WO2022262183A1 - 联邦计算的处理方法、装置、电子设备和存储介质 - Google Patents

联邦计算的处理方法、装置、电子设备和存储介质 Download PDF

Info

Publication number
WO2022262183A1
WO2022262183A1 PCT/CN2021/130307 CN2021130307W WO2022262183A1 WO 2022262183 A1 WO2022262183 A1 WO 2022262183A1 CN 2021130307 W CN2021130307 W CN 2021130307W WO 2022262183 A1 WO2022262183 A1 WO 2022262183A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
task
federated
task identifier
data processing
Prior art date
Application number
PCT/CN2021/130307
Other languages
English (en)
French (fr)
Inventor
季石磊
廖源
黄海平
刘吉
Original Assignee
百度在线网络技术（北京）有限公司
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 百度在线网络技术（北京）有限公司 filed Critical 百度在线网络技术（北京）有限公司
Priority to US18/012,967 priority Critical patent/US20230251898A1/en
Priority to EP21945762.9A priority patent/EP4160440A4/en
Priority to JP2022581625A priority patent/JP2023534917A/ja
Priority to KR1020227045230A priority patent/KR20230006033A/ko
Publication of WO2022262183A1 publication Critical patent/WO2022262183A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/48—Program initiating; Program switching, e.g. by interrupt
    • G06F9/4806—Task transfer initiation or dispatching
    • G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
    • G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/50—Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems
    • G06F21/57—Certifying or maintaining trusted computer platforms, e.g. secure boots or power-downs, version controls, system software checks, secure updates or assessing vulnerabilities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5066—Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • the present disclosure relates to the technical field of artificial intelligence, in particular to the technical fields of big data, cloud computing, and deep learning, and in particular to a processing method, device, electronic device, and storage medium for federated computing.
  • Federated computing refers to the use of privacy and security technologies such as Multi-Party Computation (MPC), Trusted Execution Environment (TEE), and Federated Learning (FL) to provide data for multi-party data security cooperation.
  • MPC Multi-Party Computation
  • TEE Trusted Execution Environment
  • FL Federated Learning
  • the disclosure proposes a federated computing processing method, device, electronic equipment and storage medium.
  • a federated computing processing method including:
  • n data slices corresponding to each participant are generated;
  • the federation calculation result corresponding to the task identifier is determined.
  • a federated computing processing device including:
  • An acquisition module configured to acquire metadata to be processed from each participant based on the task identifier to be executed
  • a segmentation module configured to segment each of the metadata to obtain n data slices corresponding to each participant, where n is a positive integer greater than 1;
  • a generating module configured to generate n data sets according to n data slices corresponding to each participant
  • a processing module configured to call n data processing services corresponding to the task identifier, so as to process each of the data sets by using each of the data processing services;
  • the determination module is configured to determine the federated calculation result corresponding to the task identifier according to the processing result of each data processing service.
  • an electronic device including:
  • the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can perform the federated computing described in the above-mentioned one aspect embodiment. Approach.
  • a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the federated computing according to the above-mentioned one embodiment Approach.
  • a computer program product including a computer program.
  • the computer program is executed by a processor, the federated computing processing method according to the above-mentioned one embodiment is implemented.
  • the metadata to be processed is obtained from each participant based on the task identifier to be executed, and each metadata is segmented to obtain n data slices corresponding to each participant, and then According to the n data slices corresponding to each participant, n data sets are generated, and then n data processing services corresponding to the task identifier are called to process each data set with each data processing service, and finally according to each The processing result of the data processing service determines the federation calculation result corresponding to the task ID. Therefore, when performing federated computing, the parallelization of federated learning is realized by using data parallelism, which greatly improves the speed of federated computing and provides technical feasibility for the use of large-scale data privacy computing.
  • FIG. 1 is a schematic flowchart of a federated computing processing method provided by an embodiment of the present disclosure
  • FIG. 2 is a schematic diagram of data segmentation of a hash bucket provided by an embodiment of the present disclosure
  • FIG. 3 is a schematic diagram of sequential bucketing data segmentation provided by an embodiment of the present disclosure
  • FIG. 4 is a schematic flowchart of another federated computing processing method provided by an embodiment of the present disclosure.
  • FIG. 5 is a schematic diagram of the architecture of a federated computing platform provided by an embodiment of the present disclosure
  • FIG. 6 is a schematic flowchart of another federated computing processing method provided by an embodiment of the present disclosure.
  • FIG. 7 is a schematic structural diagram of another federated computing platform provided by an embodiment of the present disclosure.
  • FIG. 8 is a schematic structural diagram of a federated computing processing device provided by an embodiment of the present disclosure.
  • FIG. 9 is a block diagram of an electronic device for implementing the federated computing processing method of the embodiment of the present disclosure.
  • Artificial intelligence is a discipline that studies the use of computers to simulate certain human thinking processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), including both hardware-level technical fields and software-level technologies.
  • Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies include computer vision technology, speech recognition technology, natural language processing technology, deep learning, big data Processing technology, knowledge map technology and other major directions.
  • Cloud computing refers to access to elastic and scalable shared physical or virtual resource pools through the network.
  • Resources include servers, operating systems, networks, software, applications, and storage devices, etc., and resources can be allocated in an on-demand, self-service manner.
  • the federated computing processing method provided by the present disclosure will be described in detail below in combination with various embodiments of the present disclosure.
  • the federated computing processing method provided by the present disclosure is executed by a federated computing platform.
  • FIG. 1 is a schematic flowchart of a federated computing processing method provided by an embodiment of the present disclosure.
  • the processing methods of the federated computing include:
  • Step 101 based on the identifier of the task to be executed, metadata to be processed is obtained from each participant.
  • the task identifier to be executed may be any identifier that can uniquely identify a federated computing task.
  • the federated computing platform may first determine the identifier of the task to be executed, and then obtain metadata related to the identifier of the task to be executed from each participant.
  • the task to be performed by the federated computing platform is a speech recognition model
  • the participants include A and B, where the metadata related to the speech recognition model in participant A is a, and the metadata related to the speech recognition model in participant B is for b.
  • the federated computing platform can obtain metadata a from participant A and metadata b from participant B.
  • the metadata obtained by the federated computing platform from each participant is intermediary data related to the identification of tasks to be executed, that is, the metadata can describe the characteristics of each participant's own data and be used for model training, etc. Federated computing tasks, but not the own data itself.
  • Step 102 segment each metadata to obtain n data slices corresponding to each participant.
  • the data splitting mode may be determined first according to the identifier of the task to be executed.
  • the data segmentation mode may include hash bucket segmentation and sequential bucket segmentation.
  • each metadata can be segmented based on the data segmentation mode to obtain n data slices corresponding to each participant, where n is a positive integer greater than 1 .
  • hash bucket segmentation is that the speed of finding and inserting data is very fast, and the consumption of space is reduced. Therefore, when performing security encryption, unique identification, data verification, hash functions, load balancing (such as consistent hashing) or distributed caching, hash bucket segmentation can be used.
  • load balancing such as consistent hashing
  • hash bucket segmentation can be used.
  • PSI Primary Data Intersection
  • sequential bucket segmentation is that data segmentation is more efficient and stable. Therefore, when the data sequence is relatively uniform, sequential bucket segmentation can be used.
  • FIG. 2 is a schematic diagram of data segmentation in a hash bucket provided by an embodiment of the present disclosure.
  • the metadata to be processed can be ⁇ D1, D2, D3, D4, D5, D6 ⁇
  • the metadata to be processed is segmented into hash buckets, and the resulting data segmentation results can be ⁇ [D1, D3, D4], [D2, D5, D6] ⁇ . That is, the metadata to be processed is divided into two smaller data slices, and because hash bucket segmentation is hashed according to the unique identification value of each data, it is put into the corresponding bucket, Therefore, the obtained data segmentation result is not divided according to the order of the metadata to be processed.
  • FIG. 3 is a schematic diagram of sequential bucketing data segmentation provided by an embodiment of the present disclosure.
  • the metadata to be processed can be ⁇ D1, D2, D3, D4, D5, D6 ⁇
  • the metadata to be processed is segmented into buckets sequentially, and the result of the data segmentation can be ⁇ [D1, D2, D3], [D4, D5, D6] ⁇ .
  • sequential bucket segmentation refers to putting each data into each bucket evenly in order.
  • each participant determines the metadata to be processed based on the task identifier to be executed. After determining the metadata to be processed, each participant separately divides each metadata to obtain the The corresponding n data slices. After that, each participant can synchronize the n data slices after segmentation to the federated computing platform.
  • Step 103 generating n data sets according to n data slices corresponding to each participant.
