Consistency data set generation method and device, equipment and storage medium

Abstract

The invention discloses a consistency data set generation method and device, equipment and a storage medium, relates to the technical field of artificial intelligence, and aims to generate a result data set with consistency by using the same problem data set through at least one neural network, so that the tampering risk caused by direct transmission of the result data set can be avoided, and the user experience is improved. And the security of generating the consistent data set is ensured. The method comprises the following steps: selecting any neural network from at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to a problem data set; selecting at least one node in the system and / or the network, so that a neural network deployed on the at least one node generates at least one second result data set according to the problem data set; and performing target operation by taking the first result data set and the at least one second result data set as a consistency data set.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for generating a consistent data set. Background Art

[0002] A consistent data set refers to a collection of data that maintains uniformity and reliability in multiple dimensions to ensure that the data meets predetermined constraints in different scenarios or tasks. It is usually used in systems and / or networks. With the expansion of application scenarios and the deepening of technical requirements, consistent data sets are an important feature that cannot be ignored in database systems, which directly affects the accuracy, completeness and reliability of data.

[0003] In related technologies, the generation methods of consistent data sets in different fields are different. In the medical field, consistent data sets can be obtained through sharing, for example, different hospitals share patients' diagnostic data for joint analysis; in the financial field, consistent data sets can be obtained through synchronization, for example, multiple data centers synchronize transaction information in real time to ensure the consistency of ledgers. However, data sets obtained through sharing or synchronization require direct data transmission, which increases the risk of data tampering and makes it difficult to ensure data security. Summary of the invention

[0004] In view of this, the present application provides a method, device, equipment and storage medium for generating a consistent data set, the main purpose of which is to solve the problem in the prior art that directly transmitting data increases the risk of data tampering and makes it difficult to ensure data security.

[0005] According to a first aspect of the present application, a method for generating a consistent data set is provided, comprising: Selecting any one of the at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to the problem data set, wherein the at least two neural networks have the same structural parameters and are respectively deployed on different nodes in the system and / or network, and the system and / or network is at least one; Selecting at least one node in the system and / or network so that a neural network deployed on the at least one node generates at least one second result data set based on the problem data set; The first result data set and the at least one second result data set are used as consistent data sets to perform a target operation.

[0006] Furthermore, each neural network has a preset local data set set, and the selecting at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set includes: If the problem data set is not initially stored in the local data set set of the neural network, the problem data set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; or the first storage identifier of the problem data set in the remote data set set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node requests the verification value of the problem data set and the result data set from the remote data set set through the first storage identifier, and generates at least one second result data set according to the problem data set; If the problem data set is initially stored in a local data set set of a neural network, a second storage identifier of the problem data set in the local data set set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node obtains the problem data set in the local data set set through the second storage identifier, and generates at least one second result data set based on the problem data set. The second storage identifier is used to point to the same problem data set in local data set sets of different neural networks.

[0007] Furthermore, before performing a target operation on the first result data set and the at least one second result data set as consistent data sets, the method further includes: Transmitting the verification value of the problem data set and the first result data set to at least one node in the system and / or network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; comparing the check value of the first result data set with the check value of the at least one second result data set; Correspondingly, if the comparison results are consistent, the first result data set and the at least one second result data set are used as consistent data sets to perform target operations.

[0008] Furthermore, if the comparison result is inconsistent, the first result data set and the at least one second result data set are treated as inconsistent data sets for troubleshooting.

[0009] Furthermore, before comparing the check value of the first result data set with the check value of the at least one second result data set, the method further includes: Calculating a check value of the first result data set to uniquely identify a feature of the first result data set through the check value; and A check value of the second result data set is calculated to uniquely identify a feature of the second result data set through the check value.

[0010] Further, the calculating the check value of the first result data set includes: Converting the first result data set into a string of fixed length using a hash function to obtain a check value of the first result data set; or Performing feature extraction on the first result data set to obtain a verification value of the first result data set; Correspondingly, calculating the check value of the second result data set includes: Converting the second result data set into a string of fixed length by using a hash function to obtain a check value of the second result data set; or Feature extraction is performed on the second result data set to obtain a verification value of the second result data set.

[0011] Furthermore, the verification values ​​of the problem data set and the first result data set are transmitted via a network or a physical carrier.

[0012] Furthermore, before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set, the method further includes: Presetting a transmission order of the data set between different nodes in the system and / or network, the transmission order being an order of unidirectional transmission from the initial neural network to a neural network deployed on at least one node in the system and / or network; Correspondingly, the problem data set is unidirectionally transmitted to at least one node in the system and / or network in the transmission order, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set.

[0013] Furthermore, during the one-way transmission process, the neural network deployed at the next node in the system and / or network is determined according to the transmission order, and one-way transmission is performed from the initial neural network to the neural network of the next node. The neural network deployed at the next node repeats the one-way transmission process until the neural network deployed at the last node.

[0014] Furthermore, before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set, the method further includes: Encrypting the problem data set to obtain a ciphertext of the data set; Correspondingly, the ciphertext of the data set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node decrypts the ciphertext of the data set to obtain the problem data set, and generates at least one second result data set based on the problem data set.

[0015] According to a second aspect of the present application, a device for generating a consistent data set is provided, comprising: A selection unit, configured to select any one of at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to the problem data set, wherein the at least two neural networks have the same structural parameters and are respectively deployed on different nodes in a system and / or network, wherein the system and / or network is at least one; A selecting unit, configured to select at least one node in the system and / or network, so that a neural network deployed on the at least one node generates at least one second result data set according to the problem data set; A generating unit is configured to perform a target operation on the first result data set and the at least one second result data set as consistent data sets.

[0016] Furthermore, each neural network has a preset local data set set, and the selected unit includes: A first transmission module is used to transmit the problem data set to at least one node in the system and / or network if the problem data set is not initially stored in the local data set set of the neural network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; or transmit the first storage identifier of the problem data set in the remote data set set to at least one node in the system and / or network, so that the neural network deployed on the at least one node requests the verification value of the problem data set and the result data set from the remote data set set through the first storage identifier, and generates at least one second result data set according to the problem data set; A second transfer module is used to transfer a second storage identifier of the problem dataset in the local dataset set to at least one node in the system and / or network if the problem dataset is initially stored in the local dataset set of the neural network, so that the neural network deployed on the at least one node obtains the problem dataset in the local dataset set through the second storage identifier, and generates at least one second result dataset based on the problem dataset, and the second storage identifier is used to point to the same problem dataset in the local dataset sets of different neural networks.

[0017] Furthermore, the selection unit further includes: a third transmission module, configured to transmit the verification value of the problem data set and the first result data set to at least one node in the system and / or network before performing a target operation on the first result data set and the at least one second result data set as a consistent data set, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; a comparing unit, configured to compare the check value of the first result data set with the check value of the at least one second result data set; Correspondingly, the generating unit is specifically configured to, if the comparison results are consistent, use the first result data set and the at least one second result data set as consistent data sets to perform a target operation.

[0018] Furthermore, the device also includes: The problem troubleshooting unit is configured to, if the comparison result is inconsistent, treat the first result data set and the at least one second result data set as inconsistent data sets for problem troubleshooting.

[0019] Furthermore, the device also includes: a first calculation unit, configured to calculate a check value of the first result data set before comparing the check value of the first result data set with the check value of the at least one second result data set, so as to uniquely identify a feature of the first result data set by the check value; and The second calculation unit is configured to calculate a check value of the second result data set, so as to uniquely identify a feature of the second result data set through the check value.

[0020] Furthermore, the first computing unit is specifically configured to: Converting the first result data set into a string of fixed length using a hash function to obtain a check value of the first result data set; or Performing feature extraction on the first result data set to obtain a verification value of the first result data set; Accordingly, the second computing unit is specifically used for: Converting the second result data set into a string of fixed length by using a hash function to obtain a check value of the second result data set; or Feature extraction is performed on the second result data set to obtain a verification value of the second result data set.

[0021] Furthermore, the verification values ​​of the problem data set and the first result data set are transmitted via a network or a physical carrier.

[0022] Furthermore, the selection unit further includes: A setting module, used for presetting a transmission order of data sets between different nodes in the system and / or network before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set, wherein the transmission order is an order of unidirectional transmission from the initial neural network to the neural network deployed on at least one node in the system and / or network; Correspondingly, the first transmission module is specifically used to unidirectionally transmit the problem data set to at least one node in the system and / or network in accordance with the transmission order, so that the neural network deployed on the at least one node generates at least one second result data set based on the problem data set.

[0023] Furthermore, during the one-way transmission process, the neural network deployed at the next node in the system and / or network is determined according to the transmission order, and one-way transmission is performed from the initial neural network to the neural network of the next node. The neural network deployed at the next node repeats the one-way transmission process until the neural network deployed at the last node.

[0024] Furthermore, the selection unit further includes: an encryption module, configured to encrypt the problem data set to obtain a data set ciphertext before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; Correspondingly, the first transmission module is further specifically used to transmit the ciphertext of the data set to at least one node in the system and / or network, so that the neural network deployed on the at least one node decrypts the ciphertext of the data set to obtain the problem data set, and generates at least one second result data set based on the problem data set.

[0025] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the first aspect when executing the computer program.

[0026] According to a fourth aspect of the present application, a readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0027] By means of the above technical scheme, the present application provides a method, device, equipment and storage medium for generating a consistent data set. Compared with the current prior art method of generating a consistent data set by sharing, synchronizing or exchanging, the present application selects any one of at least two neural networks as the initial neural network, so that the initial neural network generates a first result data set according to a pre-constructed problem data set, and at least two neural networks have the same structural parameters, which are respectively deployed to different nodes in the system and / or network, and the system and / or network are at least one; at least one node in the system and / or network is selected, so that the neural network deployed on at least one node generates at least one second result data set according to the problem data set; the first result data set and at least one second result data set are used as consistent data sets for target operation. The whole process does not obtain a consistent data set by sharing or synchronizing, but realizes the transmission of the same problem data set by deploying at least two neural networks in the system and / or network. Since at least two neural networks have the same structural parameters, at least one neural network can use the same problem data set to generate a consistent result data set, which can avoid the risk of tampering caused by direct transmission of the result data set and ensure the security of generating a consistent data set.

[0028] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a flowchart of a method for generating a consistent data set in an embodiment of the present application; Figure 2 is a flowchart of a method for generating a consistent data set in another embodiment of the present application; Figure 3 is a flowchart of a method for generating a consistent data set in another embodiment of the present application; Figure 4 is a flowchart of a method for generating a consistent data set in another embodiment of the present application; Figure 5 is a flowchart of a method for generating a consistent data set in another embodiment of the present application; Figure 6 is a network architecture diagram of a consistent data set generation process in an embodiment of the present application; Figure 7 is a schematic diagram of the structure of a device for generating a consistent data set in an embodiment of the present application; Figure 8 It is a schematic diagram of the device structure of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] Now the content of the present invention will be discussed with reference to several exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and thus implement the content of the present invention, rather than implying any limitation on the scope of the present invention.

[0031] As used herein, the term "including" and variations thereof are to be interpreted as open-ended terms meaning "including but not limited to." The term "based on" is to be interpreted as "based, at least in part, on." The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment." The term "another embodiment" is to be interpreted as "at least one other embodiment."

[0032] In related technologies, the generation methods of consistent data sets in different fields are different. In the medical field, consistent data sets can be obtained through sharing, for example, different hospitals share patients' diagnostic data for joint analysis; in the financial field, consistent data sets can be obtained through synchronization, for example, multiple data centers synchronize transaction information in real time to ensure the consistency of ledgers. However, data sets obtained through sharing or synchronization require direct data transmission, which increases the risk of data tampering and makes it difficult to ensure data security.

[0033] In order to solve this problem, this embodiment provides a method for generating a consistent data set, such as Figure 1 As shown, the following steps are included: 101. Select any one neural network from at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to the problem data set.

[0034] In today's era of rapid development of digitalization and intelligence, system and / or network computing and neural network technology have become an important driving force for promoting scientific and technological progress. With the widespread application of human functional intelligence, especially in the fields of medical diagnosis, financial analysis, and autonomous driving, distributed neural networks formed by systems and / or networks have gradually become the core tools for solving large-scale data processing and complex tasks. Generally speaking, in the application scenarios of distributed neural networks, from cloud computing platforms to edge computing devices, and then to IoT terminals, it is necessary to ensure data consistency and accuracy in the process of multiple node collaboration. For example, in the medical field, different hospitals need to share patients' diagnostic data for joint analysis; in the financial field, multiple data centers need to synchronize transaction information in real time to ensure the consistency of ledgers; in the field of autonomous driving, the data exchange between vehicles and cloud servers must be highly consistent to ensure driving safety. These application scenarios have placed extremely high demands on the data consistency of distributed neural networks.

[0035] Among them, at least two neural networks have the same structural parameters and are respectively deployed on different nodes in the system and / or network, and the system and / or network is at least one. The execution subject of the embodiment of the present invention can be the system or server where the at least two neural networks are located. Through at least two neural networks with the same network structure, the same result data set can be output based on the same problem data set.

[0036] A twin neural network can be used here. The core structure of the twin neural network is composed of at least two neural networks. The neural network in the twin neural network is deployed to at least one network and / or system. That is to say, the twin neural network can exist in different networks and / or systems. For example, the twin neural network includes neural network A and neural network B. Neural network A and neural network B in the twin neural network can be deployed together on different nodes in the same network, or on different nodes in the same system. Neural network A and neural network B in the twin neural network can also be deployed separately on any node in different networks, or on any node in different systems. Neural network A in the neural network can also be deployed on any node in the network, and neural network B can be deployed on any node in the system.

[0037] It should be noted that each neural network in the twin neural network shares the same weights and parameters, that is, they have the same structural parameters, which ensures that each neural network in the twin neural network outputs the same result data set when the same problem data set is input. For example, if the same image is input into different neural networks in the twin neural network for feature extraction, in theory each neural network will output the same features.

[0038] In this embodiment, the pre-constructed question data set may be randomly generated by the initial neural network, that is, after the initial neural network generates the question data set, it generates the first result data set according to the question data set. The question data set may also be imported from the outside, that is, after the initial neural network receives the external input question data set, it generates the first result data set according to the question data set. The specific question data set may be in various forms, such as pictures, vectors, and any other data that can be used as neural network input values.

[0039] 102. Select at least one node in the system and / or network so that a neural network deployed on the at least one node generates at least one second result data set according to the problem data set; In this embodiment, at least one node in the selected system and / or network can specify a node, and the neural network deployed on the node can generate at least one second result data set according to the problem data set. Here, the problem data set can be obtained through a local data set set, can be obtained through a remote data set set, or can be transmitted by the initial neural network, which is not limited in this embodiment.

[0040] That is to say, at least one node in the selected system and / or network has the ability to obtain the problem data set, and a neural network is deployed on the node. After receiving the problem data set, the neural network deployed on the node can generate a second result data set according to the problem data set.

[0041] 103. Perform a target operation on the first result data set and the at least one second result data set as consistent data sets.

[0042] It can be understood that since at least two neural networks have the same structural parameters, theoretically, at least two neural networks can generate a consistent result data set through the same problem data set, that is, the first result data set and at least one second result data set are consistent, and the first result data set and the second result data set can be used as consistent data sets for target operations, for example, merging data of the consistent data sets, triggering a business process, or starting a distributed transaction.

[0043] Specifically, in the scenario of order data synchronization on an e-commerce platform, the initial neural network can generate order data based on the source data, and transmit the source data to at least one node in the network and / or system. The neural network deployed on at least one node can generate at least one second order data based on the source data, and the first order data and the at least one second order data are treated as consistent order data to achieve order data synchronization.

[0044] The method for generating a consistent data set provided in the embodiment of the present application is compared with the method of generating a consistent data set by sharing, synchronizing or exchanging in the prior art. The present application selects any neural network as the initial neural network in at least two neural networks so that the initial neural network generates a first result data set according to a pre-constructed problem data set, and at least two neural networks have the same structural parameters and are respectively deployed to different nodes in the system and / or network, and the system and / or network are at least one; at least one node in the system and / or network is selected so that the neural network deployed on at least one node generates at least one second result data set according to the problem data set; the first result data set and at least one second result data set are used as consistent data sets for target operation. The whole process does not obtain a consistent data set by sharing or synchronizing, but realizes the transmission of the same problem data set by deploying at least two neural networks in the system and / or network. Since at least two neural networks have the same structural parameters, at least one neural network can use the same problem data set to generate a consistent result data set, which can avoid the risk of tampering caused by direct transmission of the result data set and ensure the security of generating a consistent data set.

[0045] In actual application scenarios, each neural network has a preset local data set set. In the process of selecting at least one node in a system and / or network so that the neural network deployed on at least one node generates at least one second result data set based on the problem data set, the problem data set may be initially stored in the local data set set or may not be initially stored in the local data set set.

[0046] Specifically, if the problem data set is not initially stored in the local data set set of the neural network, the problem data set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; or the first storage identifier of the problem data set in the remote data set set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node requests the verification value of the problem data set and the result data set from the remote data set set through the first storage identifier, and generates at least one second result data set according to the problem data set; In the scenario of transmitting the problem data set or the first storage identifier of the problem data set in the remote data set set, the problem data set or the first storage identifier can be transmitted to a selected node in the system and / or network, for example, to the node A1 in the distributed system of the system. The problem data set or the first storage identifier can also be transmitted to multiple nodes in the system and / or network, for example, to the node A1 in the distributed system and the node B1 in the distributed network.

[0047] It should be noted that in the scenario where the problem dataset is first stored in the remote dataset set, the remote dataset set can also generate a corresponding result dataset in advance through a standard neural network at the same time as the problem dataset, where the standard neural network has the same structural parameters as at least two neural networks, and the result dataset generated by the standard neural network is theoretically consistent with the result dataset generated by at least two neural networks. The corresponding verification value of the result dataset is calculated, and when a data acquisition request is received, the verification values ​​of the problem dataset and the result dataset are fed back to the neural network, so that after the neural network generates at least one second result dataset according to the problem dataset, the consistency of the at least one second result dataset is verified according to the verification value of the result dataset.

[0048] In this scenario, the problem dataset is not initially stored in the local dataset set of the neural network, which means that the problem dataset is not initially stored in the local dataset set of the neural network being transmitted. In this case, the problem dataset needs to be transmitted to at least one node in the system and / or network, or the problem dataset needs to be requested from a remote dataset set. In this case, the first storage identifier of the problem dataset in the remote dataset set needs to be transmitted. Through the first storage identifier, a data acquisition request can be initiated to the remote dataset set, and the same problem dataset can be obtained accordingly.

[0049] Specifically, if the problem data set is initially stored in a local data set set of a neural network, a second storage identifier of the problem data set in the local data set set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node obtains the problem data set in the local data set set through the second storage identifier, and generates at least one second result data set based on the problem data set. The second storage identifier is used to point to the same problem data set in local data set sets of different neural networks.

[0050] In this scenario, the problem dataset is initially stored in the local dataset set of the neural network. In this way, there is no need to transfer the problem dataset between neural networks. Instead, only the second storage identifier of the problem dataset in the local dataset set needs to be transferred. Since the second storage identifier is used to point to the same problem dataset in the local dataset sets of different neural networks, the neural network at the receiving end can obtain the same problem dataset from the local dataset set and generate a consistent result dataset.

[0051] In this embodiment, the local data set set is a data set set stored locally by each neural network in the system and / or network, that is, the local data set set of each neural network contains at least one identical problem data set, and by transmitting the second storage identifier of the problem data set in the local data set set, the receiving neural network can obtain the same problem data set from the local data set set according to the second storage identifier, and then generate a second result data set based on the same problem data set.

[0052] Furthermore, in order to verify the second result data set, the second storage identifier of the problem data set in the local data set set and the verification value of the first result data set can also be transmitted to at least one node in the system and / or network, so that the neural network deployed on at least one node obtains the problem data through the storage identifier, generates at least one second result data set based on the problem data set, and compares the verification value of the first result data set with the verification value of at least one second result data set to verify the consistency of the generated result data set.

[0053] It should be noted that in order to reduce the amount of data transmission, the local data set set can also generate a corresponding result data set in advance through a standard neural network while storing the problem data set. Here, the standard neural network has the same structural parameters as at least two neural networks, and the result data set generated by the standard neural network is theoretically consistent with the result data sets generated by at least two neural networks. Accordingly, the receiving end neural network can directly verify the consistency of the generated second result data set through the result data set stored in the local data set set.

[0054] In actual application scenarios, in order to verify the consistency of at least two neural network generated result data sets, in the scenario where the problem data set is not initially stored in the local data set set of the neural network, that is, before the problem data set is transmitted, the check value of the result data set can be pre-calculated, and the consistency verification can be performed through the check value of the result data set. Further, Figure 2 As shown, before step 103, the method further includes: 201. Transmit the verification values ​​of the problem data set and the first result data set to at least one node in the system and / or network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set.

[0055] 202. Compare the verification value of the first result data set with the verification value of the at least one second result data set.

[0056] In this embodiment, the check value of the first result data set is used to check the integrity of the first result data set to ensure that the first result data set has not been tampered with or damaged. The check value of the specific first result data set can be generated by calculating the content of the first result data set using a specific algorithm, and is usually a unique string of fixed length, which is used to identify the content of the second result data set. As long as any part of the first result data set is changed, this string will be different.

[0057] It should be noted that the check value of the second result data set is generated in the same way as the check value of the first result data set, that is, a specific algorithm is used to calculate and generate the content of the second result data set to obtain a unique fixed-length string for identifying the content of the second result data set.

[0058] In this embodiment, the process of calculating the check value of the first result data is performed after the initial neural network generates the first result data set. Similarly, the process of calculating the check value of the second result data set is performed after the receiving end neural network generates at least one second result data set. Theoretically, the structural parameters of at least one neural network are the same as those of the initial neural network. When the input problem data sets are the same, the corresponding output first result data set and at least one second result data set should also be the same. Here, by comparing the check value of the first result data with the check value of the second result data set, it is possible to determine whether the data sets are consistent without comparing the data sets byte by byte.

[0059] However, if the problem data set is damaged or tampered with during the transmission process, the problem data set received by at least one node being transmitted will be different from the problem data set used by the initial neural network, so that the second result data set generated by the transmitted neural network is different from the first result data set. If the comparison results are inconsistent, the first result data set and at least one second result data set are treated as inconsistent data sets for troubleshooting.

[0060] Correspondingly, 103 , if the comparison results are consistent, the first result data set and the at least one second result data set are used as consistent data sets to perform a target operation.

[0061] In actual application scenarios, directly transmitting the result data set often increases the risk of data being tampered with during the transmission process. Due to the large amount of data in the result data set, it is difficult to perform detailed security authentication on each data fragment. Once the data is maliciously tampered with during the transmission process, it is difficult for the receiving node to detect it, resulting in subsequent processing based on erroneous data, affecting the correctness of the entire system and / or network. Furthermore, if Figure 3 As shown, before step 202, the method further includes: 301. Calculate a check value of the first result data set to uniquely identify a feature of the first result data set through the check value; and 302. Calculate a check value of the second result data set to uniquely identify a feature of the second result data set through the check value.

[0062] It can be understood that the calculation of the verification value of the first result data set can be triggered after the initial neural network generates the first result data set based on the pre-constructed problem data set, and correspondingly, the calculation of the verification value of the second result data set can be triggered after the neural network deployed on at least one node generates at least one second result data set based on the problem data set.

[0063] Specifically, in the process of calculating the check value of the first result data set, the first result data set can be converted into a string of fixed length through a hash function to obtain the check value of the first result data set; or the first result data set can be feature extracted to obtain the check value of the first result data set. Here, the hash function is a common method for generating check values, including SHA-256, MD5, etc. The hash function can convert data of any length into a string of fixed length, which is collision-resistant and deterministic, and is suitable for verifying the integrity and consistency of the data. Here, the feature extraction process does not rely on the byte-by-byte integrity of the data, but identifies the core content of the data set through semantic or structural features, which is suitable for verifying the content consistency of the data set. The specific feature extraction process can select a feature extraction method according to the type of the first result data set, convert the extracted features into a vector of fixed dimension or a structured summary, and obtain the check value of the first result data set.

[0064] Accordingly, in the process of calculating the check value of the second result data set, the second result data set can be converted into a string of fixed length by a hash function to obtain the check value of the second result data set; or feature extraction is performed on the second result data set to obtain the check value of the second result data set. This process is the same as the process of calculating the check value of the first result data set, and will not be repeated here.

[0065] It should be noted that the same hash function should be used to calculate the check value of the second result data set and the check value of the first result data set, so as to ensure the accuracy of the data verification result.

[0066] In actual application scenarios, in order to effectively manage and transmit data, in scenarios where the problem data set is not initially stored in the local data set set of the neural network, that is, in the process of transmitting the verification value between the problem data set and the result data set, the verification value between the problem data set and the first result data set is transmitted through a network carrier or a physical carrier.

[0067] Here, the network transmission can be implemented by data encapsulation and the like, and the check values ​​of the problem data set and the first result data set are encapsulated into a data packet and then transmitted. Here, the physical carrier transmission can be implemented by a USB flash drive and the like, and the check values ​​of the problem data set and the first result data set are transmitted through the USB flash drive.

[0068] Specifically, in the process of using data encapsulation to achieve transmission, the problem data set and the verification value of the first result data set can be combined together and encapsulated according to a certain format and protocol to form a data packet. Here, other metadata can also be added to the data packet during the encapsulation process, such as the data set identifier corresponding to the verification value, the generation time, etc.

[0069] Correspondingly, after receiving the data packet, the neural network at the receiving end will parse it according to the previously encapsulated format and protocol. The parsing process is to extract the various parts of the data packet, including the problem data set, the result data set check value, and related metadata.

[0070] In practical application scenarios, considering the order of data set transmission between nodes in the system and / or network, the transmission order can be predefined in the scenario where the problem data set is not initially stored in the local data set set of the neural network, that is, before the problem data set is transmitted. Further, Figure 4 As shown, before step 102, the method further includes the following steps: 401. Preset a transmission order of data sets between different nodes in the system and / or network.

[0071] Correspondingly, in step 102, the problem data set is unidirectionally transmitted to at least one node in the system and / or network in the transmission order, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set.

[0072] In this embodiment, the transmission order is the order of unidirectional transmission from the initial neural network to the neural network deployed on at least one node in the system and / or network. Here, unidirectional transmission refers to transmission in only one direction and not in the reverse direction, that is, the transmitted neural network can only receive data sets and cannot transmit data sets to the initial neural network. For example, the system includes nodes A1 and A2. If the initial neural network is deployed in node A1, the unidirectional transmission order is that the neural network deployed in node A1 transmits the data set to the neural network deployed in node A2, while the neural network deployed in node A2 cannot transmit the data set to the neural network deployed in node A1. Specifically, the unidirectional transmission process can be achieved through data diodes, which can ensure that the data set is only transmitted from the initial neural network to the neural network deployed in at least one node in the system and / or network, while completely preventing reverse communication.

[0073] Specifically, in the one-way transmission process, the neural network deployed at the next node in the system and / or network is determined according to the transmission order, and the initial neural network is transmitted one-way to the neural network of the next node. The neural network deployed at the next node is used as the initial neural network, and the above one-way transmission process is repeated until the neural network deployed at the last node.

[0074] In an application scenario, the transmission order may be transmission in a system or network, and the specific transmission method may be traversal transmission, in which case the data set is traversed and transmitted from the initial neural network in the system or network to the neural networks deployed in all nodes. Taking the transmission of data sets in a distributed system as an example, three nodes A1, A2, and A3 are deployed in the distributed system, and the neural network deployed in node A2 is used as the initial neural network. The transmission order is node A2-node A1-node A3, so that the check value of the data set is transmitted from the initial neural network deployed in node A2 to the neural network deployed in node A1. After the neural network deployed in node A1 generates a consistent data set, the neural network deployed in node A1 is used as the initial neural network, and the check value of the corresponding data set is transmitted from the initial neural network deployed in node A1 to the neural network deployed in node A3. After the neural network deployed in node A3 generates a consistent data set, the neural networks deployed in all nodes in the distributed system generate consistent data sets. The specific transmission method may also be distributed transmission, in which case the data set is distributed and transmitted from the initial neural network in the system or network to the neural networks deployed in multiple or all nodes. Continuing with the above example, the neural network deployed at node A2 is used as the initial neural network, and the transmission order is node A2-node A1, node A2-node A3. In this way, the check value of the data set is transmitted from the initial neural network deployed at node A2 to the neural network deployed at node A1, and the neural network deployed at node A1 generates a consistent data set. At the same time, the check value of the data set is transmitted from the initial neural network deployed at node A2 to the neural network deployed at node A3, and the neural network deployed at node A3 generates a consistent data set. In this way, the neural networks deployed at all nodes in the distributed system generate consistent data sets. The specific transmission method can also be a combination of traversal transmission and decentralized transmission. Taking the transmission of data sets in a distributed system as an example, there are 4 nodes A1, A2, A3, and A4 deployed in the distributed system. The neural network deployed at node A2 is used as the initial neural network, and the transmission order is node A2-node A1-node A3, node A2-node A4.

[0075] In another application scenario, the transmission order can be interactive transmission in the system and network. Similarly, the specific transmission method can be traversal transmission. At this time, the data set is traversed and transmitted by the initial neural network in the system or network to the neural network deployed in other networks or systems. The data set is transmitted between the neural networks deployed in the nodes of distributed system system 1 and distributed system 2. For example, distributed system 1 includes nodes A1 and A2, and distributed system 2 includes nodes B1 and B2. The initial neural network is the neural network deployed by node A1, and the transmission order is node A1-node B1-node B2-node A2. The corresponding process of generating consistent data is the same as the process of traversal transmission above, which is not repeated here. The specific transmission method can also be decentralized transmission. At this time, the data set is decentralized from the initial neural network in the system or network to the neural networks deployed in other networks or systems. Continuing with the above example, the initial neural network is the neural network deployed by node A1, and the transmission order is node A1-node B1, node A1-node A2, node A1-node B2. The corresponding process of generating consistent data is the same as the process of decentralized transmission above, which is not repeated here. The specific transmission method can also be a combination of traversal transmission and decentralized transmission. Continuing with the above example, the initial neural network is the neural network deployed by node A1, and the transmission order is node A1-node B1, node A1-node A2-node B2. The corresponding process of generating consistent data is the same as the process of combining traversal transmission and decentralized transmission above, and will not be repeated here.

[0076] In actual application scenarios, considering the privacy security of the problem dataset during transmission, the problem dataset can be encrypted before it is transmitted, in the scenario where the problem dataset is not initially stored in the local dataset set of the neural network. Figure 5 As shown, further, before step 102, the method further includes the following steps: 501. Encrypt the problem data set to obtain a ciphertext of the data set.

[0077] Correspondingly, in step 102, the ciphertext of the data set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node decrypts the ciphertext of the data set to obtain the problem data set, and generates at least one second result data set based on the problem data set.

[0078] In this embodiment, the problem data set can be in plain text or cipher text during transmission. Considering the flexibility in the data transmission process, the problem data set can also be encrypted before transmission. If the network and / or system has an encryption and decryption system, the problem data set can be encrypted so that the neural network receiving the problem data set can decrypt the data set cipher text according to the encryption and decryption system to protect data privacy and integrity and prevent the problem data set from being stolen or tampered with during transmission. Specific encryption processing methods include but are not limited to symmetric encryption algorithms, asymmetric algorithms, hash algorithms, etc.

[0079] Similarly, in the scenario where the problem dataset is not initially stored in the local dataset set of the neural network, the problem dataset and the first result dataset can also be encrypted before being transmitted. In this case, the problem dataset and the first result dataset are encrypted before the verification values ​​of the problem dataset and the first result dataset are transmitted to at least one node in the system and / or network. Accordingly, the dataset ciphertext obtained by encrypting the dataset and the first result dataset is transmitted to at least one node in the system and / or network, so that the neural network deployed on at least one node decrypts the dataset ciphertext to obtain the verification values ​​of the problem dataset and the first result dataset, generates at least one second result dataset based on the problem dataset, and performs consistency verification on at least one second result dataset based on the verification value of the first result dataset.

[0080] In actual application scenarios, the network architecture of the consistent dataset generation process is as follows: Figure 6 As shown in Figure 1. A large financial institution needs to conduct real-time risk assessment of its customers' transaction behaviors and implement target operations between multiple data centers based on the assessment results. In order to improve computing efficiency and security, the institution adopts a distributed neural network architecture to distribute risk assessment tasks to multiple data centers (nodes) for processing. Among them, one data center (neural network A) is responsible for generating problem sets and result data sets, and transmitting relevant information through a one-way transmission channel, while the other data center (neural network B) uses the same result data set to complete specific business operations.

[0081] Specifically, in the process of generating a consistent data set, neural network A receives an external input problem data set, for example, a group of customer transaction behavior characteristics, including "transaction amount range", "transaction frequency", "transaction time distribution", etc. Then, neural network A generates a first result data set based on the problem data set, that is, a real-time risk score for each customer, for example, low, medium, and high risk levels. However, these scores are highly sensitive privacy data and should be encrypted to calculate the verification values ​​of the problem data set and the first result data set. For example, the SHA-256 hash function is used to generate unique summary information for the problem data set and the first result data set as the verification value. Further, neural network A encapsulates the verification values ​​of the problem data set and the result data set into a data packet, and transmits it to neural network B through a unidirectional transmission channel. Correspondingly, after receiving the data packet, neural network B parses the problem data set therein and generates a second result data set based on the same problem set, that is, the real-time risk score for each customer. Since neural network A and neural network B have exactly the same structural parameters, the result data sets generated by the two should be completely consistent in theory, but there is still a risk of inconsistency. Further, neural network B uses the same processing method to calculate the check value of the second result data set it generates, and compares the check value of the second result data set with the check value of the first result data set received from neural network A. If the two are consistent, it means that the result data sets generated by neural network A and neural network B are completely consistent, and the consistency verification is successful. In other words, neural network B can use the generated second result data set to complete specific business operations. For example, according to the customer's risk score, restrict certain transaction permissions of high-risk customers (such as large transfers or night transactions); provide personalized financial product recommendations or preferential activities for medium and low-risk customers; when high-risk behavior is detected, automatically trigger the internal alarm system to remind relevant departments to conduct manual verification. If the two are inconsistent, it means that there may be data transmission errors or other abnormalities in the data set, which need to be further investigated. Correspondingly, neural network B can trigger an error handling mechanism, for example, notifying the administrator to intervene.

[0082] In the above-mentioned consistency data generation process, firstly, by only transmitting the problem set and the check value instead of the complete result data set, the communication bandwidth occupancy and delay are greatly reduced, thereby improving the real-time and scalability of the system. Secondly, the use of the check value avoids the direct exposure of sensitive data and meets the needs of privacy protection, which is particularly suitable for application scenarios involving user privacy or commercial secrets. Finally, the consistency verification mechanism based on the check value ensures that even in the case of network delays or partitions, each node can still accurately judge the data consistency, thereby improving the reliability and stability of the system. Through the above-mentioned consistency data generation process, not only the problems of high communication overhead and high privacy risks in the prior art are solved, but also a more efficient and secure solution is provided for the data consistency of distributed neural networks.

[0083] Further, as Figure 1-5 The specific implementation of the method, the embodiment of the present application provides a device for generating a consistent data set, such as Figure 7 As shown, the device includes: a selection unit 61, a selection unit 62 and a generation unit 63.

[0084] A selection unit 61 is used to select any one of the at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to the problem data set, the at least two neural networks have the same structural parameters, are respectively deployed to different nodes in the system and / or network, and each neural network has a preset local data set set; A selecting unit 62, configured to select at least one node in the system and / or network, so that a neural network deployed on the at least one node generates at least one second result data set according to the problem data set; The generating unit 63 is configured to perform a target operation on the first result data set and the at least one second result data set as consistent data sets.

[0085] The device for generating a consistent data set provided by an embodiment of the present application selects any one neural network as an initial neural network from at least two neural networks, so that the initial neural network generates a first result data set according to a pre-constructed problem data set, and at least two neural networks have the same structural parameters, and are respectively deployed to different nodes in the system and / or network, and the system and / or network are at least one; at least one node in the system and / or network is selected, so that the neural network deployed on at least one node generates at least one second result data set according to the problem data set; the first result data set and at least one second result data set are used as consistent data sets for target operation. The whole process does not obtain a consistent data set through sharing or synchronization, but realizes the transmission of the same problem data set by deploying at least two neural networks in the system and / or network. Since at least two neural networks have the same structural parameters, at least one neural network can use the same problem data set to generate a consistent result data set, which can avoid the risk of tampering caused by direct transmission of the result data set and ensure the security of generating a consistent data set.

[0086] In actual application scenarios, each neural network has a preset local data set set, and the selected unit includes: A first transmission module is used to transmit the problem data set to at least one node in the system and / or network if the problem data set is not initially stored in the local data set set of the neural network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; or transmit the first storage identifier of the problem data set in the remote data set set to at least one node in the system and / or network, so that the neural network deployed on the at least one node requests the verification value of the problem data set and the result data set from the remote data set set through the first storage identifier, and generates at least one second result data set according to the problem data set; A second transfer module is used to transfer a second storage identifier of the problem dataset in the local dataset set to at least one node in the system and / or network if the problem dataset is initially stored in the local dataset set of the neural network, so that the neural network deployed on the at least one node obtains the problem dataset in the local dataset set through the second storage identifier, and generates at least one second result dataset based on the problem dataset, and the second storage identifier is used to point to the same problem dataset in the local dataset sets of different neural networks.

[0087] In an actual application scenario, the selection unit further includes: a third transmission module, configured to transmit the verification value of the problem data set and the first result data set to at least one node in the system and / or network before performing a target operation on the first result data set and the at least one second result data set as a consistent data set, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; a comparing unit, configured to compare the check value of the first result data set with the check value of the at least one second result data set; Correspondingly, the generating unit is specifically configured to, if the comparison results are consistent, use the first result data set and the at least one second result data set as consistent data sets to perform a target operation.

[0088] In an actual application scenario, the device further includes: The problem troubleshooting unit is configured to, if the comparison result is inconsistent, treat the first result data set and the at least one second result data set as inconsistent data sets for problem troubleshooting.

[0089] In an actual application scenario, the device further includes: a first calculation unit, configured to calculate a check value of the first result data set before comparing the check value of the first result data set with the check value of the at least one second result data set, so as to uniquely identify a feature of the first result data set by the check value; and The second calculation unit is configured to calculate a check value of the second result data set, so as to uniquely identify a feature of the second result data set through the check value.

[0090] In an actual application scenario, the first computing unit is specifically used to: Converting the first result data set into a string of fixed length using a hash function to obtain a check value of the first result data set; or Performing feature extraction on the first result data set to obtain a verification value of the first result data set; Accordingly, the second computing unit is specifically used for: Converting the second result data set into a string of fixed length by using a hash function to obtain a check value of the second result data set; or Feature extraction is performed on the second result data set to obtain a verification value of the second result data set.

[0091] In an actual application scenario, the verification values ​​of the problem data set and the first result data set are transmitted via a network or a physical carrier.

[0092] In an actual application scenario, the selection unit further includes: A setting module, used for presetting a transmission order of data sets between different nodes in the system and / or network before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set, wherein the transmission order is an order of unidirectional transmission from the initial neural network to the neural network deployed on at least one node in the system and / or network; Correspondingly, the first transmission module is specifically used to unidirectionally transmit the problem data set to at least one node in the system and / or network in accordance with the transmission order, so that the neural network deployed on the at least one node generates at least one second result data set based on the problem data set.

[0093] In actual application scenarios, during the one-way transmission process, the neural network deployed at the next node in the system and / or network is determined according to the transmission order, and one-way transmission is performed from the initial neural network to the neural network of the next node. The neural network deployed at the next node repeats the one-way transmission process until the neural network deployed at the last node.

[0094] In an actual application scenario, the selection unit further includes: an encryption module, configured to encrypt the problem data set to obtain a data set ciphertext before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; Correspondingly, the first transmission module is further specifically used to transmit the ciphertext of the data set to at least one node in the system and / or network, so that the neural network deployed on the at least one node decrypts the ciphertext of the data set to obtain the problem data set, and generates at least one second result data set based on the problem data set.

