行业数据联合建模方法及系统
CN114021891A (zh) * 2021-09-30 2022-02-08 医渡云(北京)技术有限公司 基于联邦学习的生存分析方法、装置及电子设备
CN114202018A (zh) * 2021-11-29 2022-03-18 新智我来网络科技有限公司 一种模块化的联合学习方法及系统
CN114416854A (zh) * 2022-03-29 2022-04-29 北京安华金和科技有限公司 一种数据转换处理方法和系统
CN114647982A (zh) * 2022-03-31 2022-06-21 东北大学 一种基于动态契约理论的车联网联邦学习激励方法
CN114707309A (zh) * 2022-03-18 2022-07-05 浙江清华长三角研究院 一种松耦合联邦系统链路耦合度计算方法及计算机设备
CN114936892A (zh) * 2022-04-24 2022-08-23 福州大学 基于联邦迁移学习标签传播的线下商品推荐系统及方法
CN115130682A (zh) * 2022-06-29 2022-09-30 蓝象智联(杭州)科技有限公司 多模式的联邦学习方法及装置
CN115499307A (zh) * 2022-08-05 2022-12-20 中山大学 面向边缘联邦学习部署架构、方法、装置及存储介质
CN115514761A (zh) * 2021-06-03 2022-12-23 中移动信息技术有限公司 联邦学习环境下的数据共享与协同方法及系统
CN115526365A (zh) * 2021-06-24 2022-12-27 中兴通讯股份有限公司 指标优化方法及服务器、计算机可读存储介质
CN115618969A (zh) * 2022-12-19 2023-01-17 湖南工商大学 基于聚类与环形拓扑的数据选择方法
WO2023040958A1 (zh) * 2021-09-18 2023-03-23 大唐移动通信设备有限公司 一种联邦学习组处理方法、装置及功能实体
CN115914257A (zh) * 2022-11-02 2023-04-04 卫盈联信息技术(深圳)有限公司 基于联邦学习的状态交互方法、系统、设备及存储介质
CN116055150A (zh) * 2022-12-22 2023-05-02 深圳信息职业技术学院 车联网入侵检测平台、方法及相关设备
CN116050548A (zh) * 2023-03-27 2023-05-02 深圳前海环融联易信息科技服务有限公司 一种联邦学习方法、装置及电子设备
CN116128050A (zh) * 2021-11-12 2023-05-16 华为技术有限公司 一种模型训练方法、系统以及相关装置
CN116226158A (zh) * 2021-11-29 2023-06-06 新智我来网络科技有限公司 联合学习任务加入方法、装置、电子设备及存储介质
CN116245172A (zh) * 2023-03-14 2023-06-09 南京航空航天大学 跨孤岛联邦学习中面向个体模型性能优化的联盟组建博弈方法
CN116450738A (zh) * 2023-04-11 2023-07-18 浙江大学 一种基于区块链的深度神经网络模型联邦重组方法及系统
CN116567702A (zh) * 2022-01-26 2023-08-08 展讯通信(上海)有限公司 一种用户设备选择方法、装置、芯片及模组设备
CN119210465A (zh) * 2024-11-07 2024-12-27 湖南科技大学 一种自适应压缩编码的分层联邦学习方法

