WO2022257731A1

WO2022257731A1 - Method, device and system for performing algorithm negotiation on privacy computation

Info

Publication number: WO2022257731A1
Application number: PCT/CN2022/094032
Authority: WO
Inventors: 李漓春; 阮方圆; 殷山
Original assignee: 支付宝(杭州)信息技术有限公司
Priority date: 2021-06-11
Filing date: 2022-05-20
Publication date: 2022-12-15
Also published as: CN115470513A

Abstract

A method, device and system for performing algorithm negotiation on privacy computation. The method comprises: an initiating party sends a negotiation request to a participating party, the negotiation request comprising description information of a target computing task to be subjected to privacy computation, and a privacy computation algorithm list supported by the initiating party against the target computing task (S210); the participating party selects a target algorithm from the privacy computation algorithm list (S220); and the participating party sends a feedback message to the initiating party, the feedback message comprising an algorithm identifier of the target algorithm (S230).

Description

Method, device and system for algorithm negotiation for private computing

technical field

One or more embodiments of this specification relate to the field of computer technology, and in particular to a method, device, and system for algorithm negotiation for private computing.

Background technique

In order to fully tap the value of data, multi-party joint data calculation has become a major research hotspot. For example, in order to better classify and analyze merchants, the electronic payment platform will classify merchants based on the merchant transaction data it owns, the e-merchant platform based on the merchant sales data it owns, and the banking institution based on the merchant loan data it owns. analyze.

In multi-party data joint computing, in order to protect the data privacy of all parties, it is necessary to adopt Privacy-Preserving Computation technology. Privacy-preserving computing has a variety of computing routes, such as Secure Multi-Party Computation (MPC), Trusted Execution Environment (Trusted Execution Environment, TEE), Federated Learning (Federated Learning, FL), etc. For a specific privacy computing task, almost every technical route has multiple algorithms.

Contents of the invention

One or more embodiments of this specification describe a method, device, and system for algorithm negotiation for privacy computing, which supports negotiation between the initiator and other participants, so that the two parties can meet their demands in terms of performance, security, and privacy protection strength. and preferences are fully considered, resulting in a privacy computing algorithm with high feasibility and high availability.

According to the first aspect, there is provided a method for algorithm negotiation for private computing, including: the initiator sends a negotiation request to the participant, the negotiation request includes description information of the target computing task to be performed on the private computing, and the initiator provides A list of privacy computing algorithms supported by the target computing task; the participant selects a target algorithm from the list of privacy computing algorithms; the participant sends a feedback message to the initiator, and the feedback message includes the target Algorithm ID of the algorithm.

In one embodiment, the description information of the target computing task includes a task type of the target computing task and data information of participants involved in the target computing task.

In a specific embodiment, the task type includes at least one of the following: joint statistical analysis for specified data items; joint training of specified models; joint prediction of specified models; the data information of the participants includes, Party dataset ID.

In an embodiment, the negotiation request further includes the initiator's first priority information for each privacy computing algorithm in the privacy computing algorithm list.

In a specific embodiment, the participant selects a target algorithm from the list of privacy calculation algorithms, including: the participant selects several privacy calculation algorithms supported by itself from the list of privacy calculation algorithms; The first priority information of the several privacy computing algorithms, the target algorithm is selected from the several privacy computing algorithms.

In another specific embodiment, the participant selects a target algorithm from the list of privacy calculation algorithms, including: the participant selects several privacy calculation algorithms supported by itself from the list of privacy calculation algorithms; Selecting the target algorithm from the several privacy computing algorithms according to the first priority information of the several privacy computing algorithms and the second priority information of the participant for the several privacy computing algorithms.

In an embodiment, the participant selects a target algorithm from the list of privacy calculation algorithms, including: the participant selects several privacy calculation algorithms supported by itself from the list of privacy calculation algorithms; according to the The participating party selects the target algorithm from the several privacy computing algorithms for the second priority information of the several privacy computing algorithms.

In one embodiment, it further includes: the initiator and the participant using the target algorithm to execute the target computing task.

According to the second aspect, there is provided a method for algorithm negotiation for private computing, which is executed by an initiator, including: sending a negotiation request to a participant, where the negotiation request includes description information of a target computing task for which private computing is to be performed, and A list of privacy computing algorithms supported by the initiator for the target computing task; receiving a feedback message from the participant, the feedback message including the algorithm identification of the target algorithm, the target algorithm is obtained by the participant from the privacy Select from the calculation algorithm list.

In one embodiment, the method further includes: determining the privacy computing algorithm list according to the target computing task.

In a specific embodiment, determining the list of privacy computing algorithms according to the target computing task includes: acquiring a first list of algorithms used to execute the target computing task; determining the algorithm list in the first algorithm list The privacy computing algorithm supported by the party to form the list of privacy computing algorithms.

In an embodiment, the method further includes, for each privacy computing algorithm in the privacy computing algorithm list, determining first priority information, and making the negotiation request include the first priority information.