  • the executive subject of the present disclosure may combine n data slices of each participant, and then respectively combine n data slices corresponding to other participants except itself, to generate n data sets.
  • each data set includes one data slice for each participant.
  • Step 104 call n data processing services corresponding to the task identifier, so as to use each data processing service to process each data set.
  • n threads can be called to run n data processing services to process each data set separately, thereby greatly improving Increased the speed of federated computing.
  • Step 105 according to the processing result of each data processing service, determine the federated calculation result corresponding to the task identifier.
  • each data set includes a data slice of each participant, that is, the processing results obtained according to each data set can be considered as using a small amount of federated data to perform federated calculations to obtain a result, and then multiple processing results After fusion, you can get the calculation results of the federated calculation of the full amount of federated data.
  • processing results of multiple data processing services can be averaged to obtain federated calculation results, or the processing results of multiple data processing services can be weighted and fused to obtain federated calculation results. Do limited.
  • the federated calculation result may also be sent to each of the participants.
  • the metadata to be processed is obtained from each participant based on the task identifier to be executed, and each metadata is segmented to obtain n data slices corresponding to each participant, and then According to the n data slices corresponding to each participant, n data sets are generated, and then n data processing services corresponding to the task identifier are called to process each data set with each data processing service, and finally according to each The processing result of the data processing service determines the federation calculation result corresponding to the task ID. Therefore, when performing federated computing, the parallelization of federated learning is realized by using data parallelism, which greatly improves the speed of federated computing and provides technical feasibility for the use of large-scale data privacy computing.
  • the federated calculation result corresponding to the task identifier can be determined according to the processing result of each data processing service.
  • the process of federated computing may require multiple rounds of iterative updates. Therefore, in this disclosure, after obtaining the federated computing results each time, it is possible to judge the corresponding task ID according to the convergence condition corresponding to the task ID. Whether the task ends to determine whether to continue the federated computation. The above situation will be described below with reference to FIG. 4 .
  • FIG. 4 is a schematic flowchart of another federated computing processing method provided by an embodiment of the present disclosure.
  • the processing methods of the federated computing include:
  • Step 401 based on the identifier of the task to be executed, metadata to be processed is obtained from each participant.
  • Step 402 Segment each piece of metadata to obtain n data slices corresponding to each participant, where n is a positive integer greater than 1.
  • Step 403 generating n data sets according to n data slices corresponding to each participant.
  • Step 404 call n data processing services corresponding to the task identifier, so as to use each data processing service to process each data set.
  • Step 405 acquiring the processing result returned by each data processing service.
  • Step 406 fusing the n processing results to obtain a fusion result.
  • FIG. 5 is a schematic diagram of an architecture of a federated computing platform provided by an embodiment of the present disclosure.
  • the metadata to be processed corresponding to participant A is X.
  • the metadata to be processed corresponding to participant B is Y.
  • the execution subject may call a data split (split) program to split the data of X and Y.
  • the split program is a program capable of cutting a large file into many small files.
  • the split program divides X and Y into smaller data slices. For example, after dividing X and Y respectively, n data slices corresponding to participant A and participant B are obtained, as shown in Figure 5.
  • n data processing services corresponding to the task identifiers can be invoked, so as to process each of the data sets by using each of the data processing services.
  • each data processing service can send the obtained processing results to the data fusion service, such as PSA and PSB in Figure 5, and the PSA and PSB will fuse the processing results of each data processing service to generate the corresponding task ID. Federated calculation results.
  • a corresponding data aggregator may be called to fuse the n processing results, thereby obtaining the fusion result.
  • the federated computing platform in this disclosure can set certain task end conditions for each federated computing task, so as to minimize invalidity while ensuring that the federated computing results are sufficiently accurate. number of calculations.
  • the task end condition may be set as required, such as the number of calculations, the accuracy of calculation results, etc., which is not limited in the present disclosure.
  • the current task end condition after determining the fusion result after each federation calculation, it can be judged whether the current task end condition has been met. For example, it can be judged whether the end condition is met according to the number of federation calculations that have been performed. Alternatively, it may also be determined whether the mission end condition is met or not according to the difference between the current fusion result and the fusion result obtained by the last federation calculation, which is not limited in the present disclosure.
  • Step 407 when it is determined that the task corresponding to the task identifier has not ended, distribute the fusion result to n data processing services, so that each data processing service continues to process the corresponding data set based on the fusion result until the task identifier is determined.
  • the corresponding task has ended, and the federated calculation result corresponding to the task identifier is determined according to the latest n processing results.
  • the fusion result may be distributed to n data processing services. After receiving the fusion result, each data processing service may continue to process the corresponding data set based on the fusion result until it is determined that the task corresponding to the task identifier has ended. Afterwards, the federated calculation result corresponding to the task identifier can be determined according to the latest n processing results.
  • Step 408 in a case where it is determined that the task corresponding to the task identifier has ended, determine that the fusion result is the federation calculation result corresponding to the task identifier.
  • the fusion result obtained after the task corresponding to the task identifier ends may be determined as the federation calculation result corresponding to the task identifier.
  • the task corresponding to the task identifier may be training a face recognition model, and the training times are K times. Then, the federated computing platform can judge whether the number of federated computings that have been executed has reached K after each fusion result is obtained. If not, it is determined that the task is not completed.
  • the processing result returned by each data processing service can be obtained, and the n processing results can be fused to obtain the fusion result. If the task corresponding to the task identifier is not completed, the fusion result is distributed to n data processing services, so that each data processing service continues to process the corresponding data set based on the fusion result until it is determined that the task corresponding to the task identifier has ended , according to the latest n processing results, determine the federated calculation result corresponding to the task ID, and determine that the fusion result is the federated calculation result corresponding to the task ID when it is determined that the task corresponding to the task ID has ended. Therefore, by setting the end of the federated computing task, while ensuring the accuracy of the final federated computing result, it is possible to avoid invalid computing of the task, and further improve the speed of federated computing.
  • FIG. 6 is a schematic flowchart of another federated computing processing method provided by an embodiment of the present disclosure.
  • the processing method of the federated computing includes:
  • Step 601 based on the identifier of the task to be executed, metadata to be processed is obtained from each participant.
  • Step 602 segment each piece of metadata to obtain n data slices corresponding to each participant, where n is a positive integer greater than 1.
  • Step 603 generating n data sets according to n data slices corresponding to each participant.
  • Step 604 call n data processing services corresponding to the task identifier, so as to use each data processing service to process each data set.
  • Step 605 acquiring a fusion result returned by any data processing service, wherein the fusion result is generated by any data processing service after fusing the processing results of each data processing service.
  • FIG. 7 is a schematic structural diagram of another federated computing platform provided by an embodiment of the present disclosure.
  • the metadata to be processed corresponding to participant A is data A
  • the metadata to be processed corresponding to participant B is data b.
  • the execution subject can call the data split program to split data A and data B.
  • the split program is a program capable of cutting a large file into many small files.
  • the split program divides data A and data B into smaller data slices, so that n data slices corresponding to participant A and participant B can be obtained, as shown in Figure 7 WorkerA-0, WorkerA-1, WorkerA-..., WorkerA-n, WorkerB-0, WorkerB-1, WorkerB-..., WorkerB-n. Then combine the first data slice in participant A with the first data slice in participant B to obtain a data set, and follow the above method in turn until the data in participant A and participant B are combined All slices are combined to generate n data sets (WorkerA-0, WorkerB-0), (WorkerA-1, WorkerB-1), (WorkerA-..., WorkerB-...), (WorkerA-n, WorkerB-n).
  • n data processing services corresponding to the task identifiers can be invoked, so as to process each of the data sets by using each of the data processing services.
  • n data processing services can be invoked, so as to process each of the data sets by using each of the data processing services.
