Contribution degree evaluation method and device, equipment and storage medium

Abstract

The invention provides a contribution degree evaluation method and device, equipment and a storage medium, and the method comprises the steps: determining a first model prediction value according to a trained local model corresponding to an aggregation party; obtaining a second model prediction value corresponding to the common participant; wherein the second model prediction value is determined based on the trained local models corresponding to the common participants respectively; obtaining an aggregation result according to the first model prediction value and the second model prediction value, and determining contribution degrees corresponding to the aggregation party and the common participant based on the aggregation result; according to the method, fairness and accuracy of contribution degree evaluation can be ensured according to the predicted values corresponding to the participants, participation of a third party is not needed, and the safety problem caused by the third party is avoided.

Description

Technical Field

[0001] The present application relates to the field of data contribution evaluation, and in particular to a contribution evaluation method, device, equipment and storage medium. Background Art

[0002] With the rapid development of big data and artificial intelligence, data has become the core driving force for technological progress and business innovation. With the explosive growth of data and the increase in its complexity, the value of data has become increasingly prominent. However, the reality is that data is often scattered among different institutions or departments, forming the so-called "data islands". This dispersion not only hinders the effective use of data, but also limits cross-organizational collaborative research and development. In order to break this situation and realize data sharing and collaborative learning, federated learning has been proposed as a distributed learning paradigm and has been widely used.

[0003] To cope with the challenge of feature dispersion, federated learning is divided into two main types: horizontal federated learning and vertical federated learning. Among them, vertical federated learning (VFL) focuses on dealing with situations where data features are dispersed among different participants. Specifically, VFL is suitable for situations where different participants have data sets of the same users but different features. For example, in the financial field, a bank may have a customer's transaction records, while another insurance company has the customer's health information. Through VFL, these institutions can jointly train models without sharing the original data, thereby improving model performance while protecting user privacy. In the federated learning scenario, since the participants do not share the original data, how to fairly and reasonably evaluate the data contribution of each participant has become a key issue. The data contribution evaluation is not only related to the distribution of interests among the participants, but also directly affects the stability and sustainability of the federated learning system. If the contribution of each participant cannot be accurately measured, it may lead to unreasonable resource allocation, which in turn affects the enthusiasm of all parties and even destroys the collaboration mechanism of the entire system.

[0004] In existing federated learning schemes, a third-party coordinator is usually introduced to ensure the security and transparency of the entire process. This third party is responsible for key tasks such as managing encryption keys, verifying model updates, and evaluating data contribution. However, this model of relying on a third party also brings several problems: for example, due to frequent interactions with third parties, especially when a large number of participants are involved, the communication overhead increases significantly, resulting in a decrease in overall computing efficiency. Although the existence of a third party increases security, it also introduces new trust issues, namely how to ensure that the third party itself will not become a new single point of failure or a source of privacy leakage risk. Summary of the invention

[0005] The present application provides a contribution evaluation method, device, equipment and storage medium, which ensures the fairness and accuracy of contribution evaluation based on the corresponding predicted values ​​of each participant, without the need for third-party participation, thus avoiding security issues caused by third parties.

[0006] In a first aspect, the present application provides a contribution evaluation method, which is applied to an aggregation party, including:

[0007] Determine a first model prediction value according to the trained local model corresponding to the aggregator;

[0008] Obtaining a second model prediction value corresponding to the common participant; wherein the second model prediction value is determined based on a trained local model corresponding to each of the common participants;

[0009] An aggregation result is obtained according to the first model prediction value and the second model prediction value, and the corresponding contribution of the aggregator and the ordinary participant is determined based on the aggregation result.

[0010] In one or more possible embodiments, when the model training is determined as multi-party vertical federated learning according to a preset routing rule, obtaining the second model prediction value corresponding to the ordinary participant includes:

[0011] Receive the first prediction value sent by each of the ordinary participants and the second prediction value sent by the last ordinary participant; wherein the last ordinary participant can be determined according to the corresponding routing rule.

[0012] In one or more possible embodiments, the first prediction value is obtained in the following manner:

[0013] For the i-th ordinary participant, the second model prediction value W is divided into i X i Decompose into and and the As the first prediction value;

[0014] The second predicted value is obtained in the following manner:

[0015] For the i-th ordinary participant, Sent to the i+1th ordinary participant, and according to the and the i+1th ordinary participant Regarding the Update and get the updated Until the last common participant updates The updated as the second prediction value.

[0016] In one or more possible embodiments, when the model training is determined as two-party vertical federated learning according to a preset routing rule, obtaining an aggregated result according to the first model prediction value and the second model prediction value includes:

[0017] Encrypting the first model prediction value and the second model prediction value using a preset public key;

[0018] The encrypted first model prediction value and the second model prediction value are aggregated using homomorphic encryption to obtain an aggregated result.

[0019] In one or more possible embodiments, determining the contribution degree of each of the aggregator and the ordinary participant based on the aggregation result includes:

[0020] Determining a first contribution degree corresponding to the aggregation party according to the aggregation result;

[0021] Sending the aggregation result to each of the common participants, so that each of the common participants determines the second contribution corresponding to each of them according to the aggregation result;

[0022] The first contribution degree and the second contribution degree are normalized to obtain the corresponding contribution degrees of the aggregator and the ordinary participant respectively.

[0023] In the second aspect, the present application provides a contribution evaluation method, which is applied to ordinary participants, including:

[0024] For any common participant, determine a second model prediction value according to the trained local model corresponding to the common participant;

[0025] Send the second model prediction value to the aggregator so that the aggregator obtains the aggregation result according to the first model prediction value and the second model prediction value; wherein the first model prediction value is determined based on the trained local model corresponding to the aggregator.

[0026] In a third aspect, the present application provides a contribution evaluation device, which is applied to an aggregation party, including:

[0027] A first model prediction value determination module, used to determine a first model prediction value according to a trained local model corresponding to the aggregator;

[0028] A second model prediction value acquisition module, used to acquire a second model prediction value corresponding to the ordinary participant; wherein the second model prediction value is determined based on the trained local model corresponding to each of the ordinary participants;

[0029] A contribution determination module is used to obtain an aggregation result according to the first model prediction value and the second model prediction value, and determine the contribution corresponding to each of the aggregating party and the ordinary participant based on the aggregation result.

[0030] In a fourth aspect, the present application provides a contribution evaluation device, which is applied to ordinary participants, including:

[0031] A second model prediction value determination module is used to determine, for any common participant, a second model prediction value according to a trained local model corresponding to the common participant;

[0032] The second model prediction value sending module is used to send the second model prediction value to the aggregation party so that the aggregation party obtains the aggregation result according to the first model prediction value and the second model prediction value; wherein the first model prediction value is determined based on the local model corresponding to the aggregation party that has completed training.

[0033] In a fifth aspect, the present application provides a contribution evaluation device, the device comprising:

[0034] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable 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 any one of the methods in the first aspect or the second aspect.

[0035] In a sixth aspect, the present application provides a computer storage medium storing a computer program for causing a computer to execute any one of the methods in the first aspect or the second aspect.

[0036] According to a contribution evaluation method, device, equipment and storage medium provided in this application, the fairness and accuracy of the contribution evaluation are ensured according to the corresponding predicted values ​​of each participant, without the need for third party participation, thus avoiding security issues caused by third parties. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application, and do not constitute an improper limitation on the present application.

[0038] Figure 1 A flowchart of a contribution evaluation method provided according to an embodiment;

[0039] Figure 2 A diagram of a contribution evaluation device applied to an aggregation party according to an embodiment;

[0040] Figure 3 A diagram of a contribution evaluation device for a common participant provided in accordance with an embodiment;

[0041] Figure 4 A schematic diagram of a contribution evaluation device provided according to an embodiment;

[0042] Figure 5 The present invention is a schematic diagram of a computer storage medium provided according to an embodiment. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described here can be implemented in an order other than those illustrated or described here. The implementation methods described in the following exemplary embodiments do not represent all implementation methods consistent with the present disclosure. The embodiments of the present disclosure may include multiple steps. For ease of description, these steps are numbered, but these numbers are not limitations on the execution time slots and execution order between the steps; these steps can be implemented in any order, and the embodiments of the present disclosure do not limit this. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the attached claims.

[0045] Furthermore, in the description of the embodiments of the present application, unless otherwise specified, “ / ” means or. For example, A / B can mean A or B. The “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0046] Example 1

[0047] This application provides a contribution evaluation method, such as Figure 1 As shown, including:

[0048] Step 101, determining a first model prediction value according to the trained local model corresponding to the above-mentioned aggregation party;

[0049] In one or more possible embodiments, before the vertical federated learning performs model training, initialization is required. First, the current task ID needs to be obtained from the request parameter object. The above-mentioned request parameter object refers to any participating object of the federated learning. The vertical federated learning client object, that is, all participants in the vertical federated learning, is obtained from the memory according to the above-mentioned task ID. If the client object in the memory is empty, it is necessary to request the parameter object to create a new client object and put it into the memory. The key value is the task ID, and the value object is the vertical federated learning object; then the vertical federated training task is executed. After the execution is completed, it is determined that each participant has a corresponding local model, and then the aggregator can input the corresponding local data set into the corresponding local model according to the aggregator to obtain the first model prediction value.

[0050] Step 102, obtaining the second model prediction value corresponding to the above-mentioned ordinary participants; wherein the second model prediction value is determined based on the trained local models corresponding to the above-mentioned ordinary participants;

[0051] In one or more possible embodiments, based on the same operation method, each ordinary participant also corresponds to a local model, and the local data corresponding to the ordinary participant is input into the corresponding local model to obtain the second model prediction value.

[0052] Step 103, obtaining an aggregation result according to the prediction value of the first model and the prediction value of the second model, and determining the corresponding contribution of the aggregator and the ordinary participant based on the aggregation result.

[0053] In one or more possible embodiments, the first model prediction value and the second model prediction value are aggregated to obtain an aggregation result, and the aggregation result includes the model prediction values ​​of all participants; based on the aggregation result, the corresponding contribution of the above-mentioned aggregation party and the above-mentioned ordinary participants is determined, including: determining the first contribution of the above-mentioned aggregation party according to the above-mentioned aggregation result; sending the above-mentioned aggregation result to each of the above-mentioned ordinary participants, so that each of the above-mentioned ordinary participants determines the corresponding second contribution according to the above-mentioned aggregation result; normalizing the above-mentioned first contribution and the above-mentioned second contribution to obtain the corresponding contribution of the above-mentioned aggregation party and the above-mentioned ordinary participants; specifically, the aggregation party can quantify the contribution of each participant by analyzing the influence of the prediction value provided by each participant on the overall performance, which can be achieved in the following ways: (1) Feature importance: If the participants provide different feature subsets, the contribution can be evaluated by analyzing which features have the greatest influence on the final prediction result. (2) Incremental improvement: gradually add the data of each participant, observe the changes in model performance, and measure the incremental improvement brought by each participant. (3) Shapley value: The Shapley value method in game theory is used to accurately allocate the contribution share of each participant to the overall performance improvement. Other methods are not described here. In order to make the contributions of different participants comparable, the calculated contributions need to be normalized so that the sum of the contributions of all participants is equal to 1 or a fixed value. Finally, the corresponding contributions of the above-mentioned aggregator and the above-mentioned ordinary participants are determined.

[0054] In one or more possible embodiments, first, it can be determined whether it is a two-party or multi-party vertical federated learning based on the length of the participant IP array. If the length of the participant IP array is 2, it means that it is a two-party vertical federated learning. The participant address of the configuration class of the vertical federated learning is set to the IP address, the routing rule is other, and the roles are host (the host party is the aggregator in this application) and guest (the guest party is the ordinary participant in this application). If the length of the participant IP array is greater than 2, it means that it is a three-party or more vertical federated learning. Set the routing rules of each party and proceed as follows: The first participant sets its routing rule to agg (aggregator). It needs to communicate with all ordinary participants, so The routing array of the configuration class object is set to the IP addresses of all ordinary participants. The ordinary participant sets its routing rule to simple (ordinary participant). It needs to communicate with the aggregator and the next ordinary participant. The routing array of the configuration class object is set to the IP of the aggregator and the IP of the next ordinary participant. The last ordinary participant sets its routing rule to simple-end (the last ordinary participant). It only needs to communicate with the first participant (that is, the aggregator). The routing array of the configuration class object is set to the IP of the first participant. The above routing rules are not limited to other, agg, simple and simple-end mentioned in this application, as long as they can be distinguished.

[0055] In one or more possible embodiments, since the routing rules corresponding to two-party vertical federated learning and multi-party vertical federated learning are different, it is possible to determine whether the current two-party vertical federated learning or multi-party vertical federated learning is based on the routing rules of the participants.