In a specific embodiment, determining the first priority information includes: determining the first priority information according to the operating indicators of each privacy calculation algorithm, and the operating indicators include at least one of the following: occupied Computing resources, computing speed, communication cost, and privacy protection strength.

In one embodiment, the method further includes: using the target algorithm to jointly execute the target computing task with the participant.

According to a third aspect, there is provided a method for algorithm negotiation for private computing, which is executed by a participant, including: receiving a negotiation request sent by an initiator, where the negotiation request includes description information of a target computing task to be performed for private computing, and a list of privacy computing algorithms supported by the initiator for the target computing task; selecting a target algorithm from the list of privacy computing algorithms; sending a feedback message to the initiator, the feedback message including the target algorithm Algorithm ID.

In a specific embodiment, selecting a target algorithm includes: selecting several privacy computing algorithms supported by itself from the list of privacy computing algorithms; Select the target algorithm in the privacy calculation algorithm.

In another specific embodiment, selecting a target algorithm includes: selecting several privacy computing algorithms supported by itself from the list of privacy computing algorithms; according to the first priority information of the several privacy computing algorithms, and the The participating party selects the target algorithm from the several privacy computing algorithms for the second priority information of the several privacy computing algorithms.

In a more specific embodiment, the first priority information includes a first ranking, and the second priority information includes a second ranking; the selection of the target algorithm from the plurality of privacy calculation algorithms includes : Determine the comprehensive ranking according to the first ranking and the second ranking of the plurality of privacy computing algorithms; determine the privacy computing algorithm with the highest comprehensive ranking as the target algorithm.

In one embodiment, selecting a target algorithm includes: selecting several privacy computing algorithms supported by itself from the list of privacy computing algorithms; according to the second priority information of the participants for the several privacy computing algorithms, from The target algorithm is selected from the plurality of privacy calculation algorithms.

According to the fourth aspect, there is provided a system for algorithm negotiation for private computing, including: an initiator, configured to send a negotiation request to a participant, where the negotiation request includes description information of a target computing task to be performed for private computing, and A list of privacy computing algorithms supported by the initiator for the target computing task; the participant is configured to select a target algorithm from the list of privacy computing algorithms; the participant is also configured to send A feedback message, where the feedback message includes the algorithm identifier of the target algorithm.

According to the fifth aspect, there is provided a device for algorithm negotiation for privacy calculation, the device is integrated in the initiator, and includes: a request sending unit configured to send a negotiation request to the participant, the negotiation request includes the privacy calculation to be performed Description information of the target computing task, and a list of privacy computing algorithms supported by the initiator for the target computing task; a message receiving unit configured to receive a feedback message from the participant, the feedback message including the algorithm identification of the target algorithm , the target algorithm is selected by the participant from the list of privacy calculation algorithms.

According to the sixth aspect, there is provided a device for algorithm negotiation for privacy calculation, the device is integrated with the participants, including: a request receiving unit configured to receive the negotiation request sent by the initiator, the negotiation request includes, the privacy calculation is to be performed The description information of the target computing task, and the list of privacy computing algorithms supported by the initiator for the target computing task; the algorithm selection unit is configured to select the target algorithm from the list of privacy computing algorithms; the message sending unit is configured to Sending a feedback message to the initiator, where the feedback message includes the algorithm identifier of the target algorithm.

According to a seventh aspect, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed in a computer, it causes the computer to execute the method provided in the first aspect or the second aspect or the third aspect .

According to an eighth aspect, there is provided a computing device, including a memory and a processor, wherein executable code is stored in the memory, and when the processor executes the executable code, the first aspect or the second aspect or the third aspect is realized. methods provided in the aspect.

According to the method and device provided in the embodiments of this specification, the initiator of privacy computing sends a negotiation request to the participant, and the negotiation request includes the task information of the target computing task and the list of privacy computing algorithms supported by the initiator for the target computing task, so as to participate in The party can select the target algorithm from the privacy computing algorithm list based on its own demands, and feed it back to the initiator, and then the two parties can jointly complete the target computing task based on the mutually negotiated high-feasibility and high-availability target algorithm.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without making creative efforts.

FIG. 1 shows a schematic diagram of an interaction scenario of a two-party negotiation privacy calculation algorithm according to an embodiment;

Fig. 2 shows a schematic diagram of two-party interaction for algorithm negotiation for privacy calculation according to an embodiment;

Fig. 3 shows a schematic structural diagram of a system for algorithm negotiation for privacy calculation according to an embodiment;

Fig. 4 shows a schematic structural diagram of a device for algorithm negotiation for privacy calculation according to an embodiment;

Fig. 5 shows a schematic structural diagram of an apparatus for algorithm negotiation for privacy calculation according to another embodiment.

Detailed ways

The solutions provided in this specification will be described below in conjunction with the accompanying drawings.

Usually, for a certain privacy computing task, the initiator specifies the privacy computing algorithm (or privacy computing scheme). This specification provides another solution to determine the privacy calculation algorithm.