  • each data processing service After each data processing service obtains a data processing result, it can synchronize the processing result to other data processing services, so that each data processing service can be based on Multiple processing results, update its own processing results, and obtain the fused processing results.
  • the federated calculation results corresponding to the task identifiers are determined.
  • Step 606 in a case where it is determined that the task corresponding to the task identifier has ended, determine that the fusion result is the federation calculation result corresponding to the task identifier.
  • Step 607 when it is determined that the task corresponding to the task identifier is not completed, send the fusion result to any data processing service, so that each data processing service continues to process the corresponding data set based on the fusion result until the task identifier is determined.
  • the corresponding task has ended, and the latest fusion result is used to determine the federation calculation result corresponding to the task identifier.
  • the fusion result may be sent to any data processing service. After any data processing service receives the fusion result, it may continue to process the corresponding data set based on the fusion result until it is determined that the task corresponding to the task identifier has ended. After the task corresponding to the task identifier is completed, the execution subject of the present disclosure can determine the federated calculation result corresponding to the task identifier with the latest fusion result.
  • Step 608 send the federated calculation result to each participant.
  • the fusion result returned by any data processing service can be obtained, wherein the fusion result is that any data processing service combines each data processing service It is generated after fusion of the processing results of the task identifier. If it is determined that the task corresponding to the task identifier has ended, the fusion result is determined to be the federated calculation result corresponding to the task identifier.
  • the fusion result Send it to any data processing service, so that each data processing service continues to process the corresponding data set based on the fusion result until it is determined that the task corresponding to the task identifier has ended, and the latest fusion result is used to determine the federated computing corresponding to the task identifier result. Therefore, by setting the end of the federated computing task, while ensuring the accuracy of the final federated computing result, it is possible to avoid invalid computing of the task, and further improve the speed of federated computing.
  • FIG. 8 is a schematic structural diagram of a federated computing processing device provided by an embodiment of the present disclosure.
  • the processing device of the federated computing includes: an acquisition module 810 , a segmentation module 820 , a generation module 830 , a processing module 840 and a determination module 850 .
  • An acquisition module configured to acquire metadata to be processed from each participant based on the task identifier to be executed
  • a segmentation module configured to segment each of the metadata to obtain n data slices corresponding to each participant, where n is a positive integer greater than 1;
  • a generating module configured to generate n data sets according to n data slices corresponding to each participant
  • a processing module configured to call n data processing services corresponding to the task identifier, so as to process each of the data sets by using each of the data processing services;
  • the determination module is configured to determine the federated calculation result corresponding to the task identifier according to the processing result of each data processing service.
  • the segmentation module 820 is specifically configured to:
  • each of the metadata is segmented to obtain n data slices corresponding to each participant.
  • the determining module 850 is specifically configured to:
  • the fusion result is determined to be a federated calculation result corresponding to the task identifier.
  • the determining module 850 is further configured to:
  • the fusion result is distributed to n data processing services, so that each data processing service continues to process the corresponding data set based on the fusion result Perform processing until it is determined that the task corresponding to the task identifier has ended, and determine the federated calculation result corresponding to the task identifier according to the latest n processing results.
  • the determining module 850 is specifically configured to:
  • the fusion result is sent to any one of the data processing services, so that each of the data processing services continues to process the corresponding data set based on the fusion result Processing is performed until it is determined that the task corresponding to the task identifier has ended, and the latest fusion result is used to determine the federated calculation result corresponding to the task identifier.
  • the determination module 850 is further configured to:
  • the federated computing processing device of the embodiment of the present disclosure obtains the metadata to be processed from each participant based on the task identifier to be executed, and divides each metadata to obtain the n corresponding to each participant n data slices, and then generate n data sets according to the n data slices corresponding to each participant, and then call n data processing services corresponding to the task ID to use each data processing service to process each data set , and finally, according to the processing result of each data processing service, determine the federated calculation result corresponding to the task ID.
  • the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.
  • FIG. 9 shows a schematic block diagram of an example electronic device 900 that may be used to implement embodiments of the present disclosure.
  • Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers.
  • Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices.
  • the components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the disclosure described and/or claimed herein.
  • the device 900 includes a computing unit 901, which can be loaded into a RAM (Random Access Memory, random access/ accesses the computer program in the memory) 903 to execute various appropriate actions and processes.
  • RAM Random Access Memory
  • various programs and data necessary for the operation of the device 900 can also be stored.
  • the computing unit 901, ROM 902, and RAM 903 are connected to each other through a bus 904.
  • An I/O (Input/Output, input/output) interface 905 is also connected to the bus 904 .
  • the I/O interface 905 includes: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a magnetic disk, an optical disk, etc. ; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, and the like.
  • the communication unit 909 allows the device 900 to exchange information/data with other devices over a computer network such as the Internet and/or various telecommunication networks.
  • the computing unit 901 may be various general-purpose and/or special-purpose processing components having processing and computing capabilities. Some examples of computing unit 901 include but are not limited to CPU (Central Processing Unit, central processing unit), GPU (Graphic Processing Units, graphics processing unit), various dedicated AI (Artificial Intelligence, artificial intelligence) computing chips, various operating The computing unit of the machine learning model algorithm, DSP (Digital Signal Processor, digital signal processor), and any appropriate processor, controller, microcontroller, etc.
  • the computing unit 901 executes various methods and processes described above, for example, a processing method of federated computing.
  • the processing method of federated computing may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 908 .
  • part or all of the computer program may be loaded and/or installed on the device 900 via the ROM 902 and/or the communication unit 909.
  • the computer program When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the processing method of the federated computing described above can be executed.
  • the computing unit 901 may be configured in any other appropriate way (for example, by means of firmware) to execute the processing method of federated computing.
  • programmable processor can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.
  • Program codes for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special purpose computer, or other programmable data processing devices, so that the program codes, when executed by the processor or controller, make the functions/functions specified in the flow diagrams and/or block diagrams Action is implemented.
  • the program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.
  • a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device.
  • a machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium.
  • a machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing.
  • machine-readable storage media would include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory, Erasable Programmable Read-Only Memory) Or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory, portable compact disk read-only memory), optical storage device, magnetic storage device, or any suitable combination of the above.
  • the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (Cathode-Ray Tube) or LCD ( Liquid Crystal Display (LCD) monitor); and a keyboard and pointing device (such as a mouse or trackball) through which a user can provide input to a computer.
  • a display device e.g., a CRT (Cathode-Ray Tube) or LCD ( Liquid Crystal Display (LCD) monitor
  • a keyboard and pointing device such as a mouse or trackball
  • Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.
  • the systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system.
  • the components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: LAN (Local Area Network, local area network), WAN (Wide Area Network, wide area network), the Internet, and blockchain networks.
  • a computer system may include clients and servers.
  • Clients and servers are generally remote from each other and typically interact through a communication network.
  • the relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.
  • the server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the problems existing in traditional physical hosts and VPS services (Virtual Private Server, virtual private server). The defects of difficult management and weak business expansion.
  • the server can also be a server of a distributed system, or a server combined with a blockchain.
  • the present disclosure also provides a computer program product, when the instruction processor in the computer program product executes, executes the federated computing processing method proposed by the above-mentioned embodiments of the present disclosure.
  • steps may be reordered, added or deleted using the various forms of flow shown above.
  • each step described in the present disclosure may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present disclosure can be achieved, no limitation is imposed herein.