[0056] In one or more possible embodiments, when the model training is determined as multi-party vertical federated learning according to the preset routing rules, the obtaining of the second model prediction value corresponding to the ordinary participant includes: receiving the first prediction value sent by each of the ordinary participants and the second prediction value sent by the last ordinary participant; wherein the last ordinary participant can be determined according to the corresponding routing rules; the first prediction value sent by each of the ordinary participants is determined in the following manner: for the i-th ordinary participant, the second model prediction value W is divided according to the preset sharding method. i X i Decompose into and And the above as the first prediction value; the second prediction value is obtained in the following way: for the i-th ordinary participant, the above Sent to the i+1th ordinary participant, and according to the above and the i+1th ordinary participant For the above Update and get the updated Until the last common participant updates The updated as the second predicted value.

[0057] In one or more possible embodiments, a specific process of a contribution method of multi-party vertical federated learning is given:

[0058] The aggregator determines the predicted value W based on the local data set and the corresponding trained local model. agg X agg , and generate an empty set Z part1 ;

[0059] The first ordinary participant determines the corresponding prediction value W1X1 based on the local data set and the corresponding trained local model, and decomposes the prediction value into and Will Sent to the above aggregator so that the aggregator can and the empty set Z part1 Update Z after merging part1 , and Send to the second ordinary party;

[0060] The second ordinary participant determines the corresponding prediction value W2X2 based on the local data set and the corresponding trained local model, and decomposes the prediction value into and Will Sent to the above aggregator so that the aggregator can With the updated Z part1 Merge and update Z again part1 ; and and After merging, they are sent to a third common participant;

[0061] The i-th ordinary participant determines the corresponding prediction value W based on the local data set and the corresponding trained local model i X i , and decompose the predicted value into and Will Sent to the above aggregator so that the aggregator can Update Z part1 , and and After merging, they are sent to the next common participant;

[0062] The last ordinary participant determines the corresponding prediction value W based on the local data set and the corresponding trained local model. end X end , and decompose the predicted value into and And use the same method as above to and To process, in fact, all ordinary participants Merge into Z part2 , and Z part2 Send to the above aggregator;

[0063] The above-mentioned aggregation party converts the above-mentioned predicted value W agg X agg , Z part1 and Z part2 Polymerization to obtain Z * , and Z * Encrypted broadcast to all other ordinary participants, the aggregator uses Z * Calculate your first contribution, and all other ordinary participants receive and decrypt to get Z * , and similarly calculates its own second contribution and returns it to the aggregator. After the aggregator successfully receives it, it normalizes the above-mentioned first contribution and the above-mentioned second contribution, and finally obtains the corresponding contribution of the aggregator and all other ordinary participants.

[0064] In one or more possible embodiments, when it is determined that the model training is two-party vertical federated learning according to the preset routing rule, the above-mentioned obtaining the aggregation result according to the above-mentioned first model prediction value and the above-mentioned second model prediction value includes: encrypting the above-mentioned first model prediction value and the above-mentioned second model prediction value using a preset public key; aggregating the encrypted first model prediction value and the above-mentioned second model prediction value using a homomorphic encryption method to obtain an aggregation result;

[0065] In one or more possible embodiments, when determining that model training is two-party vertical federated learning according to a preset routing rule, a specific process of a contribution method of two-party vertical federated learning is provided:

[0066] The aggregator determines the predicted value W1X1 based on the local data set X1 and the corresponding trained local model W1 and encrypts W1X1. The common participant determines the predicted value W2X2 based on the local data set X2 and the corresponding trained local model W2 and encrypts W2X2.

[0067] The encrypted W1X1 can be sent to the common participants, and then the common participants perform ciphertext aggregation on the encrypted W1X1 and the encrypted W2X2 according to the homomorphism of the Paillier key to obtain an aggregation result, and then send the above aggregation result to the aggregation party, or the common participants send the encrypted W2X2 to the aggregation party, and the aggregation party performs ciphertext aggregation on the encrypted W1X1 and the encrypted W2X2 according to the homomorphism of the Paillier key to obtain an aggregation result;

[0068] The aggregator decrypts the above aggregation result to obtain Z, and adds a one-time random number mask r to Z to obtain Z * And send it to ordinary participants in encrypted form, and the aggregator will * Calculate the first contribution, and the ordinary participants decrypt to get Z * , calculate the second contribution, and send the second contribution to the aggregator; finally, the aggregator normalizes the first contribution and the second contribution, and finally obtains the corresponding contribution of the aggregator and all other ordinary participants.

[0069] According to a contribution evaluation method provided by the present application, in a two-party federated learning scenario, the aggregator and the ordinary participant are allowed to share the model prediction value in an encrypted manner without exposing their respective original data. By utilizing the characteristics of the Paillier homomorphic encryption technology, both parties can perform aggregation operations on the prediction value in a ciphertext state, thereby avoiding the risk of data leakage. The aggregator and the ordinary participant calculate their respective model prediction values ​​respectively, and send them to each other after encrypting them with the public key. After receiving the prediction value encrypted by the other party, the ciphertext aggregation is performed using the homomorphism of the Paillier key, and then the first contribution and the second contribution of each are calculated, and finally the corresponding contribution is obtained after normalization; in a multi-party federated learning scenario, due to the large number of participants, direct global data aggregation and contribution calculation will face huge communication and computing overheads. Therefore, the present application adopts a shard aggregation method to shard the model prediction value of the ordinary participant and transmit it to the adjacent participant or the aggregator. In this way, the number of participants in direct communication is reduced, and the communication cost is reduced. At the same time, the aggregator, as a central node, is responsible for collecting all shards and performing global merging to finally obtain a complete set of prediction values. After obtaining the global prediction value, the aggregator uses homomorphic encryption technology to encrypt it and broadcast it to all participants. Each participant calculates its second contribution based on the received global data and returns it to the aggregator. The aggregator finally normalizes the first contribution and the second contribution to obtain the corresponding contribution of all participants. This shard aggregation and global evaluation strategy not only improves the efficiency of multi-party federated learning, but also ensures the accurate evaluation of the contribution of each participant.

[0070] Example 2

[0071] Based on the same inventive concept, the present application also provides a contribution evaluation method, which is applied to ordinary participants, including: for any ordinary participant, according to the local model that has completed training corresponding to the above-mentioned ordinary participant, determine the second model prediction value, and send it to the aggregator, so that the above-mentioned aggregator obtains the aggregation result according to the first model prediction value and the above-mentioned second model prediction value; wherein the above-mentioned first model prediction value is determined based on the local model that has completed training corresponding to the above-mentioned aggregator; the description of the above-mentioned ordinary participant has been shown in Example 1 and will not be repeated here.

[0072] Example 3

[0073] This application provides a contribution evaluation device, such as Figure 2 As shown, it is applied to the aggregation side, including:

[0074] A first model prediction value determination module 201 is used to determine a first model prediction value according to the trained local model corresponding to the above-mentioned aggregation party;

[0075] The second model prediction value acquisition module 202 is used to acquire the second model prediction value corresponding to the above-mentioned ordinary participants; wherein the above-mentioned second model prediction value is determined based on the local model corresponding to each of the above-mentioned ordinary participants after training;

[0076] The contribution determination module 203 is used to obtain an aggregation result according to the first model prediction value and the second model prediction value, and determine the contribution corresponding to each of the aggregation party and the ordinary participant based on the aggregation result.

[0077] Corresponding to the above contribution evaluation method, the present invention also provides a contribution evaluation device. Since the device embodiment of the present invention corresponds to the above method embodiment, details not disclosed in the device embodiment can be referred to the above method embodiment, and will not be repeated in the present invention.

[0078] Example 4

[0079] This application provides a contribution evaluation device, such as Figure 3 As shown, it applies to ordinary participants, including:

[0080] The second model prediction value determination module 301 is used to determine the second model prediction value for any common participant according to the trained local model corresponding to the common participant;

[0081] The second model prediction value sending module 302 is used to send the above-mentioned second model prediction value to the aggregation party, so that the above-mentioned aggregation party obtains the aggregation result according to the first model prediction value and the above-mentioned second model prediction value; wherein the above-mentioned first model prediction value is determined based on the local model corresponding to the above-mentioned aggregation party that has completed training.

[0082] Example 5

[0083] The present application also provides a contribution assessment device, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable 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 above-mentioned contribution assessment method.

[0084] like Figure 4 As shown, the device includes a processor 401, a memory 402, a communication interface 403 and a bus 404. The processor 401, the memory 402 and the communication interface 403 are connected to each other via the bus 404.

[0085] The processor 401 is used to read and execute instructions in the memory 402, so that at least one processor can execute the contribution evaluation method provided in the above embodiment.

[0086] The memory 402 is used to store various instructions and programs of the contribution evaluation method provided in the above embodiment.