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363305B (zh) * 2019-07-17 2023-09-26 深圳前海微众银行股份有限公司 联邦学习方法、系统、终端设备及存储介质
CN110851482B (zh) * 2019-11-07 2022-02-18 支付宝(杭州)信息技术有限公司 为多个数据方提供数据模型的方法及装置
CN110990870A (zh) * 2019-11-29 2020-04-10 上海能塔智能科技有限公司 运维、使用模型库的处理方法、装置、设备与介质
CN111222646B (zh) * 2019-12-11 2021-07-30 深圳逻辑汇科技有限公司 联邦学习机制的设计方法、装置和存储介质
CN113065610B (zh) * 2019-12-12 2022-05-17 支付宝(杭州)信息技术有限公司 基于联邦学习的孤立森林模型构建和预测方法和装置
CN111062493B (zh) * 2019-12-20 2021-06-15 深圳前海微众银行股份有限公司 基于公共数据的纵向联邦方法、装置、设备及介质
CN111340558B (zh) * 2020-02-28 2021-07-27 深圳前海微众银行股份有限公司 基于联邦学习的线上信息处理方法、装置、设备及介质
WO2021179176A1 (en) * 2020-03-10 2021-09-16 Asiainfo Technologies (China) , Inc. Federated learning in telecom communication system
CN113496291A (zh) * 2020-03-18 2021-10-12 索尼公司 用于联邦学习的装置、方法和存储介质
CN111428620B (zh) * 2020-03-20 2025-02-07 深圳前海微众银行股份有限公司 基于联邦活体检测模型的身份识别方法、装置、设备和介质
CN111723942B (zh) * 2020-06-29 2024-02-02 南方电网科学研究院有限责任公司 一种企业用电负荷预测方法、电网业务子系统及预测系统
CN111935157B (zh) * 2020-08-12 2022-05-06 科技谷(厦门)信息技术有限公司 一种基于安全防御的联盟学习系统
CN112329947A (zh) * 2020-10-28 2021-02-05 广州中国科学院软件应用技术研究所 一种基于差分进化的联邦学习激励方法和系统
CN114490705B (zh) * 2020-11-13 2026-02-06 深圳前海微众银行股份有限公司 信息处理方法、装置、设备及存储介质
CN114625490B (zh) * 2020-12-08 2025-08-29 新奥新智科技有限公司 一种基于联合学习的任务分配方法及装置
CN114840736B (zh) * 2021-01-14 2025-05-30 新奥新智科技有限公司 一种联合学习平台的整体交互方法
US11755954B2 (en) 2021-03-11 2023-09-12 International Business Machines Corporation Scheduled federated learning for enhanced search
CN113160021B (zh) * 2021-03-18 2022-04-12 天津中科物联科技有限公司 一种基于多源异构数据联邦学习的安全生产预警系统
WO2022236469A1 (en) * 2021-05-08 2022-11-17 Asiainfo Technologies (China), Inc. Customer experience perception based on federated learning
CN116226649A (zh) * 2021-11-29 2023-06-06 新智我来网络科技有限公司 联合学习训练数据筛选方法、装置、电子设备及存储介质
CN116226204B (zh) * 2021-11-29 2026-01-02 新奥新智科技有限公司 基于联合学习平台的场景确定方法、装置、设备及存储介质
CN116227616A (zh) * 2021-11-29 2023-06-06 新智我来网络科技有限公司 一种基于物联网的联合学习方法及装置
CN116226157A (zh) * 2021-11-29 2023-06-06 新智我来网络科技有限公司 联合学习任务创建方法、装置、电子设备及存储介质
CN116307138A (zh) * 2023-03-01 2023-06-23 中国科学院上海微系统与信息技术研究所 基于联邦学习的电气火灾预警方法、装置、终端及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109167695A (zh) * 2018-10-26 2019-01-08 深圳前海微众银行股份有限公司 基于联邦学习的联盟网络构建方法、设备及可读存储介质
CN109165515A (zh) * 2018-08-10 2019-01-08 深圳前海微众银行股份有限公司 基于联邦学习的模型参数获取方法、系统及可读存储介质
CN109299728A (zh) * 2018-08-10 2019-02-01 深圳前海微众银行股份有限公司 联邦学习方法、系统及可读存储介质
CN109492420A (zh) * 2018-12-28 2019-03-19 深圳前海微众银行股份有限公司 基于联邦学习的模型参数训练方法、终端、系统及介质
CN110363305A (zh) * 2019-07-17 2019-10-22 深圳前海微众银行股份有限公司 联邦学习方法、系统、终端设备及存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090254392A1 (en) * 2006-03-30 2009-10-08 Zander Van S Method and system for enterprise network access control and management for government and corporate entities
CN101106449B (zh) * 2006-07-13 2010-05-12 华为技术有限公司 实现多方通信安全的系统和方法
CN102663202A (zh) * 2012-04-25 2012-09-12 清华大学 基于联邦模式的动态产品协同开发平台及方法
CN103514321A (zh) * 2013-08-12 2014-01-15 北京理工大学 一种应用于hla分布式仿真方法的通用联邦成员
CN109840686A (zh) * 2018-12-24 2019-06-04 万翼科技有限公司 一种合伙人联盟平台及事件创建管理方法和系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109165515A (zh) * 2018-08-10 2019-01-08 深圳前海微众银行股份有限公司 基于联邦学习的模型参数获取方法、系统及可读存储介质
CN109299728A (zh) * 2018-08-10 2019-02-01 深圳前海微众银行股份有限公司 联邦学习方法、系统及可读存储介质
CN109167695A (zh) * 2018-10-26 2019-01-08 深圳前海微众银行股份有限公司 基于联邦学习的联盟网络构建方法、设备及可读存储介质
CN109492420A (zh) * 2018-12-28 2019-03-19 深圳前海微众银行股份有限公司 基于联邦学习的模型参数训练方法、终端、系统及介质
CN110363305A (zh) * 2019-07-17 2019-10-22 深圳前海微众银行股份有限公司 联邦学习方法、系统、终端设备及存储介质

Cited By (40)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113158223A (zh) * 2021-01-27 2021-07-23 深圳前海微众银行股份有限公司 基于状态转移核优化的数据处理方法、装置、设备及介质
CN113157434B (zh) * 2021-02-26 2024-05-07 西安电子科技大学 一种横向联邦学习系统用户节点的激励方法及系统
CN113157434A (zh) * 2021-02-26 2021-07-23 西安电子科技大学 一种横向联邦学习系统用户节点的激励方法及系统
CN112949866A (zh) * 2021-03-25 2021-06-11 深圳前海微众银行股份有限公司 泊松回归模型的训练方法、装置、电子设备及存储介质
CN113034295A (zh) * 2021-04-21 2021-06-25 中国建设银行股份有限公司 险种推荐方法、装置、电子设备及存储介质
CN113268920A (zh) * 2021-05-11 2021-08-17 西安交通大学 一种基于联邦学习的无人机群感知数据安全共享方法
CN113505520A (zh) * 2021-05-17 2021-10-15 京东科技控股股份有限公司 用于支持异构联邦学习的方法、装置和系统
CN115514761A (zh) * 2021-06-03 2022-12-23 中移动信息技术有限公司 联邦学习环境下的数据共享与协同方法及系统
CN115526365A (zh) * 2021-06-24 2022-12-27 中兴通讯股份有限公司 指标优化方法及服务器、计算机可读存储介质
CN113409134A (zh) * 2021-06-30 2021-09-17 中国工商银行股份有限公司 基于联邦学习的企业融资授信方法及装置
CN113469371A (zh) * 2021-07-01 2021-10-01 建信金融科技有限责任公司 联邦学习方法和装置
CN113569151A (zh) * 2021-09-18 2021-10-29 平安科技(深圳)有限公司 基于人工智能的数据推荐方法、装置、设备及介质
US12574300B2 (en) 2021-09-18 2026-03-10 Datang Mobile Communications Equipment Co., Ltd. Federated learning group processing method, device and functional entity
WO2023040958A1 (zh) * 2021-09-18 2023-03-23 大唐移动通信设备有限公司 一种联邦学习组处理方法、装置及功能实体
CN113836809B (zh) * 2021-09-26 2023-12-01 上海万向区块链股份公司 基于区块链和联邦学习的跨行业数据联合建模方法及系统
CN113836809A (zh) * 2021-09-26 2021-12-24 上海万向区块链股份公司 基于区块链和联邦学习的跨行业数据联合建模方法及系统
CN114021891A (zh) * 2021-09-30 2022-02-08 医渡云(北京)技术有限公司 基于联邦学习的生存分析方法、装置及电子设备
CN116128050A (zh) * 2021-11-12 2023-05-16 华为技术有限公司 一种模型训练方法、系统以及相关装置
CN114202018B (zh) * 2021-11-29 2025-04-15 新奥新智科技有限公司 一种模块化的联合学习方法及系统
CN114202018A (zh) * 2021-11-29 2022-03-18 新智我来网络科技有限公司 一种模块化的联合学习方法及系统
CN116226158A (zh) * 2021-11-29 2023-06-06 新智我来网络科技有限公司 联合学习任务加入方法、装置、电子设备及存储介质
CN116567702A (zh) * 2022-01-26 2023-08-08 展讯通信(上海)有限公司 一种用户设备选择方法、装置、芯片及模组设备
CN114707309B (zh) * 2022-03-18 2024-12-06 浙江清华长三角研究院 一种松耦合联邦系统链路耦合度计算方法及计算机设备
CN114707309A (zh) * 2022-03-18 2022-07-05 浙江清华长三角研究院 一种松耦合联邦系统链路耦合度计算方法及计算机设备
CN114416854A (zh) * 2022-03-29 2022-04-29 北京安华金和科技有限公司 一种数据转换处理方法和系统
CN114647982A (zh) * 2022-03-31 2022-06-21 东北大学 一种基于动态契约理论的车联网联邦学习激励方法
CN114936892A (zh) * 2022-04-24 2022-08-23 福州大学 基于联邦迁移学习标签传播的线下商品推荐系统及方法
CN115130682A (zh) * 2022-06-29 2022-09-30 蓝象智联(杭州)科技有限公司 多模式的联邦学习方法及装置
CN115499307A (zh) * 2022-08-05 2022-12-20 中山大学 面向边缘联邦学习部署架构、方法、装置及存储介质
CN115914257A (zh) * 2022-11-02 2023-04-04 卫盈联信息技术(深圳)有限公司 基于联邦学习的状态交互方法、系统、设备及存储介质
CN115618969A (zh) * 2022-12-19 2023-01-17 湖南工商大学 基于聚类与环形拓扑的数据选择方法
CN115618969B (zh) * 2022-12-19 2023-04-07 湖南工商大学 基于聚类与环形拓扑的数据选择方法
CN116055150B (zh) * 2022-12-22 2023-10-27 深圳信息职业技术学院 车联网入侵检测平台、方法及相关设备
CN116055150A (zh) * 2022-12-22 2023-05-02 深圳信息职业技术学院 车联网入侵检测平台、方法及相关设备
CN116245172A (zh) * 2023-03-14 2023-06-09 南京航空航天大学 跨孤岛联邦学习中面向个体模型性能优化的联盟组建博弈方法
CN116245172B (zh) * 2023-03-14 2023-10-17 南京航空航天大学 跨孤岛联邦学习中优化个体模型性能的联盟组建方法
CN116050548A (zh) * 2023-03-27 2023-05-02 深圳前海环融联易信息科技服务有限公司 一种联邦学习方法、装置及电子设备
CN116450738A (zh) * 2023-04-11 2023-07-18 浙江大学 一种基于区块链的深度神经网络模型联邦重组方法及系统
CN119210465A (zh) * 2024-11-07 2024-12-27 湖南科技大学 一种自适应压缩编码的分层联邦学习方法
CN119210465B (zh) * 2024-11-07 2025-02-14 湖南科技大学 一种自适应压缩编码的分层联邦学习方法

Also Published As

Publication number Publication date
CN110363305A (zh) 2019-10-22
CN110363305B (zh) 2023-09-26

Similar Documents

Publication Publication Date Title
WO2021008017A1 (zh) 联邦学习方法、系统、终端设备及存储介质
WO2020147385A1 (zh) 数据录入方法、装置、终端及计算机可读存储介质
WO2020073495A1 (zh) 基于人工智能的复审方法、装置、设备及存储介质
WO2021003930A1 (zh) 客服录音的质检方法、装置、设备及计算机可读存储介质
WO2020168606A1 (zh) 广告视频优化方法、装置、设备及计算机可读存储介质
WO2020111532A1 (ko) 복수 챗봇의 협업에 의한 대화형 정보제공 시스템 및 그 방법
WO2019144585A1 (zh) 虚拟机器人的控制方法、装置、设备及可读存储介质
WO2018082484A1 (zh) 一种电子设备的截屏方法、系统及电子设备
WO2015137641A2 (ko) 계층적인 룰 구조를 가지고 있는 비즈니스 룰 관리 시스템 및 그 표현 방법
WO2020153552A1 (ko) 기록된 거래의 수정이 가능한 블록체인을 위한 방법 및 장치
WO2018004236A1 (ko) 개인정보의 비식별화 방법 및 장치
WO2018076829A1 (zh) 终端数据的处理方法、装置、系统、存储介质及服务器
WO2018166098A1 (zh) 数据稽核方法、装置、移动终端及计算机可读存储介质
WO2019019245A1 (zh) 基金组合购买方法、系统及存储介质
WO2018201774A1 (zh) 数据审批方法、装置、设备和计算机可读存储介质
WO2018084326A1 (ko) 실시간 상담을 제공하기 위한 방법 및 서버
WO2011010815A2 (ko) 온라인 가치공학 프로세스 처리 시스템
WO2025178213A1 (ko) 서비스 데이터 처리 방법, 장치 및 기록 매체
WO2018205816A1 (zh) 基于线上线下一体化服务的方法、设备、装置和存储介质
WO2024117292A1 (ko) 메타버스 연동 챗봇 로봇 시스템
WO2020101081A1 (ko) 고 영향 노드의 결정 방법, 고 영향 노드의 결정 장치, 바이럴 마케팅에서의 마케팅 비용 최소화 방법 및 이를 수행하기 위한 시스템
WO2023224423A1 (ko) 웹사이트 자동 생성 플랫폼 운영 방법 및 시스템
WO2024257937A1 (ko) 전자 장치 및 그의 정보 제공 방법
WO2021042564A1 (zh) 检索方法
WO2023229089A1 (ko) 주문 관리를 위한 전자 장치 및 그 방법

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19937770

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19937770

Country of ref document: EP

Kind code of ref document: A1