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Computer Security & Cryptography (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本公开提供了联邦计算的处理方法、装置、电子设备和存储介质，涉及人工智能技术领域，尤其涉及大数据、云计算、深度学习技术领域。具体实现方案为：基于待执行的任务标识，从每个参与方获取待处理的元数据；将每个元数据进行切分，以获取每个参与方对应的n个数据切片；根据每个参与方对应的n个数据切片，生成n个数据集；调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每个数据集进行处理；根据每个数据处理服务的处理结果，确定任务标识对应的联邦计算结果。由此，在进行联邦计算时，通过利用数据并行的方式实现联邦学习的并行化，极大的提升了联邦计算的速度，为大规模数据隐私计算的使用提供了技术可行性。

Description

联邦计算的处理方法、装置、电子设备和存储介质
相关申请的交叉引用
本公开要求百度在线网络技术(北京)有限公司于2021年06月18日提交的、发明名称为“联邦计算的处理方法、装置、电子设备和存储介质”的、中国专利申请号“202110677839.9”的优先权。
技术领域
本公开涉及人工智能技术领域，尤其涉及大数据、云计算、深度学习技术领域，具体涉及一种联邦计算的处理方法、装置、电子设备和存储介质。
背景技术
联邦计算，是指使用多方安全计算(Multi-Party Computation，MPC)、可信执行环境(Trusted Execution Environment，TEE)、联邦学习(Federated Learning，FL)等隐私安全技术，为多方数据安全合作提供数据可用不见的数据分析环境。对数据在流通过程的隐私安全产生着重要的影响。
如何提升联邦计算的性能，是亟待解决的问题。
发明内容
本公开提出一种联邦计算的处理方法、装置、电子设备和存储介质。
根据本公开的一方面，提供了一种联邦计算的处理方法，包括：
基于待执行的任务标识，从每个参与方获取待处理的元数据；
将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数；
根据每个参与方对应的n个数据切片，生成n个数据集；
调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理；
根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果。
根据本公开的另一方面，提供了一种联邦计算的处理装置，包括：
获取模块，用于基于待执行的任务标识，从每个参与方获取待处理的元数据；
切分模块，用于将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数；
生成模块，用于根据每个参与方对应的n个数据切片，生成n个数据集；
处理模块，用于调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理；
确定模块，用于根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果。
根据本公开的另一方面，提供了一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述一方面实施例所述的联邦计算的处理方 法。
根据本公开的另一方面，提供了一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使所述计算机执行根据上述一方面实施例所述的联邦计算的处理方法。
根据本公开的另一方面，提供了一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现根据上述一方面实施例所述的联邦计算的处理方法。
本公开实施例中，通过基于待执行的任务标识，从每个参与方获取待处理的元数据，并将每个元数据进行切分，以获取每个参与方对应的n个数据切片，之后根据每个参与方对应的n个数据切片，生成n个数据集，再调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每个数据集进行处理，最后根据每个数据处理服务的处理结果，确定任务标识对应的联邦计算结果。由此，在进行联邦计算时，通过利用数据并行的方式实现联邦学习的并行化，极大的提升了联邦计算的速度，为大规模数据隐私计算的使用提供了技术可行性。
应当理解，本部分所描述的内容并非旨在标识本公开的实施例的关键或重要特征，也不用于限制本公开的范围。本公开的其它特征将通过以下的说明书而变得容易理解。
附图说明
附图用于更好地理解本方案，不构成对本公开的限定。其中：
图1为本公开实施例所提供的一种联邦计算的处理方法的流程示意图；
图2为本公开实施例所提供的一种哈希分桶的数据切分的示意图；
图3为本公开实施例所提供的一种顺序分桶的数据切分的示意图；
图4为本公开实施例所提供的另一种联邦计算的处理方法的流程示意图；
图5为本公开实施例提供的一种联邦计算平台的架构示意图；
图6为本公开实施例提供的另一种联邦计算的处理方法的流程示意图；
图7为本公开实施例提供的另一种联邦计算平台的架构示意图；
图8为本公开实施例提供的一种联邦计算的处理装置的结构示意图；
图9是用来实现本公开实施例的联邦计算的处理方法的电子设备的框图。
具体实施方式
以下结合附图对本公开的示范性实施例做出说明，其中包括本公开实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本公开的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。
下面参考附图描述本公开实施例的联邦计算的处理方法、装置、电子设备和存储介质。
人工智能是研究使用计算机来模拟人的某些思维过程和智能行为(如学习、推理、思考、规划等)的学科，既有硬件层面的技术领域也有软件层面的技术。人工智能硬件技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理等技术；人工智能软件技术包括计算机视觉技术、语音识别技术、自然语言处理技术以及深度学习、大数据处理技术、知识图谱技术等几大方向。
云计算，指的是通过网络接入弹性可扩展的共享物理或虚拟资源池，资源包括服务器、操作系统、网络、软件、应用和存储设备等，并可以以按需、自服务的方式对资源进行部署和管理的技术体系。通过云计算技术，可以为人工智能、区块链等技术应用、模型训练提供高效强大的数据处理能力。
下面结合本公开各实施例对本公开提供的联邦计算的处理方法进行详细说明。本公开提供的联邦计算的处理方法的执行主体为联邦计算平台。
图1为本公开实施例所提供的一种联邦计算的处理方法的流程示意图。
如图1所示，该联邦计算的处理方法包括：
步骤101，基于待执行的任务标识，从每个参与方获取待处理的元数据。
其中，待执行的任务标识，可以为任一可以唯一标识某一联邦计算任务的标识。
本公开中联邦计算平台，可以首先确定待执行的任务标识，进而从每个参与方获取与待执行的任务标识相关的元数据。
例如，联邦计算平台要执行的任务为语音识别模型，且参与方包括A和B，其中，参与方A中与语音识别模型相关的元数据为a，参与方B与语音识别模型相关的元数据为b。从而联邦计算平台，即可从参与方A中获取元数据a，并从参与方B中获取元数据b。
可以理解的是，联邦计算平台从各个参与方获取的元数据，为与待执行的任务标识相关的中介数据，即该元数据可以描述各个参与方自有数据的特性，并用于进行模型训练等联邦计算任务，但是并非自有数据本身。
步骤102，将每个元数据进行切分，以获取每个参与方对应的n个数据切片。
可选的，可以先根据所述待执行的任务标识，确定数据切分模式。其中，数据切分模式可以包括哈希分桶切分和顺序分桶切分。
在确定数据切分模式之后，可以基于所述数据切分模式，将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数。
需要说明的是，可以根据需要，为不同类型的任务标识，配置不同的切分模式。例如，哈希分桶切分的优势在于查找和插入数据的速度非常快，并减少了空间的消耗。因此，在进行安全加密、唯一标识、数据校验、散列函数、负载均衡(比如一致性哈希)或分布式缓存时，可以使用哈希分桶切分。特别的，在两个参与方数据量大小不一样的场景中，如PSI(隐私数据求交集)算法就必须要采用哈希分桶切分。顺序分桶切分的优势在于数据切分的效率较高，并且比较稳定。因此，在数据序列比较均匀的情况下，可以使用顺序分桶切分。
下面结合图2，对哈希分桶的数据切分进行详细说明。图2为本公开实施例提供的一种哈希分桶的数据切分的示意图。
如图2所示，待处理的元数据可以是{D1，D2，D3，D4，D5，D6}，对待处理的元数据进行哈希分桶的数据切分，得到的数据切分结果可以是{[D1，D3，D4]，[D2，D5，D6]}。即，将待处理的元数据切分为两个较小的数据切片，并且由于哈希分桶切分是根据每个数据的唯一的标识值求哈希后，放入相对应的桶中，所以得到的数据切分结果不是按照待处理的元数据的顺序进行切分。
下面结合图3，对顺序分桶的数据切分进行详细说明。图3为本公开实施例提供的一种顺序分桶的数据切分的示意图。
如图3所示，待处理的元数据可以是{D1，D2，D3，D4，D5，D6}，对待处理的元数据进行顺序分桶的数据切分，得到的数据切分结果可以是{[D1，D2，D3]，[D4，D5，D6]}。由此可见，顺序分桶切分是指将每个数据按照顺序均匀地放入每个桶中。
需要说明的是，上述将每个参与方对应数据，分为n个数据切片的处理过程，也可以有各个参与方单独执行。即每个参与方基于待执行的任务标识，确定待处理的元数据，在确定待处理的元数据之后，每个参与方分别将各自的每个元数据进行切分，以获取每个参与方对应的n个数据切片。之后各个参与方再将切分后的n个数据切片分别同步给联邦计算平台即可。
步骤103，根据每个参与方对应的n个数据切片，生成n个数据集。
本公开中，本公开的执行主体可以将每个参与方的n个数据切片，之后与除了自身以外的其他参与方对应的n个数据切片分别进行组合，以生成n个数据集。
可以理解的是，每个数据集中包括每个参与方的一个数据切片。
步骤104，调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每 个数据集进行处理。
