Business processing method, device and equipment based on Gaussian error linear unit

Abstract

The invention provides a service processing method, device and equipment based on a Gaussian error linear unit, and belongs to the technical field of data processing. Obtaining a service feature element obtained by processing the service attribute contained in the target data by the service processing model before the Gaussian error linear unit, and splitting the service feature element to obtain a first sub-service feature element and a second sub-service feature element; obtaining a first sub-feature obtained by processing the first sub-business feature element by the first service node according to the target function, obtaining a second sub-feature obtained by processing the second sub-business feature element by the second service node according to the target function, determining a target sub-business feature according to the first sub-feature and the second sub-feature, and sending the target sub-business feature to the server; and the target service vector is generated according to the target sub-service feature, and data analysis is performed according to the target service vector, so that the service processing efficiency can be improved while the service data security is ensured.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a business processing method, device, and equipment based on a Gaussian error linear unit. Background Art

[0002] With the development of information technology, business processing models can be used to analyze business data to automatically provide business services to target objects. For example, disease diagnosis models can be used to classify disease cases of target objects. However, in medical and other business fields, the business data of objects may have protection requirements, and the security of the business data of objects needs to be ensured during the data analysis process.

[0003] Related technologies use a business processing model to perform a secure multi-party computation (SMPC) algorithm to protect the privacy of business data. However, the business processing model usually involves a Gaussian Error Linear Unit (GELU), which contains a large number of nonlinear operations on SMPC, reducing the efficiency of business processing. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a business processing method, device, and equipment based on a Gaussian error linear unit, aiming to improve the efficiency of business processing while ensuring the security of business data during the data analysis process.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a service processing method based on a Gaussian error linear unit, the method comprising:

[0006] Acquire target data of a target object for a target business, wherein the target data includes at least one business attribute;

[0007] Inputting the target data into a business processing model, and obtaining processing progress information of the target data by the business processing model;

[0008] When the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, obtaining the objective function corresponding to the Gaussian error linear unit, and obtaining the business feature elements obtained by the business processing model before the Gaussian error linear unit processes the business attributes contained in the target data;

[0009] Splitting the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element;

[0010] Acquire a first sub-feature obtained by processing the first sub-service feature element by the first service node according to the objective function, and acquire a second sub-feature obtained by processing the second sub-service feature element by the second service node according to the objective function;

[0011] Determining a target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature;

[0012] A target business vector corresponding to the target data is generated according to the target sub-business characteristics corresponding to the business attributes, and data analysis is performed according to the target business vector.

[0013] In some embodiments, obtaining the first sub-feature obtained by the first service node processing the first sub-service feature element according to the objective function includes:

[0014] Acquire a privacy protection algorithm according to the objective function;

[0015] The first service node calls the privacy protection algorithm to perform privacy calculation on the first sub-service feature element to obtain the first sub-feature.

[0016] In some embodiments, the invoking the privacy protection algorithm by the first service node to perform privacy calculation on the first sub-service feature element to obtain the first sub-feature includes:

[0017] Invoking the privacy protection algorithm by the first service node to perform a privacy comparison calculation on the first sub-service feature element to obtain first privacy comparison information and second privacy comparison information;

[0018] Performing privacy multiplication calculation on the first sub-service characteristic element to obtain an initial sub-service characteristic;

[0019] The first sub-feature is determined according to the first privacy comparison information, the second privacy comparison information, and the initial sub-service feature.

[0020] In some embodiments, determining the first sub-feature according to the first privacy comparison information, the second privacy comparison information, and the initial sub-service feature includes:

[0021] Determine target privacy comparison information according to the first privacy comparison information and the second privacy comparison information;

[0022] Performing error calculation on the initial sub-service characteristic to obtain an error characteristic;

[0023] Performing privacy multiplication calculation on the target privacy comparison information and the error feature to obtain an intermediate sub-service feature;

[0024] The first sub-feature is determined according to the first privacy comparison information, the second privacy comparison information, and the intermediate sub-service feature.

[0025] In some embodiments, the Gaussian error linear unit is provided with a Gaussian error linear function, the Gaussian error linear function includes an initial sub-error function, and obtaining the objective function corresponding to the Gaussian error linear unit includes:

[0026] Performing function fitting on the initial sub-error function to obtain a target sub-error function;

[0027] The Gaussian error linear function is updated according to the target sub-error function to obtain the target function.

[0028] In some embodiments, splitting the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element includes:

[0029] Randomly generate reference feature elements;

[0030] The service characteristic element is split according to the reference characteristic element to obtain the first sub-service characteristic element and the second sub-service characteristic element.