[0087] The bus 404 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0088] The processor 401 may be a central processing unit (CPU), a network processor (NP), a graphic processing unit (GPU), or any combination of CPU, NP, and GPU. It may also be a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0089] Example 6

[0090] In addition, the present application also provides a computer-readable storage medium, such as Figure 5 As shown, the computer storage medium stores a computer program, and the computer program is used to enable a computer to execute any one of the methods in the above embodiments.

[0091] The memory may include a readable medium in the form of a volatile memory, such as a random access memory (RAM) 501 and / or a cache memory 502 , and may further include a read-only memory (ROM) 503 .

[0092] The memory may also include a program / utility 505 having a set (at least one) of program modules 504, such program modules 504 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0093] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0094] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0095] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0097] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A contribution evaluation method, characterized in that: Applicable to the aggregator, including: Determine a first model prediction value according to the trained local model corresponding to the aggregator; Obtaining a second model prediction value corresponding to the common participant; wherein the second model prediction value is determined based on a trained local model corresponding to each of the common participants; An aggregation result is obtained according to the first model prediction value and the second model prediction value, and the corresponding contribution of the aggregator and the ordinary participant is determined based on the aggregation result.

2. The method according to claim 1, characterized in that When the model training is determined as multi-party vertical federated learning according to the preset routing rule, obtaining the second model prediction value corresponding to the ordinary participant includes: Receive the first prediction value sent by each of the ordinary participants and the second prediction value sent by the last ordinary participant; wherein the last ordinary participant can be determined according to the corresponding routing rule.

3. The method according to claim 2, characterized in that The first prediction value is obtained in the following manner: For the i-th ordinary participant, the second model prediction value W is divided into i X i Decompose into and and the As the first prediction value; The second predicted value is obtained in the following manner: For the i-th ordinary participant, Sent to the i+1th ordinary participant, and according to the and the i+1th ordinary participant Regarding the Update and get the updated Until the last common participant updates The updated as the second predicted value.

4. The method according to claim 1, characterized in that When the model training is determined as two-party vertical federated learning according to the preset routing rule, obtaining the aggregation result according to the first model prediction value and the second model prediction value includes: Encrypting the first model prediction value and the second model prediction value using a preset public key; The encrypted first model prediction value and the second model prediction value are aggregated using homomorphic encryption to obtain an aggregated result.

5. The method according to claim 1, characterized in that: The determining, based on the aggregation result, the contribution corresponding to each of the aggregator and the ordinary participant includes: Determining a first contribution degree corresponding to the aggregation party according to the aggregation result; Sending the aggregation result to each of the common participants, so that each of the common participants determines the second contribution corresponding to each of them according to the aggregation result; The first contribution degree and the second contribution degree are normalized to obtain the corresponding contribution degrees of the aggregator and the ordinary participant respectively.

6. A contribution evaluation method, characterized in that: Applicable to ordinary participants, including: For any common participant, determine a second model prediction value according to the trained local model corresponding to the common participant; Send the second model prediction value to the aggregator so that the aggregator obtains the aggregation result according to the first model prediction value and the second model prediction value; wherein the first model prediction value is determined based on the trained local model corresponding to the aggregator.

7. A contribution evaluation device, characterized in that: Applicable to the aggregator, including: A first model prediction value determination module, used to determine a first model prediction value according to a trained local model corresponding to the aggregator; A second model prediction value acquisition module, used to acquire a second model prediction value corresponding to the ordinary participant; wherein the second model prediction value is determined based on the trained local model corresponding to each of the ordinary participants; A contribution determination module is used to obtain an aggregation result according to the first model prediction value and the second model prediction value, and determine the contribution corresponding to each of the aggregating party and the ordinary participant based on the aggregation result.

8. A contribution evaluation device, characterized in that: Applicable to ordinary participants, including: A second model prediction value determination module is used to determine, for any common participant, a second model prediction value according to a trained local model corresponding to the common participant; The second model prediction value sending module is used to send the second model prediction value to the aggregation party so that the aggregation party obtains the aggregation result according to the first model prediction value and the second model prediction value; wherein the first model prediction value is determined based on the local model corresponding to the aggregation party that has completed training.

9. A contribution evaluation device, characterized in that: The device comprises: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor so that the at least one processor can execute any one of the methods of claims 1-5 or 6.

10. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and the computer program is used to make a computer execute any one of the methods according to claims 1-5 or 6.

Images