本公开中，在生成了与待执行的任务标识对应的n个数据集后，即可调用n个线程，以运行n个数据处理服务，分别对每个数据集进行处理，从而极大的提升了联邦计算的速度。
步骤105，根据每个数据处理服务的处理结果，确定任务标识对应的联邦计算结果。
具体的，由于每个数据集中都包括各个参与的一个数据切片，即根据每个数据集得到的处理结果，可以认为是利用少量的联邦数据进行联邦计算得到一个结果，之后，将多个处理结果进行融合，即可得到全量的联邦数据进行联邦计算后的计算结果。
可选的，可以将多个数据处理服务的处理结果进行平均，以得到联邦计算结果，或者也可以将多个数据处理服务的处理结果进行加权融合，以得到联邦计算结果，本公开对此不做限定。
可选的，在确定任务标识对应的联邦计算结果之后，还可以将所述联邦计算结果发送给各个所述参与方。
本公开实施例中，通过基于待执行的任务标识，从每个参与方获取待处理的元数据，并将每个元数据进行切分，以获取每个参与方对应的n个数据切片，之后根据每个参与方对应的n个数据切片，生成n个数据集，再调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每个数据集进行处理，最后根据每个数据处理服务的处理结果，确定任务标识对应的联邦计算结果。由此，在进行联邦计算时，通过利用数据并行的方式实现联邦学习的并行化，极大的提升了联邦计算的速度，为大规模数据隐私计算的使用提供了技术可行性。
通过上述分析可知，本公开中，可以根据每个数据处理服务的处理结果，确定任务标识对应的联邦计算结果。在一种可能的实现形式中，联邦计算的过程可能需要多轮迭代更新，因此本公开中，可以在每次获取到联邦计算结果后，根据该任务标识对应的收敛条件，判断任务标识对应的任务是否结束，以确定是否继续进行联邦计算。下面结合图4对上述情况进行说明，图4为本公开实施例所提供的另一种联邦计算的处理方法的流程示意图。
如图4所示，该联邦计算的处理方法包括：
步骤401，基于待执行的任务标识，从每个参与方获取待处理的元数据。
步骤402，将每个元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数。
步骤403，根据每个参与方对应的n个数据切片，生成n个数据集。
步骤404，调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每个数据集进行处理。
本公开中，步骤401-404的具体实现方式可参照本公开中其他实施例的详细描述，在此不再赘述。
步骤405，获取每个数据处理服务返回的处理结果。
步骤406，将n个处理结果进行融合，以获取融合结果。
本公开中，联邦计算平台，可以采用图5所示的架构。图5为本公开实施例提供的一种联邦计算平台的架构示意图。
图5中，以参与方为参与方A和参与方B为例，如图5所示，参与方A对应的待处理的元数据为X。参与方B对应的待处理的元数据为Y。执行主体可以在基于待执行的任务标识，从参与方A和参与方B获取X和Y之后，调用数据切分(split)程序对X和Y进行数据切分。其中，split程序是能够将大文件切割成很多小文件的程序。这里，split程序将X和Y切分为更小的数据切片，比如，将X和Y分别切分后，得到参与方A和参与方B对应的n个数据切片，如图5中的FLA-0、FLA-1、FLA-…、FLA-n、FLB-0、FLB-1、FLB-…、FLB-n。再将参与方A中的第一个数据切片和参与方B中的第一个数据切片进行组合，得到一个数据集，并按照上述方法依次类推，直到将参与方A与参与方B中的数据 切片全部进行组合，生成n个数据集(FLA-0、FLB-0)、(FLA-1、FLB-1)、(FLA-…、FLB-…)、(FLA-n、FLB-n)。
在得到n个数据集之后，即可调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理。最后每个数据处理服务即可将得到的处理结果，发送给数据融合服务，如图5中的PSA、PSB，由PSA及PSB对各个数据处理服务的处理结果进行融合，以生成任务标识对应的联邦计算结果。
本公开中，在利用每个数据处理服务对每个数据集进行处理，得到n个处理结果后，可以调用相应的数据汇总器对n个处理结果进行融合，从而，得到融合结果。
可以理解的是，为了避免联邦计算任务不停循环，本公开中联邦计算平台可以为每个联邦计算任务都设置一定的任务结束条件，从而在保证联邦计算结果足够准确的情况下，尽量减少无效计算的次数。其中任务结束条件，可以根据需要设置，比如计算次数、计算结果的精度等等，本公开对此不做限定。
本公开中，在确定每次联邦计算后的融合结果后，即可判断当前是否已达到任务结束条件。比如，可以根据已执行的联邦计算次数，判断是否达到结束条件。或者，也可以根据当前的融合结果与上一次联邦计算得到的融合结果间的差异，确定是否达到任务结束条件等等，本公开对此不做限定。
步骤407，在确定任务标识对应的任务未结束的情况下，将融合结果分发给n个数据处理服务，以使每个数据处理服务基于融合结果继续对对应的数据集进行处理，直至确定任务标识对应的任务已结束，根据最新的n个处理结果，确定任务标识对应的联邦计算结果。
本公开中，可以先确定任务标识对应的任务是否结束，在确定任务标识对应的任务未结束的情况下，可以将融合结果分发给n个数据处理服务。每个数据处理服务在接收到融合结果之后，可以基于融合结果继续对对应的数据集进行处理，直至确定任务标识对应的任务已结束。之后，即可根据最新的n个处理结果，确定任务标识对应的联邦计算结果。
步骤408，在确定任务标识对应的任务已结束的情况下，确定融合结果为任务标识对应的联邦计算结果。
本公开中，在确定任务标识对应的任务已结束的情况下，可以将任务标识对应的任务结束后得到的融合结果，确定为任务标识对应的联邦计算结果。
例如，任务标识对应的任务可以是训练人脸识别模型，且训练次数为K次。那么联邦计算平台，即可在每次得到融合结果后，即判断已执行的联邦计算次数，是否达到K，若未达到，则确定任务未结束，若已达到，则确定任务结束。
本公开实施例中，在利用每个数据处理服务对每个数据集进行处理后，可以获取每个数据处理服务返回的处理结果，并将n个处理结果进行融合，以获取融合结果，在确定任务标识对应的任务未结束的情况下，将融合结果分发给n个数据处理服务，以使每个数据处理服务基于融合结果继续对对应的数据集进行处理，直至确定任务标识对应的任务已结束，根据最新的n个处理结果，确定任务标识对应的联邦计算结果，在确定任务标识对应的任务已结束的情况下，确定融合结果为任务标识对应的联邦计算结果。由此，通过为联邦计算任务设置结束，从而在保证最终的联邦计算结果准确的情况下，尽量避免了对任务进行无效计算的情况，进一步提高了联邦计算的速度。
下面结合图6进行说明，图6为本公开实施例所提供的另一种联邦计算的处理方法的流程示意图。
如图6所示，该联邦计算的处理方法包括：
步骤601，基于待执行的任务标识，从每个参与方获取待处理的元数据。
步骤602，将每个元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数。
步骤603，根据每个参与方对应的n个数据切片，生成n个数据集。
步骤604，调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每个数据集进行处理。
本公开中，步骤601-604的具体实现方式可参照本公开中其他实施例的详细描述，在此不再赘述。
步骤605，获取任一数据处理服务返回的融合结果，其中，融合结果为任一数据处理服务将每个数据处理服务的处理结果进行融合后生成的。
本公开中，联邦计算平台可以采用如图7所示的架构。图7为本公开实施例提供的另一种联邦计算平台的架构示意图。
图7中，以参与方为参与方A和参与方B为例，如图7所示，参与方A对应的待处理的元数据是数据A，参与方B对应的待处理的元数据是数据B。执行主体可以在基于待执行的任务标识，从参与方A和参与方B获取数据A和数据B之后，调用数据切分split程序对数据A和数据B进行数据切分。其中，split程序是能够将大文件切割成很多小文件的程序。这里，split程序将数据A和数据B切分为更小的数据切片，从而，可以得到参与方A和参与方B对应的n个数据切片，如图7中的WorkerA-0、WorkerA-1、WorkerA-…、WorkerA-n、WorkerB-0、WorkerB-1、WorkerB-…、WorkerB-n。再将参与方A中的第一个数据切片和参与方B中的第一个数据切片进行组合，得到一个数据集，并按照上述方法依次类推，直到将参与方A与参与方B中的数据切片全部进行组合，生成n个数据集(WorkerA-0、WorkerB-0)、(WorkerA-1、WorkerB-1)、(WorkerA-…、WorkerB-…)、(WorkerA-n、WorkerB-n)。
在得到n个数据集之后，即可调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理。在数据处理服务对每个所述数据集进行处理的过程中，每个数据处理服务在得到一次数据处理结果后，可以将该处理结果同步给其余数据处理服务，从而使得各个数据处理服务可以基于多个处理结果，对自身的处理结果进行更新，得到融合后的处理结果。最后，根据每个所述数据处理服务更新后的处理结果，确定所述任务标识对应的联邦计算结果。
步骤606，在确定任务标识对应的任务已结束的情况下，确定融合结果为任务标识对应的联邦计算结果。
步骤607，在确定任务标识对应的任务未结束的情况下，将融合结果发送给任一数据处理服务，以使每个数据处理服务基于融合结果继续对对应的数据集进行处理，直至确定任务标识对应的任务已结束，将最新的融合结果，确定任务标识对应的联邦计算结果。
本公开中，可以先确定任务标识对应的任务是否结束，在确定任务标识对应的任务未结束的情况下，可以将融合结果发送给任一数据处理服务。任一数据处理服务在接收到融合结果之后，可以基于融合结果继续对对应的数据集进行处理，直至确定任务标识对应的任务已结束。本公开的执行主体可以在任务标识对应的任务结束之后，将最新的融合结果，确定任务标识对应的联邦计算结果。
步骤608，将联邦计算结果发送给各个参与方。
本公开实施例中，在利用每个数据处理服务对每个数据集进行处理后，可以获取任一数据处理服务返回的融合结果，其中，融合结果为任一数据处理服务将每个数据处理服务的处理结果进行融合后生成的，在确定任务标识对应的任务已结束的情况下，确定融合结果为任务标识对应的联邦计算结果，在确定任务标识对应的任务未结束的情况下，将融合结果发送给任一数据处理服务，以使每个数据处理服务基于融合结果继续对对应的数据集进行处理，直至确定任务标识对应的任务已结束，将最新的融合结果，确定任务标识对应的联邦计算结果。由此，通过为联邦计算任务设置结束，从而在保证最终的联邦计算结果准确的情况下，尽量避免了对任务进行无效计算的情况，进一步提高了联邦计算的速度。