[0031] In some embodiments, determining the target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature includes:

[0032] Adding the first sub-feature and the second sub-feature to obtain a reference sub-service feature;

[0033] The reference sub-service characteristic is subjected to modulo processing to obtain the target sub-service characteristic.

[0034] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a service processing device based on a Gaussian error linear unit, the device comprising:

[0035] A first acquisition module is used to acquire target data of a target object for a target business, wherein the target data includes at least one business attribute;

[0036] An input module, used to input the target data into the business processing model and obtain processing progress information of the target data by the business processing model;

[0037] A second acquisition module is used to obtain, when the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, an objective function corresponding to the Gaussian error linear unit, and obtain a business feature element obtained by the business processing model before the Gaussian error linear unit processes the business attributes contained in the target data;

[0038] A splitting module, used for splitting the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element;

[0039] A third acquisition module is used to acquire a first sub-feature obtained by the first service node processing the first sub-service feature element according to the objective function, and to acquire a second sub-feature obtained by the second service node processing the second sub-service feature element according to the objective function;

[0040] A determination module, configured to determine a target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature;

[0041] The data analysis module is used to generate a target business vector corresponding to the target data according to the target sub-business characteristics corresponding to the business attributes, and perform data analysis according to the target business vector.

[0042] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the business processing method based on Gaussian error linear unit of the above-mentioned first aspect.

[0043] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the business processing method based on Gaussian error linear unit of the above-mentioned first aspect.

[0044] The business processing method based on Gaussian error linear unit, the business processing device based on Gaussian error linear unit, the electronic device and the computer-readable storage medium proposed in the present application obtain the target data of the target object for the target business, input the target data into the business processing model, and perform business processing on the target data based on the business processing model. The Gaussian error linear unit built into the business processing model involves a large number of nonlinear operations. In order to improve the business processing efficiency of the Gaussian error linear unit, the processing progress information of the business processing model on the target data is obtained. When the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model performs operations on the target data, the objective function corresponding to the Gaussian error linear unit is obtained, and the target data is operated based on the objective function, so as to improve the efficiency of data processing, and at the same time ensure that the model structure of the business processing model remains unchanged, so as to realize the reuse of model parameters in the business processing model. The business feature element obtained by the business processing model before the Gaussian error linear unit for the business attribute contained in the target data is obtained, so as to perform Gaussian error linear operation on the business feature element. In order to ensure the security of the target data, the business feature element is split to obtain the first sub-business feature element and the second sub-business feature element. The first sub-feature obtained by the first service node processing the first sub-business feature element according to the objective function and the second sub-feature obtained by the second service node processing the second sub-business feature element according to the objective function are obtained, and the sub-business feature elements are processed by two independent service nodes respectively, so as to avoid data security risks caused by data exchange between service nodes, and improve the processing efficiency of the Gaussian error linear unit while ensuring data security. According to the first sub-feature and the second sub-feature, the target sub-business feature corresponding to the business attribute is determined, and according to the target sub-business feature corresponding to the business attribute, the target business vector corresponding to the target data is generated, and data analysis is performed according to the target business vector to provide automated business services to the target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flowchart of a business processing method based on a Gaussian error linear unit provided in an embodiment of the present application;

[0046] Figure 2 yes Figure 1 Flow chart of step S130 in FIG.

[0047] Figure 3 is a function graph of the error function provided in the embodiment of the present application;

[0048] Figure 4 yes Figure 1 Flow chart of step S140 in FIG.

[0049] Figure 5 yes Figure 1Flow chart of step S150 in FIG.

[0050] Figure 6 yes Figure 5 Flow chart of step S520 in FIG.

[0051] Figure 7 yes Figure 6 Flow chart of step S630 in FIG.

[0052] Figure 8 yes Figure 1 Flow chart of step S160 in FIG.

[0053] Fig. 9 is a structural schematic diagram of a service processing device based on a Gaussian error linear unit provided in an embodiment of the present application;

[0054] Fig.10 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0058] With the development of information technology, business processing models can be used to analyze business data to automatically provide business services to target objects. For example, disease diagnosis models can be used to classify disease cases of target objects. However, in medical and other business fields, the business data of objects may have protection requirements, and the security of the business data of objects needs to be ensured during the data analysis process.

[0059] Related technologies use a business processing model to perform a secure multi-party computation (SMPC) algorithm to protect the privacy of business data. However, the business processing model usually involves a Gaussian Error Linear Unit (GELU), which contains a large number of nonlinear operations on SMPC, reducing the efficiency of business processing.

[0060] Based on this, the embodiments of the present application provide a business processing method based on a Gaussian error linear unit, a business processing device based on a Gaussian error linear unit, an electronic device and a computer-readable storage medium, aiming to improve the efficiency of business processing while ensuring the security of business data during the data analysis process.

[0061] The business processing method based on Gaussian error linear unit, the business processing device based on Gaussian error linear unit, the electronic device and the computer-readable storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the business processing method based on Gaussian error linear unit in the embodiments of the present application is described.

[0062] The business processing method based on Gaussian error linear unit provided in the embodiment of the present application relates to the field of data processing technology. The business processing method based on Gaussian error linear unit provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements a business processing method based on Gaussian error linear unit, etc., but is not limited to the above forms.

[0063] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0064] Figure 1 is an optional flowchart of a service processing method based on a Gaussian error linear unit provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S110 to S170.

[0065] Step S110, obtaining target data of a target object for a target business, where the target data includes at least one business attribute;

[0066] Step S120, inputting the target data into the business process model, and obtaining processing progress information of the target data by the business process model;

[0067] Step S130, when the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, obtain the objective function corresponding to the Gaussian error linear unit, and obtain the business feature element obtained by the business processing model before the Gaussian error linear unit processes the business attributes contained in the target data;

[0068] Step S140, splitting the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element;

[0069] Step S150, obtaining a first sub-feature obtained by processing the first sub-service feature element according to the objective function by the first service node, and obtaining a second sub-feature obtained by processing the second sub-service feature element according to the objective function by the second service node;

[0070] Step S160, determining a target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature;

[0071] Step S170: Generate a target business vector corresponding to the target data according to the target sub-business characteristics corresponding to the business attributes, and perform data analysis according to the target business vector.

[0072] In step S110 of some embodiments, the target object is a business object that produces business data. The target object may be a user object for the business service, or an object instance that participates in the business processing flow, such as an order object, an information retrieval object, etc. The target business is a business processing item under a preset business service scenario. The preset business service scenario may be order processing, disease classification and diagnosis, financial knowledge recommendation, etc. The target data is the business data generated by the target object for the target business. The target data may be order data, case data, financial knowledge query content, etc. The target data includes at least one business attribute, and the business attribute is used to describe the business characteristics of the target data at the target business level. If the target business is order processing, the business attribute may be an object attribute, a product attribute, and an order attribute, etc. The object attribute may be an object name, an object address, etc. The product attribute may be a product number, a product name, a product price, etc. The order attribute may be an order number, an order date, an order status, a payment method, etc. If the target business is disease classification and diagnosis, the business attribute may be clinical symptoms (fever, cough, etc.), object signs (body temperature, blood pressure, heart rate), etc.

[0073] In step S120 of some embodiments, the business processing model is a model for business processing of target data, which can be set according to the target business. The business processing model can be a model such as GPT, VIT, etc. based on the transformer architecture. If the target business is order processing, the business processing model can be an order processing model; if the target business is disease classification and diagnosis, the business processing model can be a disease diagnosis model. The target data is input into the business processing model for intelligent data reasoning, so as to provide automated business services to the target object using the data analysis results. The business processing model includes multiple network layers, and some network layers are built with Gaussian error linear units. In order to ensure that the model structure of the business processing model does not change and realize the reusability of the model parameters of each network layer, only when the network layer involves the operation of the Gaussian error linear unit, the operation is processed in a targeted manner, so that the business processing model can be flexibly deployed on each service node, so as to avoid the business processing model being unable to use the original model parameters due to the change of the model structure, and the original model parameters need to be retrained or fine-tuned to adapt to the new model structure.

[0074] In order to distinguish the normal processing flow of the network layer not involving nonlinear operation on the target data in the business processing model and the nonlinear operation flow of the Gaussian error linear unit, the processing progress information of the business processing model on the target data is obtained. The processing progress information is used to indicate the processing progress of the business processing model on the target data.

[0075] The Gaussian error linear unit is provided with a Gaussian error linear function, which is a GELU activation function, expressed as:

[0076]

[0077] Among them, x represents the function input parameter; ERF is the error function (Error Function, ERF).

[0078] The nonlinear calculation in the Gaussian error linear unit comes from the error function, which is expressed as:

[0079]

[0080] Among them, x represents the function input parameter, which is the upper limit of integration; t represents the independent variable of the integral operation.

[0081] In order to reduce the difficulty of nonlinear calculation and improve the data processing efficiency of Gaussian error linear unit, please refer to Figure 2 In some embodiments, step S130 may include but is not limited to steps S210 to S220:

[0082] Step S210, performing function fitting on the initial sub-error function to obtain a target sub-error function;

[0083] Step S220, updating the Gaussian error linear function according to the target sub-error function to obtain the target function.

[0084] In step S210 of some embodiments, when the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, the ERF is used as the initial sub-error function to obtain the function image of the initial sub-error function, and the function image is as follows: Figure 3 As shown in the function graph, it can be seen that the ERF function has valid values ​​only in the interval [-2.5, 2.5]. Based on the idea of ​​piecewise function, the independent variable interval of the initial sub-error function is divided into three segments: (-∞, 2.5), [-2.5, 2.5] and (2.5, +∞), and the initial sub-error function is rewritten as the following piecewise function form.

[0085]

[0086] Perform polynomial fitting on f(x) through the preset function library to obtain the fitting polynomial of the initial sub-error function in the interval [-2.5, 2.5]. The preset function library can be Python's numpy.polyfit library. The fitting polynomial is expressed as:

[0087] f(x)=-0.00316730430x 7 +0.0493278356x 5 -0.297453931x 3 +1.09952043x.

[0088] The target sub-error function is determined according to the function graph and the fitting polynomial, and the target sub-error function is a piecewise function.

[0089] In step S220 of some embodiments, the initial sub-error function in the Gaussian error linear function is replaced with the target sub-error function to obtain the target function. The target function is the Gaussian error linear function after function replacement, and the target function is a piecewise function.

[0090] In the above steps S210 to S220, the GELU activation function is approximated by a piecewise function, which avoids the complex nonlinear operation of the ERF function in the GELU activation function and improves the operation efficiency of the Gaussian error linear unit.

[0091] The business characteristic elements obtained by the business processing model before the Gaussian error linear unit for the business attributes contained in the target data are obtained, so as to perform accurate GELU function operations on the business characteristic elements through the target function. The business characteristic elements are the characteristic values ​​obtained by feature extraction of the business attributes. In order to further improve the efficiency of the GELU function operation and ensure the security of the target data, the business characteristic elements are split based on the secret sharing technology (SS) to obtain the first sub-business characteristic element and the second sub-business characteristic element, and two independent service nodes respectively operate on the first sub-business characteristic element and the second sub-business characteristic element. A service node refers to an independent server instance in a distributed server system that is responsible for processing client requests.

[0092] See also Figure 4 In some embodiments, step S140 may include but is not limited to steps S410 to S420:

[0093] Step S410, randomly generating reference feature elements;

[0094] Step S420: split the service characteristic element according to the reference characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element.

[0095] In step S410 of some embodiments, a random number is generated and used as a reference feature element.

[0096] In step S420 of some embodiments, the reference characteristic element is used as the first sub-service characteristic element, and the difference between the service characteristic element and the reference characteristic element is used as the second sub-service characteristic element. The sum of the first sub-service characteristic element and the second sub-service characteristic element is equal to the service characteristic element. If the reference characteristic element is represented by r and the service characteristic element is represented by x, then the first sub-service characteristic element is r and the second sub-service characteristic element is xr.

[0097] In the above steps S410 to S420, the business characteristic elements are split so as to subsequently process the split sub-business characteristic elements, thereby improving the processing efficiency of the business characteristic elements while ensuring the security of the business characteristic elements and the parameters of the business processing model.

[0098] As a mainstream privacy computing technology, multi-party secure computing can complete model reasoning tasks under the premise of protecting the privacy of model parameters and data. However, there are a large number of nonlinear operations on SMPC in the privacy-preserving GELU algorithm based on SMPC, which leads to low computational efficiency or insufficient computational accuracy. In order to improve the efficiency and accuracy of GELU function calculations, so that the business processing model can perform data reasoning tasks efficiently and accurately, the embodiment of the present application is jointly performed by the first service node and the second service node to perform calculations for Gaussian error linear units. The first service node performs privacy protection calculations on the first sub-service feature element locally at the first service node according to the objective function to obtain the first sub-feature. The first sub-feature is a feature obtained by performing GELU activation processing on the first sub-service feature element according to the objective function. The second service node performs privacy protection calculations on the second sub-service feature element locally at the second service node according to the objective function to obtain the second sub-feature. The second sub-feature is a feature obtained by performing GELU activation processing on the second sub-service feature element according to the objective function.

[0099] See also Figure 5 In some embodiments, step S150 may include but is not limited to steps S510 to S520:

[0100] Step S510, obtaining a privacy protection algorithm according to the objective function;

[0101] Step S520: The first service node calls a privacy protection algorithm to perform privacy calculation on the first sub-service feature element to obtain the first sub-feature.

[0102] In step S510 of some embodiments, a privacy protection algorithm is obtained according to the objective function, and the privacy protection algorithm includes a privacy protection multiplication algorithm, a privacy protection square algorithm, and a privacy protection comparison algorithm. The privacy protection multiplication algorithm is defined as Mul (·):[xy]←∏ Mul([x], [y]), where x and y are the two multipliers of the privacy-preserving multiplication operation. The privacy-preserving square algorithm is defined as ∏ Square (·):[x 2 ]←∏ Square ([x]), where x is the multiplier of the privacy-preserving square operation. The privacy-preserving comparison algorithm is defined as ∏ LT (·):[c]←∏ LT ([x],a), x and a are two data used for comparison.

[0103] In step S520 of some embodiments, the first service node calls the privacy protection multiplication algorithm, the privacy protection square algorithm, and the privacy protection comparison algorithm to perform privacy calculation on the first sub-service feature element to obtain the first sub-feature. It should be noted that the process of performing privacy protection calculation on the second sub-service feature element by the second service node to obtain the second sub-feature can refer to steps S510 to S520, which will not be repeated here.

[0104] In the above steps S510 to S520, the first service node performs privacy protection calculations on the first sub-service characteristic elements, thereby ensuring the data security of the first sub-service characteristic elements and improving the efficiency of the privacy protection calculations by utilizing the computing power resources of the first service node.

[0105] See also Figure 6 In some embodiments, step S520 may include but is not limited to steps S610 to S630:

[0106] Step S610, the first service node invokes a privacy protection algorithm to perform a privacy comparison calculation on the first sub-service characteristic element to obtain first privacy comparison information and second privacy comparison information;

[0107] Step S620, performing privacy multiplication calculation on the first sub-service characteristic element to obtain an initial sub-service characteristic;

[0108] Step S630: Determine a first sub-feature according to the first privacy comparison information, the second privacy comparison information and the initial sub-service feature.

[0109] In step S610 of some embodiments, referring to the objective function, the first service node calls the privacy protection comparison algorithm to perform a privacy comparison calculation on the first sub-service feature element and the first preset value to obtain first privacy comparison information, and calls the privacy protection comparison algorithm to perform a privacy comparison calculation on the first sub-service feature element and the second preset value to obtain second privacy comparison information. The first preset value is -2.5, and the second preset value is 2.5. The first privacy comparison information is information output by the privacy protection comparison algorithm when the first sub-service feature element is less than the first preset value. The second privacy comparison information is information output by the privacy protection comparison algorithm when the first sub-service feature element is greater than the second preset value.

[0110] Call ∏ LT (·) Calculate [c0] = ∏ LT ([x0],-2.5), [c1]=∏ LT ([x0], 2.5), c0 = (x0 < -2.5), c1 = (x0 > 2.5). Wherein, c0 is the first privacy comparison information, c1 is the second privacy comparison information, and x0 is the first sub-service characteristic element.

[0111] In step S620 of some embodiments, referring to the fitting polynomial f(x) in the objective function, the privacy-preserving square algorithm is called to perform privacy square calculation on the first sub-service characteristic element to obtain a quadratic characteristic element. The privacy-preserving multiplication algorithm is called to perform privacy multiplication calculation on the first sub-service characteristic element and the quadratic characteristic element to obtain a cubic characteristic element. The privacy-preserving multiplication algorithm is called to perform privacy multiplication calculation on the quadratic characteristic element and the cubic characteristic element to obtain a quintic characteristic element. The privacy-preserving multiplication algorithm is called to perform privacy multiplication calculation on the quadratic characteristic element and the quintic characteristic element to obtain a septad characteristic element. The quadratic characteristic element, the cubic characteristic element, the quintic characteristic element and the septad characteristic element are used as initial sub-service characteristics. The quadratic characteristic element, the cubic characteristic element, the quintic characteristic element and the septad characteristic element can be regarded as the square, the cube, the quintic and the septad characteristic element of the first sub-service, respectively.

[0112] Specifically, calling ∏ Square (·) Calculate [x0 2 ]=∏ Square ([x0]), x0 2 is a secondary characteristic element. Call ∏ Mul (·) Calculate [x0 3 ]=∏ Mul ([x0 2 ],[x0]),[x0 5 ]=∏ Mil ([x0 2 ],[x0 3 ]),[x0 7]=∏ Mul ([x0 2 ],[x0 5 ]).x0 3 、x0 5 、x0 7 They are cubic, quintic and septic elements respectively.

[0113] In step S630 of some embodiments, a privacy calculation is performed on the first privacy comparison information, the second privacy comparison information, and the initial sub-service feature to obtain the first sub-feature.

[0114] In the above steps S610 to S630, by calling the privacy protection algorithm to perform privacy protection calculation on the first sub-service characteristic element, not only the data security of the first sub-service characteristic element is guaranteed, but also the GELU activation calculation of the first sub-service characteristic element is realized.

[0115] See also Figure 7 In some embodiments, step S630 may include but is not limited to steps S710 to S740:

[0116] Step S710, determining target privacy comparison information according to the first privacy comparison information and the second privacy comparison information;

[0117] Step S720, performing error calculation on the initial sub-service feature to obtain an error feature;

[0118] Step S730, performing privacy multiplication calculation on the target privacy comparison information and the error feature to obtain the intermediate sub-service feature;

[0119] Step S740: Determine the first sub-feature according to the first privacy comparison information, the second privacy comparison information and the intermediate sub-service feature.

[0120] In step S710 of some embodiments, the second privacy comparison information is subtracted from the first privacy comparison information to obtain the target privacy comparison information. If the first privacy comparison information is represented as [c0], the second privacy comparison information is represented as [c1], and the target privacy comparison information is represented as [z 0,1 ], then [z 0,1 ]=[c1]-[c0].

[0121] In step S720 of some embodiments, the initial sub-service feature and the first sub-service feature element are substituted into the target sub-error function to perform error calculation to obtain an error feature. The error feature is the output value of the target sub-error function, that is, e=ERF(x).

[0122] In step S730 of some embodiments, a privacy-preserving multiplication algorithm is called to perform privacy multiplication calculation on the target privacy comparison information and the error feature to obtain an intermediate sub-service feature. The intermediate sub-service feature is the output feature of the privacy-preserving multiplication algorithm, represented by Π Mul ([z 0,1 ],[e]).

[0123] In step S740 of some embodiments, the first privacy comparison information is used as the first intermediate feature, the difference obtained by subtracting 1 from the second privacy comparison information is used as the second intermediate feature, the intermediate sub-service feature and the second intermediate feature are summed, and the feature obtained by summing and the first intermediate feature are calculated to obtain the first sub-feature. 0,0 ]=[c0],[z 0,1 ]=[c1]-[c0],[z 0,2 ]=1-[c1],z 0,0 is the first intermediate feature, z 0,1 For the target privacy comparison information, z 0,2 is the second intermediate feature, then the first sub-feature is expressed as:

[0124] [g]0=-[z 0,0 ]-∏ Mul ([z 0,1 ],[e])+[z 0,2 ].

[0125] Through the above steps S710 to S740, not only can the accurate calculation of the GELU function be achieved, but also the various data involved in the calculation process can be protected based on the privacy protection mechanism, thereby improving the security of business processing.

[0126] See also Figure 8 In some embodiments, step S160 may include but is not limited to steps S810 to S820:

[0127] Step S810, adding the first sub-feature and the second sub-feature to obtain a reference sub-service feature;

[0128] Step S820, performing modulo processing on the reference sub-service feature to obtain the target sub-service feature.

[0129] In step S810 of some embodiments, neither the first sub-feature alone nor the second sub-feature alone can reveal complete feature information. In order to restore the complete service feature information after processing by the Gaussian error linear unit, the first sub-feature [g]0 calculated by the first service node and the second sub-feature [g]1 calculated by the second service node are added to obtain a benchmark sub-service feature of [g]0+[g]1.

[0130] In step S820 of some embodiments, it is determined that the integer ring L=2 32 (2 64 ), modulo the reference sub-service feature according to the integer ring L to obtain the target sub-service feature corresponding to the service attribute.

[0131] The above steps S810 to S820 can restore the complete business features by integrating the sub-features obtained by the privacy protection calculations performed by the two service nodes, so as to perform data analysis based on the complete business features and provide intelligent business services to the target object.

[0132] In step S170 of some embodiments, a feature vector is constructed based on the target sub-business characteristics of each business attribute to obtain a target business vector corresponding to the target data, and data analysis is performed on the target business vector after the Gaussian error linear unit through the business processing model to achieve various tasks such as disease classification and diagnosis, financial knowledge recommendation, and order processing.

[0133] The business processing method based on Gaussian error linear unit in the embodiment of the present application can perform secure calculations on the input additive shares [x]0, [x]1 based on the existing privacy-preserving multiplication algorithm, privacy-preserving square algorithm and privacy-preserving comparison algorithm under a semi-honest model to obtain [g]0, [g]1, and satisfy ([g]0+[g]1)mod L=GELU(x), where x is a business feature element, [g]0 is the first sub-feature, [g]1 is the second sub-feature, [x]0 and [x]1 are the first sub-business feature element and the second sub-business feature element after the business feature element is split, respectively. In the privacy-preserving calculation process, the business processing method based on Gaussian error linear unit in the embodiment of the present application needs to call the privacy-preserving comparison algorithm twice and the privacy-preserving multiplication algorithm four times in total. The method only needs to perform 4+2logL rounds of communication at least, where L represents an integer ring.

[0134] From the three aspects of time overhead, communication overhead and accuracy, the business processing method based on Gaussian error linear unit of the embodiment of the present application is compared with Crypten algorithm and Puma algorithm respectively. Crypten and Puma are both privacy-preserving computing frameworks, in which Crypten uses Taylor expansion to realize the approximate calculation of ERF function, and Puma uses piecewise polynomial to directly approximate GELU function. The embodiment of the present application is based on the privacy-preserving machine learning open source framework Crypten for experimentation. The experimental environment is three Tesla V100 servers, and the bandwidth between servers is 10GB / s. The comparison results of the calculation time and communication volume of the three methods are shown in Table 1, and the comparison results of the accuracy are shown in Table 2.

[0135] Table 1

[0136]

[0137] Table 2

[0138]

[0139] As can be seen from Table 1, the business processing method based on Gaussian error linear unit in the embodiment of the present application is superior to Puma in both time overhead and communication overhead, but slightly higher than Crypten. X is the communication volume and time obtained by performing privacy protection calculations using the method of the embodiment of the present application. Taking the input dimension of 100×100 as an example, 1.46X in the column of the present application means that the communication volume of nonlinear operations on business feature elements using the GELU activation function is 1.46 times the communication volume of the method of the embodiment of the present application, and 1.29X means that the time spent on nonlinear operations on business feature elements using the GELU activation function is 1.29 times the time spent by the method of the embodiment of the present application. As can be seen from Table 2, the error mean and error variance of the embodiment of the present application are both small, and the accuracy of privacy protection calculations is high. In terms of accuracy, the Crypten algorithm can only maintain a high accuracy within a very small feature input range, namely [-1,1], which obviously cannot meet the needs of practical applications. The method of the embodiment of the present application and the Puma algorithm can realize accurate privacy protection calculations of the GELU function under any input range. Furthermore, the calculation accuracy of the method in the embodiment of the present application is higher than that of the Puma algorithm in a small input range.

[0140] See also Fig. 9 The embodiment of the present application further provides a service processing device based on a Gaussian error linear unit, which can implement the above-mentioned service processing method based on a Gaussian error linear unit. The service processing device based on a Gaussian error linear unit includes:

[0141] A first acquisition module 910 is used to acquire target data of a target object for a target business, where the target data includes at least one business attribute;

[0142] An input module 920 is used to input the target data into the business process model and obtain the processing progress information of the target data by the business process model;

[0143] The second acquisition module 930 is used to obtain the target function corresponding to the Gaussian error linear unit when the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, and obtain the business feature element obtained by the business processing model before the Gaussian error linear unit processes the business attributes contained in the target data;

[0144] A splitting module 940 is used to split the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element;

[0145] A third acquisition module 950 is used to acquire a first sub-feature obtained by the first service node processing the first sub-service feature element according to the objective function, and to acquire a second sub-feature obtained by the second service node processing the second sub-service feature element according to the objective function;

[0146] A determination module 960, configured to determine a target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature;

[0147] The data analysis module 970 is used to generate a target business vector corresponding to the target data according to the target sub-business characteristics corresponding to the business attributes, and perform data analysis according to the target business vector.

[0148] The specific implementation of the service processing device based on the Gaussian error linear unit is basically the same as the specific implementation of the service processing method based on the Gaussian error linear unit mentioned above, and will not be repeated here.

[0149] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned business processing method based on Gaussian error linear unit when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0150] See also Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0151] The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0152] The memory 1020 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1020 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1020, and the processor 1010 calls and executes the business processing method based on the Gaussian error linear unit in the embodiment of the present application;

[0153] Input / output interface 1030, used to implement information input and output;

[0154] Communication interface 1040, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0155] Bus 1050 , which transmits information between various components of the device (e.g., processor 1010 , memory 1020 , input / output interface 1030 , and communication interface 1040 );

[0156] The processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 are connected to each other in communication within the device via a bus 1050 .

[0157] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the business processing method based on the Gaussian error linear unit is implemented.

[0158] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor 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.

[0159] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0160] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0161] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0162] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0163] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0164] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0165] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0166] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0167] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0168] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0169] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A business processing method based on Gaussian error linear unit, characterized in that: The method comprises: Acquire target data of a target object for a target business, wherein the target data includes at least one business attribute; Inputting the target data into a business processing model, and obtaining processing progress information of the target data by the business processing model; When the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, obtaining the objective function corresponding to the Gaussian error linear unit, and obtaining the business feature elements obtained by the business processing model before the Gaussian error linear unit processes the business attributes contained in the target data; Splitting the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element; Acquire a first sub-feature obtained by processing the first sub-service feature element by the first service node according to the objective function, and acquire a second sub-feature obtained by processing the second sub-service feature element by the second service node according to the objective function; Determining a target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature; A target business vector corresponding to the target data is generated according to the target sub-business characteristics corresponding to the business attributes, and data analysis is performed according to the target business vector.

2. The service processing method based on Gaussian error linear unit according to claim 1, characterized in that: The obtaining of the first sub-feature obtained by the first service node processing the first sub-service feature element according to the objective function includes: Acquire a privacy protection algorithm according to the objective function; The first service node calls the privacy protection algorithm to perform privacy calculation on the first sub-service feature element to obtain the first sub-feature.

3. The service processing method based on Gaussian error linear unit according to claim 2, characterized in that: The step of performing privacy calculation on the first sub-service feature element by invoking the privacy protection algorithm through the first service node to obtain the first sub-feature includes: Invoking the privacy protection algorithm by the first service node to perform a privacy comparison calculation on the first sub-service feature element to obtain first privacy comparison information and second privacy comparison information; Performing privacy multiplication calculation on the first sub-service characteristic element to obtain an initial sub-service characteristic; The first sub-feature is determined according to the first privacy comparison information, the second privacy comparison information, and the initial sub-service feature.

4. The service processing method based on Gaussian error linear unit according to claim 3, characterized in that: The determining the first sub-feature according to the first privacy comparison information, the second privacy comparison information, and the initial sub-service feature includes: Determine target privacy comparison information according to the first privacy comparison information and the second privacy comparison information; Performing error calculation on the initial sub-service characteristic to obtain an error characteristic; Performing privacy multiplication calculation on the target privacy comparison information and the error feature to obtain an intermediate sub-service feature; The first sub-feature is determined according to the first privacy comparison information, the second privacy comparison information, and the intermediate sub-service feature.

5. The service processing method based on Gaussian error linear unit according to any one of claims 1 to 4, characterized in that: The Gaussian error linear unit is provided with a Gaussian error linear function, the Gaussian error linear function includes an initial sub-error function, and obtaining the objective function corresponding to the Gaussian error linear unit includes: Performing function fitting on the initial sub-error function to obtain a target sub-error function; The Gaussian error linear function is updated according to the target sub-error function to obtain the target function.

6. The service processing method based on Gaussian error linear unit according to any one of claims 1 to 4, characterized in that: The splitting of the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element includes: Randomly generate reference feature elements; The service characteristic element is split according to the reference characteristic element to obtain the first sub-service characteristic element and the second sub-service characteristic element.

7. The service processing method based on Gaussian error linear unit according to any one of claims 1 to 4, characterized in that: The determining, according to the first sub-feature and the second sub-feature, a target sub-service feature corresponding to the service attribute includes: Adding the first sub-feature and the second sub-feature to obtain a reference sub-service feature; The reference sub-service characteristic is subjected to modulo processing to obtain the target sub-service characteristic.

8. A service processing device based on Gaussian error linear unit, characterized in that: The device comprises: A first acquisition module is used to acquire target data of a target object for a target business, wherein the target data includes at least one business attribute; An input module, used to input the target data into the business processing model and obtain processing progress information of the target data by the business processing model; A second acquisition module is used to obtain, when the processing progress indicated by the processing progress information is that the Gaussian error linear unit built into the business processing model operates on the target data, an objective function corresponding to the Gaussian error linear unit, and obtain a business feature element obtained by the business processing model before the Gaussian error linear unit processes the business attributes contained in the target data; A splitting module, used for splitting the service characteristic element to obtain a first sub-service characteristic element and a second sub-service characteristic element; A third acquisition module is used to acquire a first sub-feature obtained by the first service node processing the first sub-service feature element according to the objective function, and to acquire a second sub-feature obtained by the second service node processing the second sub-service feature element according to the objective function; A determination module, configured to determine a target sub-service feature corresponding to the service attribute according to the first sub-feature and the second sub-feature; The data analysis module is used to generate a target business vector corresponding to the target data according to the target sub-business characteristics corresponding to the business attributes, and perform data analysis according to the target business vector.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the business processing method based on Gaussian error linear unit according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the business processing method based on Gaussian error linear unit according to any one of claims 1 to 7 is implemented.

Images