The solution proposed by the inventor allows the initiator to negotiate with other participants, and obtain a privacy computing algorithm that fully considers the needs of all parties through negotiation. In this scheme, the initiator sends a negotiation request to other participants (or participants for short in the text), which indicates the target computing task and includes a list of privacy computing algorithms supported by the initiator for the target computing task, and then the participant Select the target algorithm from the list of privacy computing algorithms based on your own demands, and feed back to the initiator. In this way, the negotiation of the privacy computing algorithm can be completed, so that subsequent parties can jointly perform target computing tasks based on the negotiated privacy computing algorithm.

To facilitate intuitive understanding, FIG. 1 shows a schematic diagram of an interaction scenario of a two-party negotiation privacy calculation algorithm according to an embodiment. As shown in Figure 1, the payment platform and the bank are the initiator and participant of the privacy computing task respectively; first, the payment platform sends a negotiation request to the bank, which includes the task information of the target computing task, for example, the joint training of the user classification model, The negotiation request also includes a list of privacy computing algorithms supported by the payment platform for the target computing task. Figure 1 shows that the list includes privacy computing algorithm A ₁ , privacy computing algorithm A ₂ , ... privacy computing algorithm A _n ; then, the bank Based on the task information of the target computing task in the negotiation request, the privacy computing algorithm A _i supported by itself is selected from the list of privacy computing algorithms, and fed back to the payment platform; furthermore, the payment platform and the bank can use the privacy computing algorithm A _i to jointly Execute the target computing task.

Below, in combination with specific embodiments, the implementation steps of the above inventive concepts are described. Fig. 2 shows a schematic diagram of two-party interaction for algorithm negotiation for privacy computing according to an embodiment. Both are data parties, and can be implemented as devices, platforms, servers or device clusters with computing and processing capabilities, so that each can independently store and process the data it holds. Exemplarily, the initiator and the participant may be different enterprises, institutions, platforms, or may be different departments in the same enterprise.

As shown in Figure 2, the interaction process includes the following steps: Step S210, the initiator sends a negotiation request to the participant, the negotiation request includes description information of the target computing task to be performed on privacy computing, and the initiator's policy on the target computing task List of supported privacy calculation algorithms.

In one embodiment, the above description information includes the task type of the target computing task. In a specific embodiment, the task type may include joint statistical analysis for specified data items. In a more specific embodiment, the specified data items may be user age, user gender, merchant average turnover, merchant category, enterprise scale, and the like. In a more specific embodiment, the joint statistical analysis may be to analyze the numerical distribution of the specified data item within its definition domain, or, between the two parties, compare the value corresponding to the specified data item, or, between the two parties Analysis and calculation operations such as summation, intersection, or averaging of specified data items. In an example, the joint statistical analysis for the specified data item may be: counting the total number of female users in the two parties. In another example, it may also be: count the number of identical users in both parties. In yet another example, it may also be: count the age distribution of all users in the two parties. In yet another example, it may be: comparing the average merchant turnover of the two parties.

In another specific embodiment, the task type of the target computing task may include joint training of specified models. In a more specific embodiment, the specified model may be a machine learning model such as a logistic regression model, a decision tree model, or a neural network model. In a more specific embodiment, the specified model may be a classification model or a regression model. In a more specific embodiment, the sample objects targeted by the specified model may be: users, merchants, commodities, events (such as login events or access events), devices (such as user terminals or Internet of Things devices), and so on. In an example, the specified model may be a user classification model or a user scoring model, such as a consumer group segmentation model, a user risk assessment model, a user anomaly identification model, and the like. In another example, the specified model may be a commodity classification model or a commodity scoring model, specifically such as a commodity popularity scoring model.

In yet another specific embodiment, the task type of the target computing task may include joint prediction of a specified model. In a more specific embodiment, the specified model may be a classification model or a regression model, and correspondingly, the joint prediction may be to predict a classification category based on a classification model, or to predict a regression value based on a regression model. In an example, the joint prediction of the above specified models may be: predicting user risk scores based on the user risk assessment model, or identifying abnormal users based on the user anomaly identification model.

The above description describes the target computing task types included in the description information. It should be understood that in addition to the above-mentioned joint statistical analysis for specified data items, joint training and joint prediction of specified models, task types can also include other types, which are not exhaustive.

In an embodiment, the description information of the target computing task further includes data information of participants involved in the target computing task. In a specific embodiment, the participant data information may include a participant dataset identifier, which indicates that the initiator hopes that the participant can use the dataset identified by the participant dataset identifier to participate in the common implement. Based on this, in a more specific embodiment, before this step, the interaction process may also include: the initiator obtains a plurality of participant data set identifiers from the participant in advance, and the corresponding plurality of participant data set For example, the obtained content can be found in Table 1; then, according to the target computing task, the initiator selects one or several dataset identifiers from the multiple participant dataset identifiers, and includes them in the above In the negotiation request, for example, the initiator selects the data set identifier 1004 from Table 1 according to the joint training task for the merchant credit evaluation model, and includes it in the above-mentioned negotiation request.

Table 1

参与方数据集标识Participant Dataset ID	简要介绍信息brief introduction information
10011001	全量用户数据集Full user data set
10021002	优质用户数据集High-quality user data set
10031003	商户数据集Merchant dataset
10041004	商户借贷数据集Merchant Loan Dataset

In another specific embodiment, the participant data information in the above description information also includes a participant data field, which indicates that the initiator hopes that the participant can use the field value corresponding to this field to participate in the above-mentioned target computing task. implement. In an example, the data field may be the category of the merchant or the time when the event occurs.

Above, the description information of the target computing task that can be included in the negotiation request, such as task type, participant data information, etc., is introduced. On the other hand, the negotiation request also includes a list of privacy computing algorithms supported by the initiator for the target computing task.

Regarding the determination of the above-mentioned privacy computing algorithm list, in one embodiment, the initiator can obtain the first algorithm list used to perform the above-mentioned target computing tasks; and then determine the privacy computing algorithm supported by the party in the first algorithm list to form A list of the above privacy calculation algorithms. It should be noted that for any privacy computing task, in general, under each privacy-preserving computing route (such as MPC, TEE, FL, etc.), one or more privacy computing algorithms for performing the task can be determined, However, different privacy computing algorithms have somewhat different requirements for the execution environment. For example, there are usually different requirements for memory space, whether to install a certain security protocol, whether to install a certain computing software, etc., and the actual environment of the task executor It is relatively fixed and limited for a period of time. Therefore, in a specific embodiment, the originator can determine that a variety of privacy computing algorithms that have been disclosed that can be used to perform the above-mentioned target computing tasks form a first algorithm list; then, the originator can The requirements of each algorithm for the execution environment (such as network environment, hardware environment, software or protocol installation, etc.), as well as the real environment of the party, determine the privacy computing algorithm supported by the party in the first algorithm list, and form the above List of privacy computing algorithms described above.

Furthermore, according to an implementation manner, the initiator further selects several algorithms from the privacy computing algorithms supported by the party to form the aforementioned list of privacy computing algorithms. The selected benchmark can be based on various reasons. For example, the data of the target computing task is too large, and some algorithms consume too much resources; the data of the target computing task is too sensitive, and algorithms with high security levels need to be used, and so on. Due to the above or other reasons, among the privacy computing algorithms supported by this party, only k of them are allowed to be used for the target computing tasks, so only these k algorithms are selected to form the above privacy computing algorithm list.

According to one embodiment, the initiator determines multiple candidate computing tasks in advance according to the data it holds and the scope of business, and determines the list of privacy computing algorithms corresponding to each candidate computing task according to its real execution environment. Correspondingly, when preparing to perform multi-party secure computing, the initiator can determine the list of privacy computing algorithms corresponding to the above target computing tasks according to the mapping relationship between its pre-established candidate computing tasks and the list of privacy computing algorithms.

It should be noted that the privacy computing algorithm includes algorithm steps for executing the target computing task. The algorithm steps are refined to the point where both parties support the privacy computing algorithm, and can jointly complete the target computing task directly based on local data and local execution environment. . In an example, assume that the task type of the above-mentioned target computing task is to count the number of identical users in the two parties. At this time, the list of privacy computing algorithms may include Algorithm A: the two parties use their stored mobile phone numbers to jointly execute a certain The Private Set Intersection (PSI) protocol under the specific algorithm, and the initiator acts as the client in the PSI protocol, and the participant acts as the server in the PSI protocol; the list can also include Algorithm B: two parties use The respective stored user ID numbers jointly execute the scheme corresponding to the pseudo-code shown in a paper, and the initiator acts as the p party recorded in the pseudo-code, and the participating party acts as the q-party recorded in the pseudo-code.

Furthermore, in some implementations, the privacy calculation algorithm is refined to correspond to information such as the version and provider of the calculation module that implements the privacy protocol. For some existing privacy protocols, they can be modularized in existing technologies and provided to data parties with secure and private computing needs. Even if the same privacy computing protocol or the same computing algorithm comes from different providers, or computing modules of different versions, there may still be differences in message format, data format, and algorithm details, resulting in the two parties being unable to communicate directly. In this case, the algorithm information of each privacy calculation algorithm may include the provider and/or version information of the calculation module in addition to the description of the algorithm used. For example, Algorithm C indicates that the two parties use the PSI calculation module 1 of version v1.0 provided by the provider ProA to perform privacy and security intersection using the user's mobile phone number stored separately. Algorithm D indicates that the two parties use the user's mobile phone number stored separately , using the PSI calculation module 2 of version v2.0 provided by the provider ProB for privacy and security calculation. Even though PSI calculation modules 1 and 2 adopt the same privacy seeking protocol, Algorithm C and Algorithm D are still regarded as different privacy calculation algorithms and are listed separately.

Based on the above, the list of privacy computing algorithms supported by the initiator for the above target computing tasks can be determined and included in the negotiation request. Further, in an embodiment, the negotiation request may also include the first priority information of each privacy calculation algorithm in the privacy calculation algorithm list. In a specific embodiment, the first priority information may include priority levels, and the priority levels of different privacy calculation algorithms may be the same or different. In another specific embodiment, the first priority information may include an algorithm score, and a higher score indicates a higher priority. According to an example, the privacy computing algorithms in the privacy computing algorithm list are arranged one by one in descending order of priority.

Regarding the determination of the above-mentioned first priority information, in a specific embodiment, the corresponding first priority information may be determined according to the operating indicators of each privacy calculation algorithm in the privacy calculation algorithm list. In a more specific embodiment, the running indicators may include: occupied computing resources, computing speed (or computing speed), communication costs, and privacy protection strength. In an example, the computing resources may include CPU resources, memory resources, hard disk resources and network resources. In one example, the communication cost is a monetary representation of resources occupied by two parties communicating. In an example, the privacy protection strength can be measured by using an existing measurement algorithm (for example, an information entropy model, etc.).

In one example, it is possible to determine the indicator scores of various operating indicators corresponding to a certain privacy computing algorithm, and then add and sum the scores of multiple indicators corresponding to multiple operating indicators, or perform weighted calculation based on artificially preset weights. and , get the algorithm score of the certain privacy calculation algorithm. Further, in a specific example, the algorithm score may be used as the first priority information. In another specific embodiment, the priority level corresponding to the algorithm score of a certain privacy calculation algorithm may be determined as the first priority level information according to the mapping relationship between the score range and the priority level established manually. In yet another specific embodiment, the index values of each privacy calculation algorithm corresponding to each operating index can be input into a pre-trained scoring model or priority prediction model, and the corresponding algorithm score or priority level can be obtained as the first priority information.

In another specific embodiment, the first priority information can be set by staff based on experience. In this way, the first priority information of each privacy computing algorithm in the privacy computing algorithm list can be determined and included in the negotiation request.

The content of the negotiation request sent by the initiator is introduced above. Correspondingly, after receiving the negotiation request, the participant may select a target algorithm from the above-mentioned privacy calculation algorithm list included in the negotiation request in step S220.

In an embodiment, this step may include: the participant selects several privacy computing algorithms supported by the participant from the aforementioned list of privacy computing algorithms. It should be understood that a participant can determine several privacy computing algorithms supported by the party based on the execution environment requirements of each privacy computing algorithm in the privacy computing algorithm list and the actual local execution environment.

Further, in a specific embodiment, the above negotiation request further includes the first priority information corresponding to each privacy calculation algorithm in the privacy calculation algorithm list. Correspondingly, the participant can select the target algorithm from the several privacy computing algorithms supported by itself according to the first priority information of the several privacy computing algorithms supported by the participant.

In a more specific embodiment, based on the first priority information, the privacy computing algorithm with the highest priority may be selected from the several privacy computing algorithms as the target algorithm. For example, the privacy calculation algorithm with the highest priority can be selected, and for another example, the privacy calculation algorithm with the highest comprehensive score can be selected. For another example, the first priority information includes the first ranking, and the first ranking indicates the corresponding ranking. At this time , the highest-ranked privacy computing algorithm among several privacy computing algorithms can be used as the above-mentioned target algorithm. Furthermore, if there are multiple parallel privacy calculation algorithms with the highest priority, one can be selected from them, or it can be handed over to manual decision-making.

In another more specific embodiment, the participant can determine its second priority information for the above-mentioned several privacy computing algorithms according to the above-mentioned target computing tasks, and then select the above-mentioned target algorithm in combination with the first priority information. It should be noted that the second priority information can reflect the preferences and tendencies of the participants on various privacy calculation algorithms, and the determination process can be similar to that of the first priority information, which will not be repeated here.

In an example, the first priority information includes the first ranking, and the second priority information includes the second ranking. At this time, the comprehensive ranking of each algorithm in the plurality of privacy calculation algorithms can be determined, and the comprehensive ranking can be the corresponding The sum of the first sorting and the second sorting; then determine the privacy calculation algorithm with the highest comprehensive ranking (top ranking) as the above target algorithm. In another example, the first priority information and the second priority information include algorithm scores. At this time, the comprehensive score of each algorithm among the several privacy calculation algorithms can be determined. Specifically, each algorithm can correspond to two parties. The scores of each algorithm are summed, averaged, or weighted and summed based on the weights set by the participants to obtain a comprehensive score; and then the privacy calculation algorithm with the highest comprehensive score among the several privacy calculation algorithms is determined as the above target algorithm .

In another specific embodiment, the participant may not consider the first priority information, but only based on the second priority information, select the privacy calculation algorithm with the highest priority from the above-mentioned several privacy calculation algorithms as the above-mentioned target algorithm. In yet another specific implementation, the participant may not consider any priority information, but directly randomly selects one of the above-mentioned several privacy calculation algorithms as the above-mentioned target algorithm.

In this way, the participants can select the target algorithm from the list of privacy calculation algorithms. Afterwards, in step S230, the participant sends a feedback message to the initiator, and the feedback message includes the algorithm identifier of the above-mentioned target algorithm. In one embodiment, the list of privacy computing algorithms includes the algorithm identification of each privacy computing algorithm, and correspondingly, the participant can obtain the algorithm identification of the target algorithm from it, and include it in the feedback message. In another embodiment, the participant can use the position of the target algorithm in the list of privacy computing algorithms (such as which row or number) as its algorithm identifier, and include it in the feedback message. In one embodiment, the above-mentioned negotiation request includes the identifier of the participant's data set, and correspondingly, the feedback message may also include a corresponding feedback field, indicating whether the participant will use the participation in the process of jointly executing the target computing task The dataset identified by the party dataset id.

From the above, the initiator can receive feedback messages from the participants. In this way, through negotiation, the two parties can obtain a target algorithm with high feasibility and strong security that meets the needs of both parties at the same time.

According to another embodiment, after step S230, the above-mentioned interaction process may further include step S240: the initiator and the participant use the above-mentioned target algorithm to jointly execute the above-mentioned target computing task. In this way, the joint execution of the target computing task can be realized according to the target algorithm negotiated by the two parties.

To sum up, in the algorithm negotiation method for privacy computing disclosed in the embodiment of this specification, the initiator of the privacy computing sends a negotiation request to the participant, and the negotiation request includes the task information of the target computing task and the initiator’s policy for the target computing task. The list of supported privacy computing algorithms, so that the participants can select the target algorithm from the list of privacy computing algorithms based on their own demands, and feed back to the initiator, and then the two parties can jointly negotiate based on the high feasibility and high availability of the target algorithm. Complete the target computing task.

Corresponding to the algorithm negotiation method described above, the embodiment of this specification also discloses an algorithm negotiation device and system.

Fig. 3 shows a schematic structural diagram of a system for algorithm negotiation for privacy calculation according to an embodiment. As shown in FIG. 3 , the system 300 includes: an initiator 310, configured to send a negotiation request to a participant, the negotiation request includes description information of a target computing task to be performed on privacy calculation, and the initiator 310 for the target A list of privacy computing algorithms supported by computing tasks; the participant 320 is configured to select a target algorithm from the list of privacy computing algorithms; the participant 320 is also configured to send a feedback message to the initiator, and the The feedback message includes the algorithm identifier of the target algorithm.

In a specific embodiment, the task type includes at least one of the following: joint statistical analysis of specified data items; joint training of specified models; joint prediction of specified models. The participant data information includes a participant dataset identifier.

In a specific embodiment, the participant 320 is specifically configured to: select several privacy computing algorithms supported by itself from the list of privacy computing algorithms; according to the first priority information of the several privacy computing algorithms, select from The target algorithm is selected from the plurality of privacy calculation algorithms.

In a specific embodiment, the participant 320 is specifically configured to: select several privacy computing algorithms supported by itself from the list of privacy computing algorithms; according to the first priority information of the several privacy computing algorithms, and The participant selects the target algorithm from the several privacy computing algorithms for the second priority information of the several privacy computing algorithms.

In one embodiment, the participant 320 is specifically configured to: select several privacy computing algorithms supported by itself from the list of privacy computing algorithms; information, and select the target algorithm from the several privacy calculation algorithms.

In one embodiment, the initiator 310 and the participant 320 are further configured to: use the target algorithm to execute the target computing task.

Fig. 4 shows a schematic structural diagram of a device for algorithm negotiation for privacy calculation according to an embodiment, the device is integrated in the initiator. As shown in FIG. 4 , the apparatus 400 includes: a request sending unit 410 configured to send a negotiation request to a participant, the negotiation request includes description information of a target computing task to be performed on privacy computing, and A list of privacy computing algorithms supported by the target computing task; the message receiving unit 420 is configured to receive a feedback message from the participant, the feedback message includes the algorithm identification of the target algorithm, and the target algorithm is obtained by the participant from the selected from the list of privacy calculation algorithms mentioned above.

In one embodiment, the apparatus 400 further includes an algorithm list determining unit 430 configured to determine the privacy computing algorithm list according to the target computing task.

In one embodiment, the algorithm list determination unit 430 is specifically configured to: obtain a first algorithm list used to execute the target computing task; determine the privacy computing algorithm supported by the party in the first algorithm list, and form The list of privacy calculation algorithms.

In one embodiment, the apparatus 400 further includes a priority determination unit 440 configured to determine first priority information for each privacy calculation algorithm in the privacy calculation algorithm list, and make the negotiation request include the first priority level information.

In a specific embodiment, the priority determination unit 440 is specifically configured to: determine the first priority information according to the operation indicators of each privacy calculation algorithm, and the operation indicators include at least one of the following: Resources, computing speed, communication cost, and privacy protection strength.

In one embodiment, the apparatus 400 further includes: a task execution unit 450 configured to use the target algorithm to jointly execute the target computing task with the participant.

Fig. 5 shows a schematic structural diagram of an apparatus for algorithm negotiation for privacy calculation according to another embodiment, the apparatus is integrated in a participant. As shown in FIG. 5 , the apparatus 500 includes: a request receiving unit 510 configured to receive a negotiation request sent by the initiator, the negotiation request including description information of the target computing task to be performed on privacy calculation, and the initiator's List of privacy computing algorithms supported by the target computing task. The algorithm selection unit 520 is configured to select a target algorithm from the list of privacy calculation algorithms. The message sending unit 530 is configured to send a feedback message to the initiator, where the feedback message includes the algorithm identifier of the target algorithm.

In a specific embodiment, the algorithm selection unit 520 is specifically configured to: select several privacy computing algorithms supported by itself from the list of privacy computing algorithms; Select the target algorithm from the several privacy calculation algorithms mentioned above.

In a specific embodiment, the algorithm selection unit 520 is specifically configured to: select several privacy computing algorithms supported by itself from the list of privacy computing algorithms; according to the first priority information of the several privacy computing algorithms, and the The participant selects the target algorithm from the plurality of privacy computing algorithms for the second priority information of the plurality of privacy computing algorithms.

In a more specific embodiment, the first priority information includes a first ranking, and the second priority information includes a second ranking; the algorithm selection unit 520 is specifically configured to: according to the first ranking of the several privacy calculation algorithms The first sorting and the second sorting determine the comprehensive ranking; the privacy calculation algorithm with the highest comprehensive ranking is determined as the target algorithm.

In one embodiment, the algorithm selection unit 520 is specifically configured to: select several privacy computing algorithms supported by itself from the list of privacy computing algorithms; level information, and select the target algorithm from the several privacy calculation algorithms.

In one embodiment, the apparatus 500 further includes: a task execution unit 540 configured to use the target algorithm to jointly execute the target computing task with the initiator.

In summary, using the device and system for algorithm negotiation for privacy computing disclosed in the embodiments of this specification, the initiator of privacy computing sends a negotiation request to the participant, and the negotiation request includes task information of the target computing task and the initiator policy for the target computing task. The list of supported privacy computing algorithms, so that the participants can select the target algorithm from the list of privacy computing algorithms based on their own demands, and feed back to the initiator, and then the two parties can negotiate based on the target algorithm with high feasibility and high availability. Work together to complete the target calculation task.

According to another embodiment, there is also provided a computer-readable storage medium on which a computer program is stored. When the computer program is executed in a computer, the computer is instructed to execute the method described in conjunction with FIG. 2 .

According to yet another embodiment, there is also provided a computing device, including a memory and a processor, wherein executable code is stored in the memory, and when the processor executes the executable code, the implementation described in conjunction with FIG. 2 is implemented. method.

Those skilled in the art should be aware that, in the above one or more examples, the functions described in the present invention may be implemented by hardware, software, firmware or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.

The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. Protection scope, any modification, equivalent replacement, improvement, etc. made on the basis of the technical solution of the present invention shall be included in the protection scope of the present invention.

Claims

A method for algorithm negotiation for private computing, comprising:

The initiator sends a negotiation request to the participant, and the negotiation request includes description information of the target computing task to be performed on the privacy calculation, and a list of privacy computing algorithms supported by the initiator for the target computing task;

The participant selects a target algorithm from the list of privacy calculation algorithms;

The participant sends a feedback message to the initiator, and the feedback message includes an algorithm identifier of the target algorithm.
The method according to claim 1, wherein the description information of the target computing task includes a task type of the target computing task and data information of participants involved in the target computing task.
The method according to claim 2, wherein the task type includes at least one of the following: joint statistical analysis of specified data items; joint training of specified models; joint prediction of specified models;

The participant data information includes a participant dataset identifier.
The method according to claim 1, wherein the negotiation request further includes the initiator's first priority information for each privacy computing algorithm in the privacy computing algorithm list.
The method according to claim 4, wherein the participant selects a target algorithm from the list of privacy calculation algorithms, comprising:

The participant selects several privacy computing algorithms supported by itself from the list of privacy computing algorithms;

Selecting the target algorithm from the several privacy computing algorithms according to the first priority information of the several privacy computing algorithms.
The method according to claim 4, wherein the participant selects a target algorithm from the list of privacy calculation algorithms, comprising:

The participant selects several privacy computing algorithms supported by itself from the list of privacy computing algorithms;

Selecting the target algorithm from the several privacy computing algorithms according to the first priority information of the several privacy computing algorithms and the second priority information of the participant for the several privacy computing algorithms.
The method according to claim 1, wherein the participant selects a target algorithm from the list of privacy calculation algorithms, comprising:

The participant selects several privacy computing algorithms supported by itself from the list of privacy computing algorithms;

Selecting the target algorithm from the several privacy computing algorithms according to the participant's second priority information for the several privacy computing algorithms.
The method according to claim 1, further comprising:

The initiator and the participant execute the target computing task by using the target algorithm.
A method for algorithm negotiation for private computing, executed by the initiator, including:

Sending a negotiation request to the participant, the negotiation request including description information of the target computing task to be performed with privacy calculation, and a list of privacy computing algorithms supported by the initiator for the target computing task;

A feedback message is received from the participant, the feedback message includes an algorithm identifier of a target algorithm, and the target algorithm is selected by the participant from the list of privacy calculation algorithms.
The method according to claim 9, further comprising: determining the list of privacy computing algorithms according to the target computing task.
The method according to claim 10, wherein, according to the target computing task, determining the list of privacy computing algorithms comprises: acquiring a first list of algorithms for executing the target computing task; Determine the privacy computing algorithm supported by the party, and form the privacy computing algorithm list.
The method according to claim 9, further comprising: determining first priority information for each privacy computing algorithm in the privacy computing algorithm list, and making the negotiation request include the first priority information.
The method according to claim 12, wherein determining the first priority information comprises:

The first priority information is determined according to the operation index of each privacy calculation algorithm, and the operation index includes at least one of the following: occupied computing resources, computing speed, communication cost, and privacy protection strength.
The method according to claim 9, further comprising: utilizing the target algorithm to jointly perform the target computing task with the participant.
A method for algorithm negotiation for private computing, performed by participating parties, including:

Receive a negotiation request sent by the initiator, where the negotiation request includes description information of the target computing task to be performed on the privacy calculation, and a list of privacy computing algorithms supported by the initiator for the target computing task;

Select a target algorithm from the list of privacy calculation algorithms;

Sending a feedback message to the initiator, where the feedback message includes the algorithm identifier of the target algorithm.
The method according to claim 15, wherein the negotiation request further includes the initiator's first priority information for each privacy computing algorithm in the privacy computing algorithm list.
The method of claim 16, wherein selecting a target algorithm comprises:

From the list of privacy computing algorithms, select several privacy computing algorithms supported by itself;

Selecting the target algorithm from the several privacy computing algorithms according to the first priority information of the several privacy computing algorithms.
The method of claim 16, wherein selecting a target algorithm comprises:

From the list of privacy computing algorithms, select several privacy computing algorithms supported by itself;

Selecting the target algorithm from the several privacy computing algorithms according to the first priority information of the several privacy computing algorithms and the second priority information of the participant for the several privacy computing algorithms.
The method of claim 18, wherein the first priority information includes a first ranking and the second priority information includes a second ranking;

The selecting the target algorithm from the several privacy calculation algorithms includes:

Determine the comprehensive ranking according to the first ranking and the second ranking of the several privacy calculation algorithms;

The privacy calculation algorithm with the highest comprehensive ranking is determined as the target algorithm.
The method of claim 15, wherein selecting a target algorithm comprises:

From the list of privacy computing algorithms, select several privacy computing algorithms supported by itself;

Selecting the target algorithm from the several privacy computing algorithms according to the participant's second priority information for the several privacy computing algorithms.
A system for algorithm negotiation for private computing, including:

The initiator is configured to send a negotiation request to the participant, the negotiation request includes description information of a target computing task to be performed with privacy computing, and a list of privacy computing algorithms supported by the initiator for the target computing task;

The participant is configured to select a target algorithm from the list of privacy calculation algorithms;

The participant is further configured to send a feedback message to the initiator, where the feedback message includes an algorithm identifier of the target algorithm.
A device for algorithm negotiation for private computing, which is integrated in the initiator, including:

A request sending unit configured to send a negotiation request to a participant, the negotiation request including description information of a target computing task to be performed with privacy computing, and a list of privacy computing algorithms supported by the initiator for the target computing task;

The message receiving unit is configured to receive a feedback message from the participant, the feedback message includes an algorithm identifier of a target algorithm, and the target algorithm is selected by the participant from the list of privacy calculation algorithms.
A device for algorithm negotiation for private computing, which is integrated with participants, including:

The request receiving unit is configured to receive a negotiation request sent by the initiator, the negotiation request including description information of a target computing task to be performed with privacy calculation, and a list of privacy computing algorithms supported by the initiator for the target computing task;

an algorithm selection unit configured to select a target algorithm from the list of privacy calculation algorithms;

A message sending unit configured to send a feedback message to the initiator, where the feedback message includes the algorithm identifier of the target algorithm.
A computer-readable storage medium, on which a computer program is stored, wherein, when the computer program is executed in a computer, the computer is caused to execute the method described in any one of claims 1-20.
A computing device, comprising a memory and a processor, wherein executable code is stored in the memory, and the method according to any one of claims 1-20 is implemented when the processor executes the executable code.