为了实现上述实施例，本公开实施例还提出一种联邦计算的处理装置。图8为本公开实施例提供的一种联邦计算的处理装置的结构示意图。
如图8所示，该联邦计算的处理装置包括：获取模块810、切分模块820、生成模块830、处理模块840和确定模块850。
获取模块，用于基于待执行的任务标识，从每个参与方获取待处理的元数据；
切分模块，用于将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数；
生成模块，用于根据每个参与方对应的n个数据切片，生成n个数据集；
处理模块，用于调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理；
确定模块，用于根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果。
在本公开实施例一种可能的实现方式中，所述切分模块820，具体用于：
根据所述待执行的任务标识，确定数据切分模式；
基于所述数据切分模式，将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片。
在本公开实施例一种可能的实现方式中，所述确定模块850，具体用于：
获取每个所述数据处理服务返回的处理结果；
将n个所述处理结果进行融合，以获取融合结果；
在确定所述任务标识对应的任务已结束的情况下，确定所述融合结果为所述任务标识对应的联邦计算结果。
在本公开实施例一种可能的实现方式中，所述确定模块850，还用于：
在确定所述任务标识对应的任务未结束的情况下，将所述融合结果分发给n个所述数据处理服务，以使每个所述数据处理服务基于所述融合结果继续对对应的数据集进行处理，直至确定所述任务标识对应的任务已结束，根据最新的n个处理结果，确定所述任务标识对应的联邦计算结果。
在本公开实施例一种可能的实现方式中，所述确定模块850，具体用于：
获取任一数据处理服务返回的融合结果，其中，所述融合结果为所述任一数据处理服务将每个所述数据处理服务的处理结果进行融合后生成的；
在确定所述任务标识对应的任务已结束的情况下，确定所述融合结果为所述任务标识对应的联邦计算结果；
在确定所述任务标识对应的任务未结束的情况下，将所述融合结果发送给所述任一数据处理服务，以使每个所述数据处理服务基于所述融合结果继续对对应的数据集进行处理，直至确定所述任务标识对应的任务已结束，将最新的融合结果，确定所述任务标识对应的联邦计算结果。
在本公开实施例一种可能的实现方式中，所述确定模块850，还用于：
将所述联邦计算结果发送给各个所述参与方。
需要说明的是，前述联邦计算的处理方法实施例的解释说明，也适用于该实施例的联邦计算的处理装置，故在此不再赘述。
本公开实施例的联邦计算的处理装置，通过基于待执行的任务标识，从每个参与方获取待处理的元数据，并将每个元数据进行切分，以获取每个参与方对应的n个数据切片，之后根据每个参与方对应的n个数据切片，生成n个数据集，再调用与任务标识对应的n个数据处理服务，以利用每个数据处理服务对每个数据集进行处理，最后根据每个数据处理服务的处理结果，确定任务标识对应的联邦计算结果。由此，在进行联邦计算的处理时，利用数据并行的方式，首先，将每个参与方对应的n个数据切片进行组合生成数据集，进而再利用每个数据处理服务对每个数据集进行处理，得到联邦计算结果，从而实现联邦学习的并行化，提升了计算速度，为大规模数据隐私计算的使用提供了技术可行性。
根据本公开的实施例，本公开还提供了一种电子设备、一种可读存储介质和一种计算 机程序产品。
图9示出了可以用来实施本公开的实施例的示例电子设备900的示意性框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本公开的实现。
如图9所示，设备900包括计算单元901，其可以根据存储在ROM(Read-Only Memory，只读存储器)902中的计算机程序或者从存储单元908加载到RAM(Random Access Memory，随机访问/存取存储器)903中的计算机程序，来执行各种适当的动作和处理。在RAM 903中，还可存储设备900操作所需的各种程序和数据。计算单元901、ROM 902以及RAM 903通过总线904彼此相连。I/O(Input/Output，输入/输出)接口905也连接至总线904。
设备900中的多个部件连接至I/O接口905，包括：输入单元906，例如键盘、鼠标等；输出单元907，例如各种类型的显示器、扬声器等；存储单元908，例如磁盘、光盘等；以及通信单元909，例如网卡、调制解调器、无线通信收发机等。通信单元909允许设备900通过诸如因特网的计算机网络和/或各种电信网络与其他设备交换信息/数据。
计算单元901可以是各种具有处理和计算能力的通用和/或专用处理组件。计算单元901的一些示例包括但不限于CPU(Central Processing Unit，中央处理单元)、GPU(Graphic Processing Units，图形处理单元)、各种专用的AI(Artificial Intelligence，人工智能)计算芯片、各种运行机器学习模型算法的计算单元、DSP(Digital Signal Processor，数字信号处理器)、以及任何适当的处理器、控制器、微控制器等。计算单元901执行上文所描述的各个方法和处理，例如联邦计算的处理方法。例如，在一些实施例中，联邦计算的处理方法可被实现为计算机软件程序，其被有形地包含于机器可读介质，例如存储单元908。在一些实施例中，计算机程序的部分或者全部可以经由ROM 902和/或通信单元909而被载入和/或安装到设备900上。当计算机程序加载到RAM 903并由计算单元901执行时，可以执行上文描述的联邦计算的处理方法的一个或多个步骤。备选地，在其他实施例中，计算单元901可以通过其他任何适当的方式(例如，借助于固件)而被配置为执行联邦计算的处理方法。
本文中以上描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、FPGA(Field Programmable Gate Array，现场可编程门阵列)、ASIC(Application-Specific Integrated Circuit，专用集成电路)、ASSP(Application Specific Standard Product，专用标准产品)、SOC(System On Chip，芯片上系统的系统)、CPLD(Complex Programmable Logic Device，复杂可编程逻辑设备)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。
用于实施本公开的方法的程序代码可以采用一个或多个编程语言的任何组合来编写。这些程序代码可以提供给通用计算机、专用计算机或其他可编程数据处理装置的处理器或控制器，使得程序代码当由处理器或控制器执行时使流程图和/或框图中所规定的功能/操作被实施。程序代码可以完全在机器上执行、部分地在机器上执行，作为独立软件包部分地在机器上执行且部分地在远程机器上执行或完全在远程机器或服务器上执行。
在本公开的上下文中，机器可读介质可以是有形的介质，其可以包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的程序。机器可读 介质可以是机器可读信号介质或机器可读储存介质。机器可读介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备，或者上述内容的任何合适组合。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、RAM、ROM、EPROM(Electrically Programmable Read-Only-Memory，可擦除可编程只读存储器)或快闪存储器、光纤、CD-ROM(Compact Disc Read-Only Memory，便捷式紧凑盘只读存储器)、光学储存设备、磁储存设备、或上述内容的任何合适组合。
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(Cathode-Ray Tube，阴极射线管)或者LCD(Liquid Crystal Display，液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：LAN(Local Area Network，局域网)、WAN(Wide Area Network，广域网)、互联网和区块链网络。
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。服务器可以是云服务器，又称为云计算服务器或云主机，是云计算服务体系中的一项主机产品，以解决了传统物理主机与VPS服务(Virtual Private Server，虚拟专用服务器)中，存在的管理难度大，业务扩展性弱的缺陷。服务器也可以为分布式系统的服务器，或者是结合了区块链的服务器。
根据本公开的实施例，本公开还提供了一种计算机程序产品，当计算机程序产品中的指令处理器执行时，执行本公开上述实施例提出的联邦计算的处理方法。
根据本公开实施例的技术方案，涉及大数据、云计算、深度学习技术领域，在进行联邦计算的处理时，利用数据并行的方式，首先，将每个参与方对应的n个数据切片进行组合生成数据集，进而再利用每个数据处理服务对每个数据集进行处理，得到联邦计算结果，由此，在进行联邦计算时，通过利用数据并行的方式实现联邦学习的并行化，极大的提升了联邦计算的速度，为大规模数据隐私计算的使用提供了技术可行性。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本发公开中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本公开公开的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本公开保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本公开的精神和原则之内所作的修改、等同替换和改进等，均应包含在本公开保护范围之内。