[0095] It should be noted that for other corresponding descriptions of the functional units involved in the apparatus for generating a consistent data set provided in this embodiment, reference can be made to Figure 1-Figure 5 The corresponding description in will not be repeated here.

[0096] Based on the above Figure 1-Figure 5 The method shown in the embodiment of the present application accordingly provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned Figure 1-Figure 5 The method for generating the consistency dataset shown.

[0097] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0098] Based on the above Figure 1-Figure 5 The method shown, and Figure 7 In order to achieve the above-mentioned purpose, the embodiment of the present application also provides a physical device for generating a consistent data set, which can be a computer, a smart phone, a tablet computer, a smart watch, a server, or a network device, etc. The physical device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1-Figure 5 The method for generating the consistency dataset shown.

[0099] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0100] In an exemplary embodiment, see Figure 8 The physical device includes a communication bus, a processor, a memory and a communication interface, and may also include an input / output interface and a display device, wherein each functional unit may communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the method for generating a consistent data set in the above embodiment.

[0101] Those skilled in the art will appreciate that the physical device structure for generating a consistent data set provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0102] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device that generates the above-mentioned consistent data set, and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0103] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. By applying the technical solution of the present application, compared with the current existing methods, the present application implements the transmission of the same problem data set by deploying at least two neural networks in the system and / or network. Since at least two neural networks have the same structural parameters, at least one neural network can use the same problem data set to generate a consistent result data set, which can avoid the risk of tampering caused by the direct transmission of the result data set and ensure the security of generating a consistent data set.

[0104] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0105] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A method for generating a consistent data set, characterized in that: include: Selecting any one of the at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to the problem data set, wherein the at least two neural networks have the same structural parameters and are respectively deployed on different nodes in the system and / or network, and the system and / or network is at least one; Selecting at least one node in the system and / or network so that a neural network deployed on the at least one node generates at least one second result data set based on the problem data set; The first result data set and the at least one second result data set are used as consistent data sets to perform a target operation.

2. The method according to claim 1, characterized in that Each neural network has a preset local data set set, and the selecting at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set includes: If the problem data set is not initially stored in the local data set set of the neural network, the problem data set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; or the first storage identifier of the problem data set in the remote data set set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node requests the verification value of the problem data set and the result data set from the remote data set set through the first storage identifier, and generates at least one second result data set according to the problem data set; If the problem data set is initially stored in a local data set set of a neural network, a second storage identifier of the problem data set in the local data set set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node obtains the problem data set in the local data set set through the second storage identifier, and generates at least one second result data set based on the problem data set. The second storage identifier is used to point to the same problem data set in local data set sets of different neural networks.

3. The method according to claim 2, characterized in that Before performing a target operation on the first result data set and the at least one second result data set as consistent data sets, the method further includes: Transmitting the verification value of the problem data set and the first result data set to at least one node in the system and / or network, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set; comparing the check value of the first result data set with the check value of the at least one second result data set; Correspondingly, if the comparison results are consistent, the first result data set and the at least one second result data set are used as consistent data sets to perform target operations.

4. The method according to claim 3, characterized in that If the comparison result is inconsistent, the first result data set and the at least one second result data set are treated as inconsistent data sets for troubleshooting.

5. The method according to claim 3, characterized in that: Before comparing the check value of the first result data set with the check value of the at least one second result data set, the method further includes: Calculating a check value of the first result data set to uniquely identify a feature of the first result data set through the check value; and A check value of the second result data set is calculated to uniquely identify a feature of the second result data set through the check value.

6. The method according to claim 5, characterized in that The calculating the check value of the first result data set includes: Converting the first result data set into a string of fixed length using a hash function to obtain a check value of the first result data set; or Performing feature extraction on the first result data set to obtain a verification value of the first result data set; Correspondingly, calculating the check value of the second result data set includes: Converting the second result data set into a string of fixed length by using a hash function to obtain a check value of the second result data set; or Feature extraction is performed on the second result data set to obtain a verification value of the second result data set.

7. The method according to claim 3, characterized in that The verification values ​​of the problem data set and the first result data set are transmitted through a network or a physical carrier.

8. The method according to claim 2, characterized in that: Before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set, the method further includes: Presetting a transmission order of the data set between different nodes in the system and / or network, the transmission order being an order of unidirectional transmission from the initial neural network to a neural network deployed on at least one node in the system and / or network; Correspondingly, the problem data set is unidirectionally transmitted to at least one node in the system and / or network in the transmission order, so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set.

9. The method according to claim 8, characterized in that During the one-way transmission process, the neural network deployed at the next node in the system and / or network is determined according to the transmission order, and one-way transmission is performed from the initial neural network to the neural network of the next node. The neural network deployed at the next node repeats the one-way transmission process until the neural network deployed at the last node.

10. The method according to claim 2, characterized in that Before transmitting the problem data set to at least one node in the system and / or network so that the neural network deployed on the at least one node generates at least one second result data set according to the problem data set, the method further includes: Encrypting the problem data set to obtain a ciphertext of the data set; Correspondingly, the ciphertext of the data set is transmitted to at least one node in the system and / or network, so that the neural network deployed on the at least one node decrypts the ciphertext of the data set to obtain the problem data set, and generates at least one second result data set based on the problem data set.

11. A device for generating a consistent data set, characterized in that: include: A selection unit, configured to select any one of at least two neural networks as an initial neural network, so that the initial neural network generates a first result data set according to the problem data set, wherein the at least two neural networks have the same structural parameters and are respectively deployed on different nodes in a system and / or network, wherein the system and / or network is at least one; A selecting unit, configured to select at least one node in the system and / or network, so that a neural network deployed on the at least one node generates at least one second result data set according to the problem data set; A generating unit is configured to perform a target operation on the first result data set and the at least one second result data set as consistent data sets.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method for generating a consistent data set according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for generating a consistent data set according to any one of claims 1 to 10 are implemented.

Images