Computing power service method, computing power service acquisition method and electronic device

Abstract

The present application provides a computing power service method, a computing power service acquisition method and an electronic device. The computing power service method comprises: acquiring a training data set and model description data of a model training task initiated by a client (S310); according to the model description data, generating a zero-knowledge proof circuit, an encryption key and a decryption key corresponding thereto (S320); sending the training data set, the model description data, the zero-knowledge proof circuit and the encryption key to a computing power node, such that the computing power node performs training on the basis of the training data set and the model description data so as to obtain a first target model (S330); acquiring a test data set of the model training task, and sending the test data set to the computing power node, such that the computing power node utilizes the zero-knowledge proof circuit and the encryption key to process model data of the first target model and the test data set so as to obtain a corresponding zero-knowledge proof, a prediction result corresponding to the test data set and a first hash value corresponding to the first target model, the zero-knowledge proof being used for proving that the prediction result is output by the first target model on the basis of the test data set (S340); when the zero-knowledge proof passes verification, acquiring the model data of the first target model from the computing power node, the zero-knowledge proof being verified by means of the decryption key, the test data set, the prediction result and the first hash value (S350); and determining a second hash value on the basis of the model data, and when the first hash value is equal to the second hash value, sending the model data to the client (S360).

Description

Computing power service method, computing power service acquisition method, and electronic device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application is based on the Chinese patent application with application number 202310784045.1 and application date of June 28, 2023, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby introduced into this application as a reference. Technical Field

[0003] The present application relates to the field of machine learning technology, and in particular to a computing power service method, a computing power service acquisition method, and an electronic device. Background Art

[0004] The computing network is a computing platform based on distributed systems and blockchain technology. In current computing network systems, computing nodes typically complete machine learning training tasks. After transmitting the trained model data to the task initiator, the task initiator tests the model and evaluates the completion of the task.

[0005] However, in a peer-to-peer computing network, since the task initiator and the computing power node are usually equal and do not trust each other, the rights and interests of both parties to the transaction may be damaged. For example, when the computing power reward transaction occurs before the task initiator obtains the calculation results, the computing power node may not perform the calculation according to regulations. When the computing power reward transaction occurs after the task initiator obtains the calculation results, the task initiator may deliberately express dissatisfaction with the calculation results and refuse to pay the reward even if the computing power results meet the regulations.

[0006] Therefore, how to provide reliable algorithm services and protect the rights and interests of both parties in computing power transactions is a technical problem that needs to be urgently solved in this field.

[0007] Summary of the Invention

[0008] Embodiments of the present application provide a computing power service method, a computing power service acquisition method, and an electronic device.

[0009] In the first aspect, an embodiment of the present application provides a computing power service method, which is applied to an intermediate service platform, and the method includes: obtaining a training data set and model description data of a model training task initiated by a client; generating a corresponding zero-knowledge proof circuit, an encryption key, and a decryption key according to the model description data; sending the training data set, the model description data, the zero-knowledge proof circuit, and the encryption key to a computing power node, so that the computing power node obtains a first target model based on the training data set and the model description data, and determines a first hash value corresponding to the first target model; obtaining a test data set for the model training task, and sending the test data set to the computing power node, so that the computing power node uses the zero-knowledge proof circuit and The encryption key processes the model data of the first target model and the test data set to obtain a corresponding zero-knowledge proof, a prediction result corresponding to the test data set, and a first hash value corresponding to the first target model, wherein the zero-knowledge proof is used to prove that the prediction result is output by the first target model based on the test data set; if the zero-knowledge proof is verified, the model data of the first target model is obtained from the computing power node, wherein the zero-knowledge proof is verified by the decryption key, the test data set, the prediction result, and the first hash value; a second hash value is determined based on the model data, and if the first hash value is equal to the second hash value, the model data is sent to the client.

[0010] In the second aspect, an embodiment of the present application provides a computing power service method, which is applied to a computing power node, and the method includes: obtaining a training data set and model description data corresponding to a model training task from an intermediate service platform; training a first target model based on the training data set and the model description data; obtaining a zero-knowledge proof circuit, an encryption key and a test data set from the intermediate service platform; using the zero-knowledge proof circuit and the encryption key to process the model data of the first target model and the test data set to obtain corresponding zero-knowledge proof, a prediction result corresponding to the test data set and a first hash value corresponding to the first target model, the zero-knowledge proof being used to prove that the prediction result is output by the first target model based on the test data set; and sending the zero-knowledge proof, the prediction result and the first hash value to the intermediate service platform.

[0011] In a third aspect, an embodiment of the present application provides a method for obtaining a computing power service, which is applied to a client. The method includes: sending a training data set and model description data of a model training task to an intermediate service platform, so that the intermediate service platform generates a corresponding zero-knowledge proof circuit, an encryption key, and a decryption key according to the model description data; the intermediate service platform sends the training data set, the model description data, the zero-knowledge proof circuit, and the encryption key to a computing power node, so that the computing power node obtains a first target model through training according to the training data set and the model description data; sending a test data set of the model training task to the intermediate service platform, so that the intermediate service platform sends the test data set to the computing power node. The computing power node uses the zero-knowledge proof circuit and the encryption key to process the model data of the first target model and the test data set to obtain the corresponding zero-knowledge proof, the prediction result corresponding to the test data set and the first hash value corresponding to the first target model, wherein the zero-knowledge proof is used to prove that the prediction result is output by the first target model according to the test data set, and the zero-knowledge proof is verified by the decryption key, the test data set, the prediction result and the first hash value; the intermediate service platform receives the model data of the first target model obtained from the computing power node when the zero-knowledge proof passes the verification, and the second hash value of the received model data is equal to the first hash value.

[0012] In a fourth aspect, an embodiment of the present application provides an electronic device comprising: one or more processors; a memory on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the computing power service method as described in the first aspect, or the computing power service method as described in the second aspect, or the computing power service acquisition method as described in the third aspect.

[0013] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the computing power service method as described in the first aspect, or the computing power service method as described in the second aspect, or the computing power service acquisition method as described in the third aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to provide a further understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0015] FIG1 is an architectural diagram of an intermediate service platform and computing nodes provided by one embodiment of the present application;

[0016] FIG2 is a diagram showing a deployment structure of a computing service provided by one embodiment of the present application;

[0017] FIG3 is a flow chart of a computing power service method provided by one embodiment of the present application;

[0018] FIG4 is a schematic diagram of the sub-step flow of step S350 in FIG3 ;

[0019] FIG5 is a schematic diagram of the sub-step flow of step S320 in FIG3 ;

[0020] FIG6 is a flow chart of a computing power service method provided by an embodiment of the present application;

[0021] FIG7 is a flow chart of a computing power service method provided by one embodiment of the present application;

[0022] FIG8 is a flow chart of a method for obtaining computing power services provided in one embodiment of the present application;

[0023] FIG9 is a flow chart of a method for obtaining computing power services provided in one embodiment of the present application;

[0024] FIG10 is a schematic diagram of a flow chart of model prediction provided by one embodiment of the present application;

[0025] FIG11 is a schematic diagram of a process for providing computing power services by a computing power network system according to an embodiment of the present application;

[0026] FIG12 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable technical personnel in this field to better understand the technical solution of this application, the computing power service method, computing power service acquisition method, electronic device, and computer-readable storage medium provided by this application are described in detail below with reference to the accompanying drawings.

[0028] Example embodiments will be described more fully hereinafter with reference to the accompanying drawings, but the described example embodiments may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete and will fully convey the scope of this application to those skilled in the art.

[0029] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0030] The terms used herein are used only to describe specific embodiments and are not intended to limit this application. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It will also be understood that when the terms "comprising" and / or "made of" are used in this specification, they specify the presence of features, wholes, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or groups thereof.

[0031] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0032] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present application, and will not be interpreted as having an idealized or overly formal meaning, unless clearly defined in the examples of the present application.

[0033] The computing network is a computing platform based on distributed systems and blockchain technology. In current computing network systems, computing nodes typically complete machine learning training tasks. After transmitting the trained model data to the task initiator, the task initiator tests the model and evaluates the completion of the task.

[0034] However, in a peer-to-peer computing network, since the task initiator and the computing power node are usually equal and do not trust each other, the rights and interests of both parties to the transaction may be damaged. For example, when the computing power reward transaction occurs before the task initiator obtains the calculation results, the computing power node may not perform the calculation according to regulations. When the computing power reward transaction occurs after the task initiator obtains the calculation results, the task initiator may deliberately express dissatisfaction with the calculation results and refuse to pay the reward even if the computing power results meet the regulations.

[0035] Therefore, how to provide reliable algorithm services and protect the rights and interests of both parties in computing power transactions is a technical problem that needs to be urgently solved in this field.

[0036] Based on this, the embodiments of the present application provide a computing power service method, a computing power service acquisition method, and an electronic device, which can provide reliable algorithm services and protect the rights and interests of both parties to the computing power transaction.

[0037] Before introducing the technical solutions of the embodiments of the present application, the application scenarios of the embodiments of the present application will be described. Please refer to Figure 1, which is an architectural diagram of the intermediate service platform and computing power nodes provided by one embodiment of the present application. As shown in Figure 1, the intermediate service platform includes an application layer, a management layer, and a support layer. The application layer is responsible for computing power registration, computing power trading, computing power matching, and computing power verification. The management layer is responsible for task management, task scheduling, resource management, and model management. The support layer includes blockchain engines and zero-knowledge engines. The computing power nodes include access layers, device layers, and resource layers. The access layer includes a proxy gateway. The device layer includes machine learning, zero-knowledge proof, and other contents. The resource layer includes data resources, computing power resources, operator resources, and other contents. Please refer to Figure 2, which is a diagram of the computing power service deployment structure provided by one embodiment of the present application. As shown in Figure 2, in the computing power service deployment, the intermediate service platform is responsible for some of the computing content in the computing power service and manages and coordinates each computing power node. Each computing power node accepts management and coordination from the intermediate service platform through the proxy gateway, and completes the computing task through device management, computing equipment, resource access, and resource coordination.

[0038] Please refer to Figure 3, which is a flow chart of a computing power service method provided by one embodiment of the present application. The computing power service method is applied to an intermediate service platform. As shown in Figure 3, the computing power service method provided by the embodiment of the present application includes but is not limited to steps S310 to S360, which are described below.

[0039] Step S310: Obtain the training data set and model description data of the model training task initiated by the client.

[0040] In one embodiment, the model description data is Keras code. Keras is an advanced neural network application programming interface (API). By calling the API, the client can build various deep learning models, including fully connected, convolutional, recurrent and hybrid models.

[0041] Step S320: Generate corresponding zero-knowledge proof circuit, encryption key and decryption key according to the model description data.

[0042] It is understood that zero-knowledge proof is used to prove the correctness of a claim without revealing any information other than the correctness of the claim. The zero-knowledge proof circuit refers to the zero-knowledge proof circuit file used to describe the functional relationship of the model. In the embodiment of this application, the discussion of zero-knowledge proof is that the computing power node has a model Test dataset The prediction results obtained by the model are The hash result of the model is h, and the computing power node proves this statement to the client (ie the task initiator), but does not expose the model. The specific value of .

[0043] It should be noted that the encryption key is used in the generation process of zero-knowledge proof, and the decryption key is used in the verification process of zero-knowledge proof. The process and its role can be reflected in the subsequent steps.

[0044] In some embodiments, please refer to Figure 5, which shows a sub-step flow chart of step S320 in Figure 3. As shown in Figure 5, the corresponding zero-knowledge proof circuit, encryption key and decryption key are generated according to the model description data, including but not limited to steps S510 to S540.

[0045] Step S510: determining a model network structure according to the model description data, where the model network structure includes a plurality of model components.

[0046] Step S520: Generate a zero-knowledge proof circuit based on multiple model components.

[0047] Step S530: compile according to the zero-knowledge proof circuit to obtain a constraint system.

[0048] Step S540: Generate an encryption key and a decryption key according to the constraint system.

[0049] It should be understood that the model network structure including multiple model components is determined according to the model description data. The model components are calculation functions including operations such as pooling, activation functions, and intermediate layers. Then, the multiple model components are converted into a zero-knowledge proof circuit consisting of two parts. The first part of the zero-knowledge proof circuit is Among them, function F is the calculation function of the prediction process of the model, is the model input, i.e. the test dataset, is the model data, i.e. weights and biases, is the model output, i.e. the prediction result of the test data set; the second part of the zero-knowledge proof circuit is Among them, function H is a hash function, and h is the hash result of the model data.

[0050] In one embodiment, the model description data is keras code, and the intermediate service platform can export an H5 model based on the keras code uploaded by the client, where H5 refers to Hierarchical Data Format 5, a data file format for storing large-scale scientific data sets. Since keras can define the model network structure and obtain a model network structure including information such as an input layer, an output layer, and the number of neurons in the intermediate layer, this information is stored in the H5 model. The intermediate service platform can determine the model network structure by parsing the structural parameters of the H5 model.

[0051] It should be understood that after the intermediate service platform converts the model component into a zero-knowledge proof circuit, it compiles the zero-knowledge proof circuit to obtain a constraint system, namely, a rank-one constraint system used to describe a set of linear constraints. The rank-one constraint system is a formal language, including a set of variables, a set of constraints and a set of public parameters, where the variables are unknowns, i.e., solutions, the constraints are restrictions on the variables, and the public parameters are public values ​​used in the process of proving and verifying the zero-knowledge proof. The public parameters include the public input and public output of the zero-knowledge proof circuit, i.e., the test data set, the prediction results of the test data set and the hash results of the model data.

[0052] It is understandable that after the constraint system is obtained by compiling the zero-knowledge proof circuit, the constraint system of the zero-knowledge proof circuit is initialized with trust to obtain the encryption key and the decryption key.

[0053] Step S330: Send the training data set, model description data, zero-knowledge proof circuit, and encryption key to the computing power node, so that the computing power node obtains the first target model through training based on the training data set and model description data.

[0054] It should be understood that the intermediate service platform sends the training data set, model description data, and zero-knowledge proof circuit and encryption key uploaded by the client to the computing power node. After receiving the training data set, model description data, zero-knowledge proof circuit and encryption key, the computing power node trains the first target model based on the training data set and model description data.

[0055] Step S340: obtain a test data set for the model training task, and send the test data set to the computing power node, so that the computing power node uses the zero-knowledge proof circuit and the encryption key to process the model data and the test data set of the first target model to obtain the corresponding zero-knowledge proof, the prediction result corresponding to the test data set, and the first hash value corresponding to the first target model. The zero-knowledge proof is used to prove that the prediction result is output by the first target model based on the test data set.

[0056] It can be understood that after the computing power node obtains the first target model based on the training data set and model description data, the client can request to test the first target model to evaluate the quality of the model. The intermediate service platform obtains the test data set of the model training task uploaded by the client and sends the test data set to the computing power node. The computing power node uses the zero-knowledge proof circuit and the encryption key to process the model data and the test data set of the first target model to obtain the corresponding zero-knowledge proof and public output data. The public output data includes the prediction result corresponding to the test data set and the first hash value corresponding to the first target model. The zero-knowledge proof can be used to prove that the prediction result is output by the first target model based on the test data set, and the model data used during the test is consistent with the model data obtained through training.

[0057] In one embodiment, the intermediate service platform compiles a constraint system based on a zero-knowledge proof circuit and performs trusted initialization on the constraint system to obtain corresponding encryption and decryption keys. The intermediate service platform sends the constraint system and encryption key corresponding to the zero-knowledge proof circuit to the computing power node, which receives the constraint system and encryption key sent by the intermediate service platform. According to the zero-knowledge proof protocol, the input data for generating the proof includes two parts: private input data and public input data. The private input data is the model data of the trained first target model, and the public input data is the test data set. Combining the constraint system and the encryption key can generate a proof document and public output data. The public output data includes the prediction result of the test data set and the first hash value of the first target model. In other words, the computing power platform inputs the first target model, the prediction result, and the encryption key into the zero-knowledge proof circuit to obtain the corresponding zero-knowledge proof and the first hash value. The zero-knowledge proof is used to prove that the prediction result is output by the first target model based on the test data set. The first hash value can also be used to verify whether the model data used by the computing power node during testing is consistent with the model data obtained during training.

[0058] In one embodiment, the training data set, the model description data, the zero-knowledge proof circuit, and the encryption key are sent to the computing power node, so that the computing power node trains the first target model according to the training data set and the model description data, further comprising:

[0059] The training data set, model description data, zero-knowledge proof circuit, and encryption key are sent to the computing power node, so that the computing power node obtains the first target model through training according to the training data set and the model description data, and determines the third hash value corresponding to the first target model.

[0060] Obtain the third hash value from the computing power node and upload the third hash value to the blockchain.

[0061] It is understood that after the computing node trains the first target model based on the training dataset and model description data, it performs a hash operation on the model data of the first target model to obtain a third hash value corresponding to the first target model. The computing node then sends the third hash value to the intermediary service platform, which then uploads the third hash value to the blockchain. As the verifier of the zero-knowledge proof, the smart contract on the blockchain can be invoked to verify, using the hash value of the model data, whether the model data used by the computing node during testing is consistent with the model data obtained during training. Specifically, when the first hash value and the third hash value are equal, the model data used by the computing node during testing is consistent with the model data obtained during training.

[0062] Step S350: When the zero-knowledge proof is verified, the model data of the first target model is obtained from the computing power node, wherein the zero-knowledge proof is verified by the decryption key, the test data set, the prediction result and the first hash value.

[0063] It should be understood that, based on the above discussion of zero-knowledge proof, zero-knowledge proof can prove to the client by verifying that the computing power node is able to prove to the client that the prediction result is obtained by the first target model trained by the computing power node on the test data set, and the model data used in the test is consistent with the trained model data. The client can make predictions on the first target model without knowing the model data, thereby testing the performance of the first target model. This can avoid the situation where the client deliberately expresses dissatisfaction with the calculation results and refuses to support remuneration after obtaining model data that meets the requirements.

[0064] In some embodiments, at least one of the intermediate service platform and the client acts as a verifier of the zero-knowledge proof and verifies the zero-knowledge proof based on the decryption key, the test data set, the prediction result and the first hash value.

[0065] It is understandable that both the intermediate service platform and the client can serve as the verifier of the zero-knowledge proof to verify the zero-knowledge proof. When the intermediate service platform serves as the verifier of the zero-knowledge proof, the intermediate service platform receives the zero-knowledge proof, prediction result, and first hash value sent by the computing power node, and verifies the zero-knowledge proof based on the decryption key, test data set, prediction result, and first hash value. If the verification result is a passed verification result, the intermediate service platform then obtains the model data of the first target model from the computing power node; when the client serves as the verifier of the zero-knowledge proof, the intermediate service platform sends the zero-knowledge proof, decryption key, test data set, prediction result, and first hash value to the client, so that the client verifies the zero-knowledge proof based on the decryption key, test data set, prediction result, and first hash value. After that, the intermediate service platform receives the verification result sent by the client. If the verification result is a passed verification result, the intermediate service platform then obtains the model data of the first target model from the computing power node.

[0066] It should be understood that the zero-knowledge proof is generated by the computing power node based on the first target model, the prediction result and the encryption key, and the verifier verifies the zero-knowledge proof based on the decryption key, the test data set, the prediction result and the first hash value to determine that the preset result is obtained by predicting the test data set by the first target model trained by the computing node, and to determine that the model data used in the test is consistent with the model data obtained by training, so that the computing power node can allow the client to test the first target model without uploading the model data of the first target model.

[0067] In one embodiment, the intermediate service platform compiles the constraint system according to the zero-knowledge proof circuit, and performs trusted initialization on the constraint system to obtain the corresponding encryption key and decryption key. The zero-knowledge proof verifier verifies the zero-knowledge proof according to the zero-knowledge proof protocol. The input of the verification process includes a public value and a certification document obtained by the computing power node in combination with the constraint system and the encryption key. The public value includes a test training data set, a prediction result of the test data set, and a hash value of the first target model. The verification result of the zero-knowledge proof can be obtained by combining the constraint system and the decryption key, that is, the verification passed or failed.

[0068] In some embodiments, the test dataset includes multiple test samples and sample labels corresponding to each test sample, and there are multiple prediction results, each corresponding to a test sample. Referring to FIG. 4 , FIG. 4 illustrates a schematic flow diagram of the sub-steps of step S350 in FIG. 3 . As shown in FIG. 4 , if the zero-knowledge proof is verified, model data for the first target model is obtained from the computing power node, including but not limited to steps S410 to S430.

[0069] Step S410: Receive a prediction accuracy threshold for a model training task from a client.

[0070] Step S420: Determine the prediction accuracy based on each prediction result and the sample label corresponding to the prediction result.

[0071] Step S430: When the prediction accuracy is greater than or equal to the prediction accuracy threshold and the verification result is a verification pass result, the model data of the first target model is obtained from the computing power node.

[0072] It should be understood that for each test sample in the test data set, the relationship between the prediction accuracy of the first target model and the prediction accuracy threshold is determined based on the sample label corresponding to the test sample and the prediction accuracy threshold of the model training task uploaded by the client. When the prediction accuracy is greater than or equal to the prediction accuracy threshold, and when the verification result is a passed verification result, it indicates that the prediction result is obtained by predicting the test data set by the first target model trained by the computing power node, the model data used in the test is consistent with the model data obtained by training, and the first target model meets the accuracy requirements. Therefore, the intermediate service platform can obtain the model data of the first target model from the computing power node to deliver the calculation results to the client.

[0073] In one embodiment, the algorithm service method further includes:

[0074] When the prediction accuracy is less than the prediction accuracy threshold or the verification result is that the verification fails, a retraining instruction is sent to the computing power node so that the computing power node retrains the first target model based on the training data set and the model description data according to the retraining instruction.

[0075] It can be understood that the prediction accuracy is less than the prediction accuracy threshold, which indicates that the performance of the first target model has not met the requirements of the task initiator, and the zero-knowledge proof has not passed the verification, which indicates that the computing power node cannot prove to the client that the prediction result is obtained by predicting the test data set by the first target model trained by the computing power node, or that the model data used in the test is consistent with the model data obtained by training. In both cases, the computing power node needs to retrain the model and re-perform the prediction and proof process until the prediction accuracy is greater than or equal to the prediction accuracy threshold and the zero-knowledge proof passes the verification.

[0076] In some embodiments, the computing power service method further includes: uploading at least one of the first hash value, the test data set, and the prediction accuracy to the blockchain.

[0077] It can be understood that for the first hash value, the intermediate service platform can obtain the first hash value corresponding to the first target model from the computing power node after the computing power node trains the first target model and upload the first hash value to the blockchain for evidence storage. The hash operation can avoid exposing the model data of the first target model and ensure the security of the first target model. By uploading the first hash value to the blockchain for evidence storage, it can be ensured that the hash value of the current model data of the first target model cannot be tampered with, so that subsequent model data cannot be arbitrarily changed, thereby protecting the rights and interests of both parties to the computing power transaction; for the test data set, the intermediate service platform can, after receiving the test data set of the model training task sent by the client , the test data set is uploaded to the blockchain for archiving to prevent the test data set from being tampered with, and the computing power node is prevented from tampering with the test data set so that the first target model that does not meet the requirements can obtain a prediction result that meets the requirements of the task initiator, so that the testing process of the first target model is reliable and trustworthy; for the prediction accuracy threshold, the intermediate service platform can upload the prediction accuracy threshold to the blockchain for archiving after receiving the prediction accuracy threshold of the model training task uploaded by the client, so as to prevent the prediction accuracy threshold from being tampered with, and the computing power node is prevented from tampering with the prediction accuracy threshold so that the first target model that does not meet the requirements can pass the model test initiated by the client, so that the testing process of the first target model is reliable and trustworthy.

[0078] Step S360: Determine a second hash value based on the model data, and send the model data to the client if the first hash value is equal to the second hash value.

[0079] It can be understood that before the computing power node delivers the calculation results to the client, the intermediate service platform determines the corresponding second hash value based on the model data sent by the computing node, and compares the first hash value and the second hash value. When the first hash value and the second hash value are equal, it indicates that the model data submitted by the computing power node is consistent with the model data used during the test, thereby avoiding the inconsistency between the model data received by the client and the model data used during the test.

[0080] In one embodiment, the algorithm service method further includes:

[0081] When the first hash value and the second hash value are not equal, a retraining instruction is sent to the computing power node.

[0082] It can be understood that when the first hash value and the second hash value are not equal, it represents that the model data submitted by the computing power node is inconsistent with the model data used during the test. By sending a retraining instruction to the computing power node, the computing power node is required to retrain the model and re-perform the prediction and proof process until the first hash value and the second hash value are equal.

[0083] The following describes the zero-knowledge proof process of model prediction in the algorithm service method through an embodiment.

[0084] Please refer to Figure 10, which shows a schematic diagram of the model prediction process provided by one embodiment of the present application. As shown in Figure 10, the intermediate service platform, as a computing power network platform, generates a zero-knowledge proof circuit for the model prediction based on the model description data uploaded by the client as the task initiator. The zero-knowledge proof circuit is then compiled to obtain a rank-one constraint system, and the rank-one constraint system is trusted initialized to obtain an encryption key and a decryption key. The encryption key is used in the zero-knowledge proof generation process, and the decryption key is used in the zero-knowledge proof verification process. The computing power node obtains the constraint system and encryption key sent by the intermediate service platform. According to the zero-knowledge proof protocol, the model data of the first target model and the test dataset are used as inputs to generate the proof. The constraint system and encryption key are combined to obtain the proof document, the prediction result of the test dataset, and the hash value of the first target model. The client, as the task initiator and also the verifier of this zero-knowledge proof, obtains the constraint system and decryption key sent by the intermediate service platform. According to the zero-knowledge proof protocol, the proof document, the test dataset, the prediction result of the test dataset, and the hash value of the first target model are used as inputs to verify the proof. The constraint system and decryption key are combined to obtain the verification result of the zero-knowledge proof.

[0085] An embodiment of the present application provides a computing power service method. On the one hand, an intermediate service platform generates a corresponding zero-knowledge proof circuit, an encryption key, and a decryption key based on model description data. After the computing power platform trains and obtains a first target model, it generates a zero-knowledge proof, a prediction result corresponding to the test data set, and a first hash value corresponding to the first target model based on the first target model, a test data set, an encryption key, and a zero-knowledge proof circuit. By verifying the zero-knowledge proof to ensure that the prediction result is output by the first target model based on the test data set, the client can test the first target model without knowing the model data. After completing the model training task, the computing power node does not need to transmit the model data of the trained first target model to the client, thereby preventing the client from deliberately expressing dissatisfaction with the calculation results and refusing to pay remuneration when obtaining model data that meets the requirements. On the other hand, by verifying the zero-knowledge proof to prove that the model data used by the computing power node during testing is the same as the model data obtained during training, and before delivering the model data to the client, the intermediate service platform verifies the model data again through the hash value, thereby avoiding the model data used by the computing node during testing being inconsistent with the model data obtained during training, and the model data received by the client being inconsistent with the model data used during testing, thereby protecting the rights and interests of both parties to the transaction in the computing power service and providing reliable computing power services.

[0086] Please refer to Figure 6, which is a flow chart of a computing power service method provided by an embodiment of the present application. The computing power service method is applied to a computing power node. As shown in Figure 6, the computing power service method provided by the embodiment of the present application includes but is not limited to steps S610 to S650.

[0087] Step S610: Obtain the training data set and model description data corresponding to the model training task from the intermediate service platform;

[0088] Step S620, training a first target model based on the training data set and the model description data;

[0089] Step S630, obtaining a zero-knowledge proof circuit, an encryption key, and a test data set from an intermediate service platform;

[0090] Step S640: Processing the model data of the first target model and the test data set using a zero-knowledge proof circuit and an encryption key to obtain a corresponding zero-knowledge proof, a prediction result corresponding to the test data set, and a first hash value corresponding to the first target model. The zero-knowledge proof is used to prove that the prediction result is output by the first target model based on the test data set.

[0091] Step S650: Send the zero-knowledge proof, prediction result and first hash value to the intermediate service platform.

[0092] It should be understood that the embodiments of the present application provide a computing power service method applied to a computing power node. For descriptions of steps, processes, effects, etc., please refer to the relevant descriptions in the computing power service method applied to an intermediate service platform provided in the above embodiments, which will not be repeated here.

[0093] In some embodiments, referring to FIG7 , FIG7 is a flow chart of a computing power service method provided by an embodiment of the present application. As shown in FIG7 , after the zero-knowledge proof, the prediction result, and the first hash value are sent to the intermediate service platform, the computing power service method further includes steps S710 and S720.

[0094] Step S710: receiving a retraining instruction sent by the intermediate service platform.

[0095] Step S720: retrain the first target model based on the training data set and the model description data according to the retraining instruction.

[0096] It is understandable that after the zero-knowledge proof, the prediction result and the first hash value are sent to the intermediate service platform, the intermediate service platform determines the prediction accuracy based on the prediction result and the sample label corresponding to the prediction result. If the prediction accuracy is less than the prediction accuracy threshold, or the zero-knowledge proof fails to pass the verification, the intermediate service platform will send a retraining instruction to the computing power node. After the computing node receives the retraining instruction sent by the intermediate service platform, it will retrain the first target model based on the training data set and the model description data according to the retraining instruction.

[0097] Please refer to Figure 8, which is a flowchart of a method for obtaining computing power services provided by an embodiment of the present application. The computing power service acquisition method is applied to the client. As shown in Figure 8, the computing power service acquisition method provided by the embodiment of the present application includes but is not limited to steps S810 to S830.

[0098] Step S810: Send the training data set and model description data of the model training task to the intermediate service platform, so that the intermediate service platform generates the corresponding zero-knowledge proof circuit, encryption key and decryption key according to the model description data. The intermediate service platform sends the training data set, model description data, zero-knowledge proof circuit, and encryption key to the computing power node, so that the computing power node obtains the first target model through training according to the training data set and model description data.

[0099] Step S820: Send the test data set of the model training task to the intermediate service platform, so that the intermediate service platform sends the test data set to the computing power node, so that the computing power node uses the zero-knowledge proof circuit and the encryption key to process the model data and the test data set of the first target model to obtain the corresponding zero-knowledge proof, the prediction result corresponding to the test data set and the first hash value corresponding to the first target model. The zero-knowledge proof is used to prove that the prediction result is output by the first target model according to the test data set, and the zero-knowledge proof is verified by the decryption key, the test data set, the prediction result and the first hash value.

[0100] Step S830: receiving the model data of the first target model obtained by the intermediate service platform from the computing power node when the zero-knowledge proof is verified, and the second hash value of the received model data is equal to the first hash value.

[0101] It should be understood that the embodiment of the present application provides a method for obtaining computing power services applied to a client. For descriptions of steps, processes, effects, etc., please refer to the relevant descriptions in the computing power service method applied to an intermediate service platform provided in the above embodiment, which will not be repeated here.

[0102] In one embodiment, the computing power service acquisition method also includes: sending a prediction accuracy threshold of the model training task to the intermediate service platform, and the intermediate service platform obtains the model data of the first target model from the computing power node when the prediction accuracy is greater than or equal to the prediction accuracy threshold and the zero-knowledge proof passes verification.

[0103] It can be understood that after the computing power node is trained to obtain the first target model, the client sends the prediction accuracy threshold of the model training task to the intermediate service platform, so that the intermediate service platform determines the prediction accuracy based on the prediction result and the sample label corresponding to the prediction result, and further judges the relationship between the prediction accuracy and the prediction accuracy threshold, so that the intermediate service platform obtains the model data of the first target model from the computing power node when the prediction accuracy is greater than or equal to the prediction accuracy threshold and the zero-knowledge proof is verified, thereby ensuring that the client can obtain the model data of the first target model that satisfies the prediction accuracy greater than or equal to the prediction accuracy threshold.

[0104] In one embodiment, referring to FIG9 , FIG9 shows a flow chart of a method for obtaining computing power services provided in an embodiment of the present application. As shown in FIG9 , the method for obtaining computing power services also includes step S910 and step S920.

[0105] Step S910: Receive the zero-knowledge proof, decryption key, test data set, prediction result and first hash value sent by the intermediate service platform.

[0106] Step S910: Verify the zero-knowledge proof based on the decryption key, test data set, prediction result and first hash value, and send the verification result to the intermediate service platform. If the verification result is a passed result, the intermediate service platform obtains the model data of the first target model from the computing power node.

[0107] It can be understood that the client receives the zero-knowledge proof, decryption key, test data set, prediction result and first hash value sent by the intermediate service platform, and verifies the zero-knowledge proof based on the decryption key, test data set, prediction result and first hash value to determine that the prediction result is output by the computing power node based on the first target model obtained through training based on the test data set, and the hash value of the first target model is the first hash value.

[0108] The following describes the process of the computing power network system in the embodiment of the present application through an example.

[0109] Referring to FIG11 , FIG11 shows a schematic diagram of a process for providing computing power services by a computing power network system provided in one embodiment of the present application. As shown in FIG11 , the computing power network system includes a client, an intermediate service platform, and computing power nodes. The overall process for providing computing power services by the computing power network system is as follows:

[0110] Step S1010: The client uploads the training data set, model description data, and prediction accuracy threshold to the intermediate service platform.

[0111] Step S1011: The intermediate service platform generates a zero-knowledge proof circuit for the model prediction calculation based on the model description data, compiles the zero-knowledge proof circuit to obtain the constraint system of the zero-knowledge proof circuit in the zero-knowledge proof, performs trusted initialization on the constraint system to obtain the encryption key and decryption key, and stores the prediction accuracy threshold on the chain.

[0112] Step S1012: The intermediate service platform sends the model description data, training data set, zero-knowledge proof circuit of model prediction, and encryption key to the computing power node.

[0113] Step S1013: The intermediate service platform sends the decryption key to the client.

[0114] Step S1014: The computing power node performs model training based on the training data set.

[0115] Step S1015: Hash the trained model data and record it on the blockchain.

[0116] Step S1016: Notify the client that the computing task has been completed.

[0117] Step S1017: The client sends the test data set (test samples and sample labels) to the intermediate service platform.

[0118] Step S1018: The intermediate service platform stores the test data set on the chain.

[0119] Step S1019: The intermediate service platform sends the test data set (test sample) to the computing power node.

[0120] Step S1020: The computing node uses the trained model to predict the test dataset and generates a related zero-knowledge proof. This proof consists of two parts: the first part is the prediction result obtained by calculating the trained model parameters and the values ​​of the test dataset; the second part is the hash value obtained by hashing the trained model data.

[0121] Step S1021: The computing power node sends the test data set, prediction results, hash value of the model data and zero-knowledge proof to the intermediate service platform.

[0122] Step S1022: The intermediary service platform calls the smart contract on the blockchain to verify the zero-knowledge proof and detects whether the accuracy of the prediction result is greater than the prediction accuracy threshold.

[0123] Step S1023: If the verification is passed in step S1022, the intermediate service platform notifies the computing power node to send the model data to the intermediate service platform.

[0124] Step S1024: If the verification fails in step S1022, the intermediate service platform requires the computing power node to retrain the model and returns to step S1014.

[0125] Step S1025: The computing power node sends the model data to the intermediate service platform.

[0126] Step S1026: The intermediary service platform calls the smart contract on the blockchain to verify the hash value of the model data.

[0127] Step S1027: If the verification is passed in step S1026, the intermediate service platform sends the model data to the client.

[0128] Step S1028: If the verification fails in step S1026, the intermediate service platform requires the computing power node to retrain the model and returns to step S1014.

[0129] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the existing situation can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the method described in each embodiment of the present application.

[0130] Referring to FIG. 12 , an embodiment of the present application further provides an electronic device, the electronic device 100 including but not limited to:

[0131] at least one processor 110;

[0132] at least one memory 120, for storing at least one program;

[0133] When at least one program is executed by at least one processor 110, the computing power service method or computing power service acquisition method described in any of the above embodiments is executed.

[0134] It should be understood that the processor 110 and the memory 120 may be connected via a bus or other means.

[0135] It should be understood that the processor 110 may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. Alternatively, the processor 110 may be one or more integrated circuits for executing relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0136] Memory 120, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs, such as the bandwidth adjustment method executed on the electronic device side described in any embodiment of this application. Processor 110 implements the above-mentioned computing power service method or computing power service acquisition method by executing the non-transitory software program and instructions stored in memory 120.

[0137] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store and execute the above-mentioned bandwidth adjustment method. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may include a memory remotely located relative to the processor 110, and these remote memories may be connected to the processor 110 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0138] The non-transient software programs and instructions required to implement the above-mentioned computing power service method or computing power service acquisition method are stored in the memory 120. When executed by one or more processors 110, the computing power service method or computing power service acquisition method provided by any embodiment of the present application is executed.

[0139] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the computing power service method or computing power service acquisition method described in any of the above embodiments.

[0140] The embodiments of the present application provide a computing power service method, a computing power service acquisition method, and an electronic device. On the one hand, the intermediate service platform generates a corresponding zero-knowledge proof circuit, an encryption key, and a decryption key according to the model description data. After the computing power platform trains and obtains the first target model, it generates a zero-knowledge proof, a prediction result corresponding to the test data set, and a first hash value corresponding to the first target model according to the first target model, a test data set, an encryption key, and a zero-knowledge proof circuit. By verifying the zero-knowledge proof to ensure that the prediction result is output by the first target model according to the test data set, the client can test the first target model without knowing the model data. The computing power node completes the model training task. After that, there is no need to transmit the trained model data of the first target model to the client, so as to avoid the client deliberately expressing dissatisfaction with the calculation results and refusing to pay remuneration when obtaining model data that meets the requirements; on the other hand, by verifying the zero-knowledge proof to prove that the model data used in the computing power node test is the same as the model data obtained by training, and before delivering the model data to the client, the intermediate service platform verifies the model data again through the hash value, so as to avoid the inconsistency between the model data used by the computing node in the test and the model data obtained by training, and the inconsistency between the model data received by the client and the model data used in the test, thereby protecting the rights and interests of both parties in the transaction in the computing power service and providing reliable computing power services.

[0141] The computer storage medium of the embodiment of the present application can adopt any combination of one or more computer-readable media.Computer-readable media can be computer-readable signal media or computer-readable storage media.Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination thereof.Examples (non-exhaustive) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination thereof.In this document, computer-readable storage media can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0142] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0143] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0144] The computer program code for performing the operations of the present application can be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (such as through the Internet using an Internet service provider).

[0145] An embodiment of the present application further provides a computer program product, which stores program instructions. When the program instructions are executed on a computer device, the computer device implements the bandwidth adjustment method described in any of the above embodiments.

[0146] The above is an explanation of some implementations of the present application, but the present application is not limited to the above implementation methods. Technical personnel familiar with the art can also make various equivalent modifications or substitutions without violating the shared conditions of the scope of the present application. These equivalent modifications or substitutions are all included in the scope defined by the present application.

Claims

1. A computing power service method, applied to an intermediate service platform, comprising: Obtain the training data set and model description data of the model training task initiated by the client; Generate a corresponding zero-knowledge proof circuit, encryption key and decryption key according to the model description data; Sending the training data set, the model description data, the zero-knowledge proof circuit, and the encryption key to a computing power node, so that the computing power node obtains a first target model through training according to the training data set and the model description data; Acquire a test data set of the model training task, and send the test data set to the computing power node, so that the computing power node uses the zero-knowledge proof circuit and the encryption key to process the model data of the first target model and the test data set to obtain the corresponding zero-knowledge proof, the prediction result corresponding to the test data set, and the first hash value corresponding to the first target model, wherein the zero-knowledge proof is used to prove that the prediction result is output by the first target model according to the test data set; When the zero-knowledge proof is verified, obtaining model data of the first target model from the computing power node, wherein the zero-knowledge proof is verified by the decryption key, the test data set, the prediction result, and the first hash value; A second Hash value is determined according to the model data, and when the first Hash value is equal to the second Hash value, the model data is sent to the client.

2. The method according to claim 1, wherein: The test data set includes multiple test samples and sample labels corresponding to each of the test samples, and there are multiple prediction results, each of which corresponds to one of the test samples; When the zero-knowledge proof is verified, obtaining the model data of the first target model from the computing power node includes: receiving a prediction accuracy threshold of a model training task from the client; Determining prediction accuracy according to each of the prediction results and the sample labels corresponding to the prediction results; When the prediction accuracy is greater than or equal to the prediction accuracy threshold and the zero-knowledge proof passes verification, the model data of the first target model is obtained from the computing power node.

3. The method according to claim 2, wherein: The method also includes: uploading at least one of the first hash value, the test data set, and the prediction accuracy to a blockchain.

4. The method according to claim 1, wherein: The generating of the corresponding zero-knowledge proof circuit, encryption key and decryption key according to the model description data includes: Determine a model network structure according to the model description data, wherein the model network structure includes a plurality of model components; generating the zero-knowledge proof circuit according to the plurality of model components; Compile according to the zero-knowledge proof circuit to obtain a constraint system; The encryption key and the decryption key are generated according to the constraint system.

5. The method according to claim 1, wherein: At least one of the intermediate service platform and the client acts as a verifier of the zero-knowledge proof and verifies the zero-knowledge proof according to the decryption key, the test data set, the prediction result and the first hash value.

6. A computing power service method, applied to a computing power node, the method comprising: Obtain the training data set and model description data corresponding to the model training task from the intermediate service platform; Train the first target model according to the training data set and the model description data; Obtaining a zero-knowledge proof circuit, an encryption key, and a test data set from the intermediate service platform; Processing the model data of the first target model and the test data set by using the zero-knowledge proof circuit and the encryption key to obtain a corresponding zero-knowledge proof, a prediction result corresponding to the test data set, and a first hash value corresponding to the first target model, wherein the zero-knowledge proof is used to prove that the prediction result is output by the first target model according to the test data set; The zero-knowledge proof, the prediction result and the first hash value are sent to the intermediate service platform.

7. The method according to claim 6, wherein: After sending the zero-knowledge proof, the prediction result and the first hash value to the intermediate service platform, the method further includes: Receiving a retraining instruction sent by the intermediate service platform; The first target model is retrained based on the training data set and the model description data according to the retraining instruction.

8. A method for obtaining computing power services, applied to a client, the method comprising: Sending a training data set and model description data of a model training task to an intermediate service platform, so that the intermediate service platform generates a corresponding zero-knowledge proof circuit, an encryption key, and a decryption key according to the model description data; the intermediate service platform sends the training data set, the model description data, the zero-knowledge proof circuit, and the encryption key to a computing power node, so that the computing power node obtains a first target model through training according to the training data set and the model description data; Sending a test data set of the model training task to the intermediate service platform, so that the intermediate service platform sends the test data set to the computing power node, so that the computing power node uses the zero-knowledge proof circuit and the encryption key to process the model data of the first target model and the test data set to obtain the corresponding zero-knowledge proof, the prediction result corresponding to the test data set, and the first hash value corresponding to the first target model, the zero-knowledge proof is used to prove that the prediction result is output by the first target model according to the test data set, and the zero-knowledge proof is verified by the decryption key, the test data set, the prediction result and the first hash value; Receive the model data of the first target model obtained by the intermediate service platform from the computing power node when the zero-knowledge proof is verified, and the second hash value of the received model data is equal to the first hash value.

9. The method according to claim 8, wherein: The method also includes: sending a prediction accuracy threshold of the model training task to the intermediate service platform, and the intermediate service platform obtains the model data of the first target model from the computing power node when the prediction accuracy is greater than or equal to the prediction accuracy threshold and the zero-knowledge proof passes verification.

10. The method according to claim 8, wherein: The method further comprises: Receiving the zero-knowledge proof, the decryption key, the test data set, the prediction result, and the first hash value sent by the intermediate service platform; The zero-knowledge proof is verified according to the decryption key, the test data set, the prediction result and the first hash value, and the verification result is sent to the intermediate service platform. If the verification result is a passed verification result, the intermediate service platform obtains the model data of the first target model from the computing power node.

11. An electronic device, comprising: one or more processors; A memory having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement: The computing power service method according to any one of claims 1 to 5; or The computing power service method according to any one of claims 6 to 7; or A method for obtaining computing power services as described in any one of claims 8 to 10.

12. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the program implements: The computing power service method according to any one of claims 1 to 5; or The computing power service method according to any one of claims 6 to 7; or A method for obtaining computing power services as described in any one of claims 8 to 10.

Images