Claims (15)

  1. 一种联邦计算的处理方法，包括：
    基于待执行的任务标识，从每个参与方获取待处理的元数据；
    将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数；
    根据每个参与方对应的n个数据切片，生成n个数据集；
    调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理；
    根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果。
  2. 如权利要求1所述的方法，其中，所述将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，包括：
    根据所述待执行的任务标识，确定数据切分模式；
    基于所述数据切分模式，将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片。
  3. 如权利要求1所述的方法，其中，所述根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果，包括：
    获取每个所述数据处理服务返回的处理结果；
    将n个所述处理结果进行融合，以获取融合结果；
    在确定所述任务标识对应的任务已结束的情况下，确定所述融合结果为所述任务标识对应的联邦计算结果。
  4. 如权利要求3所述的方法，还包括：
    在确定所述任务标识对应的任务未结束的情况下，将所述融合结果分发给n个所述数据处理服务，以使每个所述数据处理服务基于所述融合结果继续对对应的数据集进行处理，直至确定所述任务标识对应的任务已结束，根据最新的n个处理结果，确定所述任务标识对应的联邦计算结果。
  5. 如权利要求1所述的方法，其中，所述根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果，包括：
    获取任一数据处理服务返回的融合结果，其中，所述融合结果为所述任一数据处理服务将每个所述数据处理服务的处理结果进行融合后生成的；
    在确定所述任务标识对应的任务已结束的情况下，确定所述融合结果为所述任务标识对应的联邦计算结果；
    在确定所述任务标识对应的任务未结束的情况下，将所述融合结果发送给所述任一数据处理服务，以使每个所述数据处理服务基于所述融合结果继续对对应的数据集进行处理，直至确定所述任务标识对应的任务已结束，将最新的融合结果，确定所述任务标识对应的联邦计算结果。
  6. 如权利要求1-5任一所述的方法，还包括：
    将所述联邦计算结果发送给各个所述参与方。
  7. 一种联邦计算的处理装置，包括：
    获取模块，用于基于待执行的任务标识，从每个参与方获取待处理的元数据；
    切分模块，用于将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片，其中，n为大于1的正整数；
    生成模块，用于根据每个参与方对应的n个数据切片，生成n个数据集；
    处理模块，用于调用与所述任务标识对应的n个数据处理服务，以利用每个所述数据处理服务对每个所述数据集进行处理；
    确定模块，用于根据每个所述数据处理服务的处理结果，确定所述任务标识对应的联邦计算结果。
  8. 如权利要求7所述的装置，其中，所述切分模块，具体用于：
    根据所述待执行的任务标识，确定数据切分模式；
    基于所述数据切分模式，将每个所述元数据进行切分，以获取每个参与方对应的n个数据切片。
  9. 如权利要求7所述的装置，其中，所述确定模块，具体用于：
    获取每个所述数据处理服务返回的处理结果；
    将n个所述处理结果进行融合，以获取融合结果；
    在确定所述任务标识对应的任务已结束的情况下，确定所述融合结果为所述任务标识对应的联邦计算结果。
  10. 如权利要求9所述的装置，其中，所述确定模块，还用于：
    在确定所述任务标识对应的任务未结束的情况下，将所述融合结果分发给n个所述数据处理服务，以使每个所述数据处理服务基于所述融合结果继续对对应的数据集进行处理，直至确定所述任务标识对应的任务已结束，根据最新的n个处理结果，确定所述任务标识对应的联邦计算结果。
  11. 如权利要求7所述的装置，其中，所述确定模块，具体用于：
    获取任一数据处理服务返回的融合结果，其中，所述融合结果为所述任一数据处理服务将每个所述数据处理服务的处理结果进行融合后生成的；
    在确定所述任务标识对应的任务已结束的情况下，确定所述融合结果为所述任务标识对应的联邦计算结果；
    在确定所述任务标识对应的任务未结束的情况下，将所述融合结果发送给所述任一数据处理服务，以使每个所述数据处理服务基于所述融合结果继续对对应的数据集进行处理，直至确定所述任务标识对应的任务已结束，将最新的融合结果，确定所述任务标识对应的联邦计算结果。
  12. 如权利要求7-11任一所述的装置，其中，所述确定模块，还用于：
    将所述联邦计算结果发送给各个所述参与方。
  13. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-6中任一项所述的方法。
  14. 一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使所述计算机执行根据权利要求1-6中任一项所述的方法。
  15. 一种计算机程序产品，包括计算机指令，所述计算机指令在被处理器执行时实现根据权利要求1-6中任一项所述方法的步骤。
PCT/CN2021/130307 2021-06-18 2021-11-12 联邦计算的处理方法、装置、电子设备和存储介质 WO2022262183A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
US18/012,967 US20230251898A1 (en) 2021-06-18 2021-11-12 Federated computing method, electronic device and storage medium
EP21945762.9A EP4160440A4 (en) 2021-06-18 2021-11-12 METHOD AND DEVICE FOR FEDERATED DATA PROCESSING, ELECTRONIC DEVICE AND STORAGE MEDIUM
JP2022581625A JP2023534917A (ja) 2021-06-18 2021-11-12 フェデレーションコンピューティングの処理方法、フェデレーションコンピューティングの処理装置、電子機器及び記憶媒体
KR1020227045230A KR20230006033A (ko) 2021-06-18 2021-11-12 연합 계산 처리 방법, 장치, 전자 기기 및 저장 매체

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110677839.9A CN113537508B (zh) 2021-06-18 2021-06-18 联邦计算的处理方法、装置、电子设备和存储介质
CN202110677839.9 2021-06-18

Publications (1)

Publication Number Publication Date
WO2022262183A1 true WO2022262183A1 (zh) 2022-12-22

Family

ID=78096309

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/130307 WO2022262183A1 (zh) 2021-06-18 2021-11-12 联邦计算的处理方法、装置、电子设备和存储介质

Country Status (6)

Country Link
US (1) US20230251898A1 (zh)
EP (1) EP4160440A4 (zh)
JP (1) JP2023534917A (zh)
KR (1) KR20230006033A (zh)
CN (1) CN113537508B (zh)
WO (1) WO2022262183A1 (zh)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537508B (zh) * 2021-06-18 2024-02-02 百度在线网络技术（北京）有限公司 联邦计算的处理方法、装置、电子设备和存储介质
CN113973125A (zh) * 2021-10-26 2022-01-25 杭州博盾习言科技有限公司 联邦学习中的通信方法及装置、电子设备、存储介质
CN114900325B (zh) * 2022-03-25 2024-03-26 杭州博盾习言科技有限公司 基于联邦学习的隐私集合求交方法、系统、设备及介质
US20230315604A1 (en) * 2022-03-30 2023-10-05 Lenovo Global Technology (United States) Inc. Using a task identifier to identify processing services that have processed a task
CN115378624B (zh) * 2022-04-14 2024-01-30 中国电信股份有限公司 知识图谱构建方法、装置、电子设备及存储介质
CN114969444A (zh) * 2022-05-24 2022-08-30 北京百度网讯科技有限公司 数据的处理方法、装置、电子设备和存储介质
CN114968525B (zh) * 2022-05-26 2023-03-24 深圳致星科技有限公司 隐私计算和隐私数据保护的云原生任务调度方法及装置
US20230394152A1 (en) * 2022-06-01 2023-12-07 Micron Technology, Inc. Establishing a chain of ownership of a device
CN114925072B (zh) * 2022-06-13 2023-07-21 深圳致星科技有限公司 数据管理方法、装置、系统、设备、介质
CN115827526B (zh) * 2022-11-07 2023-10-27 北京百度网讯科技有限公司 数据处理方法、装置、设备以及存储介质
CN116737348B (zh) * 2023-08-14 2024-01-02 腾讯科技（深圳）有限公司 一种多方任务的处理方法、装置及计算机设备、存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190122141A1 (en) * 2017-10-23 2019-04-25 Microsoft Technology Licensing, Llc Fast hyperparameter search for machine-learning program
US20200319922A1 (en) * 2019-04-04 2020-10-08 EMC IP Holding Company LLC Method, electronic device and computer program product for processing task
CN111898137A (zh) * 2020-06-30 2020-11-06 深圳致星科技有限公司 一种联邦学习的隐私数据处理方法、设备及系统
CN111914038A (zh) * 2020-09-21 2020-11-10 北京百度网讯科技有限公司 联邦计算方法、装置、设备及存储介质
CN112148437A (zh) * 2020-10-21 2020-12-29 深圳致星科技有限公司 用于联邦学习的计算任务加速处理方法、装置及设备
CN112200713A (zh) * 2020-10-28 2021-01-08 支付宝(杭州)信息技术有限公司 一种联邦学习中的业务数据处理方法、装置以及设备
CN113537508A (zh) * 2021-06-18 2021-10-22 百度在线网络技术（北京）有限公司 联邦计算的处理方法、装置、电子设备和存储介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114787833A (zh) * 2019-09-23 2022-07-22 普雷萨根私人有限公司 分散式人工智能(ai)/机器学习训练系统
CN112860970B (zh) * 2021-03-02 2024-03-12 百度在线网络技术（北京）有限公司 一种数据处理方法、装置、电子设备及存储介质

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190122141A1 (en) * 2017-10-23 2019-04-25 Microsoft Technology Licensing, Llc Fast hyperparameter search for machine-learning program
US20200319922A1 (en) * 2019-04-04 2020-10-08 EMC IP Holding Company LLC Method, electronic device and computer program product for processing task
CN111898137A (zh) * 2020-06-30 2020-11-06 深圳致星科技有限公司 一种联邦学习的隐私数据处理方法、设备及系统
CN111914038A (zh) * 2020-09-21 2020-11-10 北京百度网讯科技有限公司 联邦计算方法、装置、设备及存储介质
CN112148437A (zh) * 2020-10-21 2020-12-29 深圳致星科技有限公司 用于联邦学习的计算任务加速处理方法、装置及设备
CN112200713A (zh) * 2020-10-28 2021-01-08 支付宝(杭州)信息技术有限公司 一种联邦学习中的业务数据处理方法、装置以及设备
CN113537508A (zh) * 2021-06-18 2021-10-22 百度在线网络技术（北京）有限公司 联邦计算的处理方法、装置、电子设备和存储介质

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JI LIU; JIZHOU HUANG; YANG ZHOU; XUHONG LI; SHILEI JI; HAOYI XIONG; DEJING DOU: "From Distributed Machine Learning to Federated Learning: A Survey", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 10 May 2021 (2021-05-10), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081956316 *
See also references of EP4160440A4 *
VERBRAEKEN, J. ET AL.: "A survey on distributed machine learning", ACM COMPUTING SURVEYS, 31 March 2020 (2020-03-31), pages 1 - 33, XP058666756, DOI: 10.1145/3377454 *

Also Published As

Publication number Publication date
KR20230006033A (ko) 2023-01-10
CN113537508B (zh) 2024-02-02
EP4160440A4 (en) 2024-01-03
JP2023534917A (ja) 2023-08-15
CN113537508A (zh) 2021-10-22
EP4160440A1 (en) 2023-04-05
US20230251898A1 (en) 2023-08-10

Similar Documents

Publication Publication Date Title
WO2022262183A1 (zh) 联邦计算的处理方法、装置、电子设备和存储介质
KR20210090122A (ko) 분산형 모델 트레이닝 장치, 방법 및 컴퓨터 프로그램
KR20230153448A (ko) 연합 학습 방법, 장치, 전자 기기 및 저장 매체
CN114675965B (zh) 联邦学习方法、装置、设备和介质
US12034820B2 (en) Fusing and degradation method and apparatus for micro-service, device, and medium
WO2024082515A1 (zh) 一种去中心化的联邦聚类学习方法、装置、设备及介质
US20210097106A1 (en) Generation and use of a dynamic bloom filter
CN113032817B (zh) 基于区块链的数据对齐方法、装置、设备和介质
CN111767149B (zh) 调度方法、装置、设备及存储设备
US11755567B1 (en) Blockchain-based data verification method and apparatus, device, and storage medium
CN113360689B (zh) 图像检索系统、方法、相关装置及计算机程序产品
CN113935069B (zh) 一种基于区块链的数据验证方法、装置、设备及存储介质
US11809992B1 (en) Applying compression profiles across similar neural network architectures
CN114897067A (zh) 基于联邦学习的决策模型训练方法、装置和联邦学习系统
CN112328807A (zh) 反作弊方法、装置、设备以及存储介质
CN112559547A (zh) 确定多存储对象副本之间一致性的方法及装置
US20230186102A1 (en) Training method and apparatus for neural network model, device and storage medium
EP3992794A2 (en) Method and apparatus for acquiring information
US11853814B2 (en) Automatically generating events
CN114071111A (zh) 视频播放方法和装置
CN117112576A (zh) 数据同步方法及装置、服务器和存储介质
CN116069764A (zh) 数据校验方法、装置、电子设备和存储介质
CN118433039A (zh) 模型训练时的节点通信方法、装置
CN117349254A (zh) 日志文件处理方法、装置、系统、电子设备及存储介质
CN115752496A (zh) 路径搜索方法、装置、电子设备和存储介质

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 20227045230

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2022581625

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021945762

Country of ref document: EP

Effective date: 20221228

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

Ref document number: 21945762

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE