A business execution method and device based on longitudinal federal logistic regression

Abstract

The application discloses a business execution method and device based on longitudinal federal logistic regression, which is realized based on a first participant and n second participants, and the method comprises the following steps: a first participant obtains a first intermediate parameter according to current first business data characteristics and parameters of a first model deployed on the first participant and pre-trained; a second participant obtains a second intermediate parameter according to current second business data characteristics and parameters of a second model deployed on the second participant and pre-trained, obtains a first mixed intermediate parameter according to the second intermediate parameter, and sends the first mixed intermediate parameter to the first participant; the first participant sums the first intermediate parameter and all the first mixed intermediate parameters to obtain an intermediate parameter total sum; the intermediate parameter total sum is processed by a softmax function to obtain a probability distribution of target categories, and K target categories with the highest probabilities are taken as business execution targets.

Description

Technical Field

[0001] This invention relates to the field of federated learning technology, and in particular to a business execution method and apparatus based on vertical federated logistic regression. Background Technology

[0002] With the widespread adoption of the internet and mobile devices, various industries have accumulated massive amounts of business data. This data is held by different entities, and due to commercial confidentiality and legal regulations regarding data security, data sharing and integration between these entities are rare, forming isolated data silos. Meanwhile, AI and big data technologies have demonstrated immense value in many fields, such as clinical auxiliary diagnosis, new drug development, and precision medicine in the medical industry; intelligent marketing and targeted advertising in the digital advertising industry; and facial recognition and voiceprint recognition in the security industry. These AI models can only achieve good accuracy and play a crucial role in actual business operations after being trained on large amounts of data.

[0003] Federated learning is essentially a distributed machine learning framework that enables data sharing and collaborative modeling while ensuring data privacy, security, and legal compliance. Its core idea is that when multiple data sources participate in model training, the original data does not need to be transferred; instead, the model is trained jointly only by exchanging intermediate parameters, and the original data does not need to leave the local machine. This approach achieves a balance between data privacy protection and data sharing analysis, resulting in a data application model that is "usable but not visible."

[0004] Logistic regression models are widely used in industries such as finance, healthcare, and e-commerce due to their advantages of concise and efficient parameters, strong interpretability, and robustness. Currently, most privacy-preserving methods for longitudinal federated logistic regression are based on homomorphic encryption. Homomorphic encryption allows addition and multiplication operations on the encrypted ciphertext, and the decrypted result is the same as the plaintext calculation result. However, the encryption and decryption process requires exponentiation of very large integers, resulting in extremely high computation time. The loss function of logistic regression includes exponentiation, while homomorphic encryption methods do not yet support exponentiation of ciphertext. A review of current research shows that longitudinal logistic regression based on homomorphic encryption approximates the loss function using Taylor expansion, and is limited to binary classification cases.

[0005] Homomorphic encryption-based vertical federated logistic regression is difficult to implement in real-world applications due to its excessive computation time and limitation to binary classification. There is an urgent need for a method that can perform business operations while ensuring privacy. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a business execution method and apparatus based on vertical federated logical regression.

[0007] In a first aspect, embodiments of the present invention provide a business execution method based on vertical federated logistic regression, implemented based on a first participant and n second participants, the method comprising:

[0008] The first participant obtains the first intermediate parameters based on the current first business data characteristics and the parameters of the first model pre-trained on the first participant; the second participant obtains the second intermediate parameters based on the current second business data characteristics and the parameters of the second model pre-trained on the second participant, obtains the first obfuscated intermediate parameters based on the second intermediate parameters, and sends the first obfuscated intermediate parameters to the first participant;

[0009] The first participant sums the first intermediate parameter and all the first obfuscated intermediate parameters to obtain the total intermediate parameter; the sum of intermediate parameters is processed by the softmax function to obtain the probability distribution of the target categories, and the K target categories with the highest probabilities are selected as the business execution targets;

[0010] The process of training the first and second models includes:

[0011] A trusted third party generates random vectors and random numbers, which are then distributed to the first participant and all second participants.

[0012] The parameters of the first model deployed on the first participant and the second model deployed on each second participant are initialized according to the data feature dimensions.

[0013] The first participant obtains the first intermediate parameter based on the current first business data characteristics and the initialization parameters of the first model; the second participant obtains the second intermediate parameter based on the current second business data characteristics and the initialization parameters of the second model, obtains the first obfuscated intermediate parameter based on the second intermediate parameter, and sends the first obfuscated intermediate parameter to the first participant;

[0014] The first participant sums the first intermediate parameter and all first confusion intermediate parameters to obtain the total intermediate parameter; the total intermediate parameter is processed by the softmax function to obtain the probability distribution of the target category, and the probability distribution is processed according to the data label to obtain the gradient intermediate parameter;

[0015] The first participant obtains the gradient of the first node based on the current first business data characteristics and gradient intermediate parameters;

[0016] The second participant initiates a secure inner product request to the first participant, and obtains the gradient of the second node based on the random vector, random number, and gradient of the first node;

[0017] Update the parameters of the first model deployed on the first participant and the parameters of the second model deployed on the second participant based on the gradients of the first node and the second node, respectively.

[0018] Secondly, embodiments of the present invention provide a business execution apparatus based on vertical federated logistic regression, including a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned business execution method based on vertical federated logistic regression.

[0019] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the above-described business execution method based on vertical federated logical regression.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0021] 1. This invention obtains intermediate parameters through business data features, and uses the softmax function to process the sum of intermediate parameters to obtain the probability distribution of the target category. It can be extended to any number of multi-class classification functions, solving the problems of low computational efficiency and inability to support multi-class classification in the prior art.

[0022] 2. In the process of training the first model deployed on the first participant and the second model deployed on the second participant, this invention generates random vectors and random numbers. The second participant initiates a secure inner product request to the first participant, and obtains the node gradients based on the random vectors and random numbers, thereby updating the first model and the second model. This invention performs joint model training only through interactive intermediate parameters, and the original data does not need to leave the local machine, thus achieving data privacy protection and data sharing analysis. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a business execution method based on vertical federated logistic regression, provided for an embodiment of the present invention;

[0025] Figure 2 A schematic diagram illustrating the process of training the first model and the second model according to an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of the process for initializing the first model parameters and the second model parameters provided in an embodiment of the present invention;

[0027] Figure 4This is a schematic diagram of the process for obtaining the first obfuscation intermediate parameter provided in an embodiment of the present invention;

[0028] Figure 5 This is a schematic diagram of the process for obtaining intermediate gradient parameters provided in an embodiment of the present invention;

[0029] Figure 6 This is a schematic diagram of the process for performing a safe inner product according to an embodiment of the present invention;

[0030] Figure 7 This is a structural block diagram of a business execution device based on vertical federated logistic regression, provided for an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] It should be noted that, unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0033] like Figure 1 As shown, this embodiment of the invention provides a business execution method based on vertical federated logistic regression. It is implemented using a trusted third party, a first participant (guest) possessing data labels and partial data features, and n (n≥1) second participants (host) possessing partial data features. The second participants are denoted as host1, host2, ..., host. i ..., host n , i∈[1,n], where n is a positive integer, the method specifically includes:

[0034] Step S1: The first participant (guest) obtains the first intermediate parameter based on the current first business data features and the parameters of the first model pre-trained on the first participant; the second participant obtains the second intermediate parameter based on the current second business data features and the parameters of the second model pre-trained on the second participant (host), obtains the first obfuscated intermediate parameter based on the second intermediate parameter, and sends the first obfuscated intermediate parameter to the first participant.

[0035] Step S2: The first participant, guest, sums the first intermediate parameter and all the first obfuscated intermediate parameters to obtain the total intermediate parameter sum; the total intermediate parameter sum is processed by the softmax function to obtain the probability distribution of the target category, and the K target categories with the highest probability are selected as the business execution targets, where K is a positive integer.

[0036] Among them, such as Figure 2 As shown, the process of training the first and second models includes:

[0037] Step S101: Generate the random vector and random number required for the secure inner product through a trusted third party, and send them to the first participant (guest) and all second participants (host).

[0038] n+1 random vectors are generated through a trusted third party. The length of each random vector is equal to the batch size B of the training data.

[0039] Let the first random vector be... .

[0040] Let the second random vector be... ; .

[0041] The first random number r is regenerated by a trusted third party.

[0042] Furthermore, in this example, the random vector Both the random number r and the random number r are generated by a uniformly distributed sampling function with upper and lower bounds of (-1, 1).

[0043] Calculate the second random number based on the first random vector, the second random vector, and the first random number. The expression is as follows:

[0044]

[0045] A trusted third party will use the first random vector The first random number r is sent to the first participant, guest, and the second random vector is sent. Second random number Send to the second participant host i , 1≤i≤n, where n is the number of second participants.

[0046] Step S102: Initialize the parameters of the first model deployed on the first participant and the second model deployed on each second participant according to the data feature dimensions.

[0047] Specifically, such as Figure 3 As shown, it includes:

[0048] Let the first business data feature dimension of the first participant be denoted as... Let the current second business data feature dimension of the i-th second participant be denoted as... .

[0049] Second participant, host i The second business data feature dimension Send to the first participant, guest.

[0050] The first participant (guest) on the first business data feature dimension and all second business data feature dimensions Summing yields the total dimension of the data features. The total dimension D of the data features is then sent to all second-party hosts.

[0051] Based on the total dimension of data characteristics The parameters of the first model deployed on the first participant are initialized, i.e., a random number matrix is ​​generated. , The size is , where c is the number of data categories, and a random number vector I is generated as the intercept vector, with length I being c. Random number matrix. The random numbers in the random number vector I are both bounded by upper and lower limits. The uniform distribution sampling function generated

[0052] Based on the total dimension of data characteristics The parameters of the second model deployed on the second participant are initialized, i.e., a random number matrix is ​​generated. ), The size is c represents the number of data categories, and the random number matrix is... The random numbers in the data are all determined by upper and lower bounds. It is generated by a uniformly distributed sampling function.

[0053] Step S103: The first participant, guest, obtains the first intermediate parameter based on the current first business data characteristics and the initialization parameters of the first model; the second participant, host, obtains the second intermediate parameter based on the current second business data characteristics and the initialization parameters of the second model, obtains the first obfuscated intermediate parameter based on the second intermediate parameter, and sends the first obfuscated intermediate parameter to the first participant, guest.

[0054] Specifically, such as Figure 4 As shown, it includes:

[0055] Let the first business data characteristic of the current batch of the first participant, guest, be: , The size is The first intermediate parameter is calculated using the following expression:

[0056]

[0057] In the formula, Let I be the initial model parameters, i.e., the random number matrix, and let I be the intercept vector. Size is B represents the batch size of the training data, and c represents the number of data categories.

[0058] Record the second participant, host i The data characteristics of the current batch are , The size is The second intermediate parameter is calculated using the following expression:

[0059]

[0060] In the formula, The initial model parameters are the random number matrix. Size is B represents the batch size of the training data, and c represents the number of data categories.

[0061] host i A third random vector is generated using a uniformly distributed sampling function with upper and lower bounds of (-1, 1). , The length is c.

[0062] Calculate the first confusion intermediate parameter The calculation method is as follows Subtract all elements in the j-th row The j-th element, i.e. The element in the j-th row and k-th column is and the first obfuscated intermediate parameter Send to the first participant, guest.

[0063] Step S104: The first participant (guest) sums the first intermediate parameter and all the first confusion intermediate parameters to obtain the total intermediate parameter; the total intermediate parameter is processed by the softmax function to obtain the probability distribution of the target category; the probability distribution is processed according to the data label to obtain the gradient intermediate parameter;

[0064] Specifically, such as Figure 5 As shown, it includes:

[0065] The first participant (guest) sums the first intermediate parameter with the obfuscated intermediate parameters uploaded by all second participants (hosts) to obtain the total intermediate parameters, as shown in the following expression:

[0066]

[0067] In the formula, This is the sum of intermediate parameters. As the first intermediate parameter, The obfuscated intermediate parameters uploaded by the i-th second participant host.

[0068] Calculate the probability distribution of the target category using the softmax function. The size of P is B is the batch size of the training data, c is the number of data categories, and the element in the i-th row and j-th column of P is... ,in, for The element in the i-th row and j-th column.

[0069] The first participant, guest, holds the data tag for the current batch. , Let B be a vector of length B, where vector B is a vector of length B. Each element in the array is an integer ranging from 1 to c, representing which category the data belongs to.

[0070] Based on the probability distribution of the target category and the data labels held by the first participant in the current batch, the intermediate parameters of the gradient are calculated as follows:

[0071]

[0072] In the formula, This represents one-hot encoding, where the size of H is... The element in the i-th row and j-th column of H is ,in , for The i-th element.

[0073] Step S105: The first participant, guest, obtains the gradient of the first node based on the data features and intermediate gradient parameters of the current batch, as shown in the following expression:

[0074] H

[0075] In the formula, The gradient of the first node. Based on the data characteristics of the current batch, the first participant, the guest, Size is , Here, c represents the first business data feature dimension corresponding to the first participant, guest; c represents the number of data categories; and H represents the intermediate gradient parameter.

[0076] Step S106: All second-participant hosts initiate a secure inner product request to the first-participant guest. Based on the secure inner product protocol, the gradient of the second node corresponding to the second-participant host is calculated.

[0077] Specifically, such as Figure 6 As shown, it includes:

[0078] Second participant, host i Based on the data characteristics of the current batch Using a second random vector, calculate the second confusion intermediate parameter. The calculation method is as follows Subtract all elements in the j-th row The j-th element, i.e. The element in the j-th row and k-th column is The second obfuscated intermediate parameter Send to the first participant, guest.

[0079] The first participant, the guest, calculates the third confusion intermediate parameter based on the gradient intermediate parameter and the first random vector. The calculation method is to subtract the j-th row element from H. The j-th element, i.e. The element in the j-th row and k-th column is The third obfuscated intermediate parameter Send to all second-party hosts.

[0080] The first participant, guest, is based on the second obfuscated intermediate parameter. Gradient intermediate parameters and the first random number Calculate the first inner product share , Size is The first inner product share Send to the second participant host i .

[0081] Second participant, host i Based on the third obfuscation intermediate parameter Second random vector Second random number Calculate the second inner product share , Given a vector of length c, the gradient of the second node is obtained by superimposing the first inner product share and the second inner product share. The calculation method is as follows Add the elements in the k-th column The kth element, i.e. The element in the j-th row and k-th column is , Size is .

[0082] Step S107: Set the learning rate, update the parameters of the model deployed on the first participant according to the gradient of the first node, and update the parameters of the model deployed on the second participant according to the gradient of the second node.

[0083] Set learning rate ;

[0084] The parameters of the first model deployed on the first participant are updated based on the gradient of the first node, as shown in the following expression:

[0085]

[0086] In the formula, B represents the gradient of the first node, and B represents the size of the current data batch.

[0087] The parameters of the second model deployed on the second participant are updated based on the gradient of the second node, as expressed below:

[0088]

[0089] In the formula, This is the gradient of the current second node.

[0090] The process of updating the parameters of the first model deployed on the first participant based on the gradient of the first node also includes:

[0091] Update the intercept parameter using the following expression:

[0092]

[0093] In the formula, It is the element in the j-th row and i-th column of the current gradient intermediate parameter H.

[0094] Step S108: Repeat steps S103 to S107 until the preset maximum number of training updates is reached.

[0095] Example 1

[0096] Embodiment 1 of this invention will detail the specific calculation steps of the privacy-preserving method based on secure inner product vertical federated logistic regression in the context of precision marketing in digital advertising.

[0097] In the digital advertising industry, precision marketing refers to delivering ads that are most relevant to each user. Compared to randomly placed ads, this significantly increases click-through rates and improves the user experience. With the development of AI technology, more and more advertising platforms are using techniques such as logistic regression to train marketing models for ad delivery. However, a major challenge in training these models is that advertising platforms can only use their own data and lack data from the brands that deliver the ads. This makes it difficult to accurately determine user interests and preferences, hindering the development of high-quality marketing models.

[0098] This embodiment includes an advertising platform and n brand owners ( The advertising platform possesses data dimensions such as customer age, gender, platform usage time, ad click type, and ad click count. Brands possess user purchase data for their brand, including purchase frequency, product type, product price, and total spending. The goal is to collaborate with the advertising platform and brands to train a marketing model that identifies the brands users are most interested in, thereby enabling precise ad targeting.

[0099] In the following steps, the nodes participating in the training from the advertising platform are denoted as "guest," and the nodes participating in the training from the brand are denoted as "host." The n brands are numbered from 1 to n, and the node of the i-th brand is denoted as "host." i .

[0100] Step S1: The advertising platform obtains the first intermediate parameters based on the current first business data features and the parameters of the first model pre-trained on the advertising platform; the brand obtains the second intermediate parameters based on the current second business data features and the parameters of the second model pre-trained on the brand, obtains the first obfuscated intermediate parameters based on the second intermediate parameters, and sends the first obfuscated intermediate parameters to the advertising platform.

[0101] In step S2, the advertising platform sums the first intermediate parameter and all the first obfuscation intermediate parameters to obtain the total intermediate parameter; it then processes the total intermediate parameter using the softmax function to obtain the probability distribution of the target categories, and selects the K target categories with the highest probabilities as the business execution targets for brand recommendation, thereby accurately delivering advertisements.

[0102] The process of training the first and second models includes:

[0103] In step S101, a trusted third party generates a random vector and random numbers, which are then distributed to the advertising platform and all brands.

[0104] Step S102: Initialize the parameters of the first model deployed on the advertising platform and the second model deployed on each brand according to the data feature dimensions;

[0105] In step S103, the advertising platform obtains the first intermediate parameter based on the current first business data characteristics and the initialization parameters of the first model; the brand obtains the second intermediate parameter based on the current second business data characteristics and the initialization parameters of the second model, obtains the first obfuscated intermediate parameter based on the second intermediate parameter, and sends the first obfuscated intermediate parameter to the advertising platform.

[0106] Step S104: The advertising platform sums the first intermediate parameter and all first confusion intermediate parameters to obtain the sum of intermediate parameters; it processes the sum of intermediate parameters using the softmax function to obtain the probability distribution of the target category, processes the probability distribution according to the data labels, and obtains the gradient intermediate parameters.

[0107] Step S105: The advertising platform obtains the gradient of the first node based on the current first business data features and gradient intermediate parameters;

[0108] Step S106: The brand initiates a secure inner product request to the advertising platform and obtains the gradient of the second node based on the random vector, random number, and gradient of the first node.

[0109] Step S107: Update the parameters of the first model deployed on the advertising platform and the parameters of the second model deployed on the brand according to the gradient of the first node and the gradient of the second node, respectively.

[0110] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 This provides a business execution method based on vertical federated logistic regression.

[0111] The present invention also provides Figure 7 The diagram shows a schematic structure of a business execution device based on vertical federated logistic regression. Figure 7 At the hardware level, the device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for the business logic. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 The described business execution method is based on vertical federated logical regression. Of course, besides software implementation, this invention does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution entity of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0112] Improvements in a technology can be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many improvements to the methodology can now be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that an improvement in methodology cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0113] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0114] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0115] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0116] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0121] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0122] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0123] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0124] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0126] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0127] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A business execution method based on vertical federated logistic regression, characterized in that, Based on one first participant and n second participants, the method includes: The first participant obtains a first intermediate parameter based on the current first business data features and the parameters of a pre-trained first model deployed on the first participant; the second participant obtains a second intermediate parameter based on the current second business data features and the parameters of a pre-trained second model deployed on the second participant, obtains a first obfuscated intermediate parameter based on the second intermediate parameter, and sends the first obfuscated intermediate parameter to the first participant; including: the first participant uses the intercept vector of the product of the current first business data features and the parameters of the pre-trained first model as the first intermediate parameter; the second participant uses the product of the current second business data features and the parameters of the pre-trained second model as the second intermediate parameter; a third random vector is generated; the difference between the second intermediate parameter and the third random vector is used as the first obfuscated intermediate parameter, and the first obfuscated intermediate parameter is sent to the first participant; The first participant sums the first intermediate parameter and all the first obfuscated intermediate parameters to obtain the total intermediate parameter; the sum of intermediate parameters is processed by the softmax function to obtain the probability distribution of the target categories, and the K target categories with the highest probabilities are selected as the business execution targets; The process of training the first and second models includes: A trusted third party generates random vectors and random numbers, which are then distributed to the first participant and all second participants. The parameters of the first model deployed on the first participant and the second model deployed on each second participant are initialized according to the data feature dimensions. The first participant obtains a first intermediate parameter based on the current first business data features and the initialization parameters of the first model; the second participant obtains a second intermediate parameter based on the current second business data features and the initialization parameters of the second model, obtains a first obfuscated intermediate parameter based on the second intermediate parameter, and sends the first obfuscated intermediate parameter to the first participant; including: the first participant uses the intercept vector of the product of the current first business data features and the current parameters of the first model as the first intermediate parameter; the second participant uses the product of the current second business data features and the current parameters of the second model as the second intermediate parameter; a third random vector is generated; the difference between the second intermediate parameter and the third random vector is used as the first obfuscated intermediate parameter, and the first obfuscated intermediate parameter is sent to the first participant; The first participant sums the first intermediate parameter and all first confusion intermediate parameters to obtain the total intermediate parameter; the total intermediate parameter is processed by the softmax function to obtain the probability distribution of the target category; the data label corresponding to the first business data feature of the first participant is one-hot encoded; the gradient intermediate parameter is calculated based on the probability distribution of the target category and the one-hot encoded data label to obtain the gradient intermediate parameter. The first participant obtains the gradient of the first node based on the current first business data characteristics and gradient intermediate parameters; The second participant initiates a secure inner product request to the first participant, and obtains the gradient of the second node based on the random vector, random number, and gradient of the first node; including: the second participant uses the difference between the current second business data feature and the second random vector as the second obfuscation intermediate parameter; and sends the second obfuscation intermediate parameter to the first participant; the first participant uses the sum of the gradient intermediate parameter and the first random vector as the third obfuscation intermediate parameter; the first participant uses the difference between the product of the second obfuscation intermediate parameter and the gradient intermediate parameter and the first random number as the first inner product share; and sends the first inner product share to the second participant; the second participant uses the difference between the product of the second random vector and the third obfuscation intermediate parameter and the second random number as the second inner product share; and the sum of the first inner product share and the second inner product share as the gradient of the second node; Update the parameters of the first model deployed on the first participant and the parameters of the second model deployed on the second participant based on the gradients of the first node and the second node, respectively.

2. The business execution method based on vertical federated logistic regression according to claim 1, characterized in that, A trusted third party generates random vectors and random numbers, which are then distributed to the first participant and all second participants, including: Obtain the first random vector ; Obtain the second random vector ; ; Get the first random number r; Calculate n second random numbers based on the first random vector, the second random vector, and the random number. The expression is as follows: ； The first random vector The first random number r is sent to the first participant, and the second random vector is sent to the first participant. Second random number Send it to the i-th second participant, 1≤i≤n, where n is the number of second participants.

3. The business execution method based on vertical federated logistic regression according to claim 1, characterized in that, The parameters of the first model deployed on the first participant and the second model deployed on each second participant are initialized according to the data feature dimensions, including: Obtain the first business data feature dimension of the first participant. ; Obtain the current second business data feature dimension of the i-th second participant. ; The second participant will use the second business data feature dimension. Send to the first participant; the first participant selects the first business data feature dimension. and all second business data feature dimensions Summation yields the total dimension of the data features; The parameters of the first model deployed on the first participant are initialized based on the total dimension of the data features, and a random number vector is generated as the intercept parameter. The parameters of the second model deployed on the second participant are initialized based on the total dimension of the data features.

4. The business execution method based on vertical federated logistic regression according to claim 1, characterized in that, The parameters of the first model deployed on the first participant and the parameters of the second model deployed on the second participant are updated based on the gradients of the first and second nodes, respectively, including: Set learning rate ; The parameters of the first model deployed on the first participant are updated based on the gradient of the first node, as shown in the following expression: ； In the formula, B is the gradient of the first node, and B is the size of the current data batch. The parameters of the second model deployed on the second participant are updated based on the gradient of the second node, as shown in the following expression: ； In the formula, This is the gradient of the current second node.

5. The business execution method based on vertical federated logistic regression according to claim 4, characterized in that, Updating the parameters of the first model deployed on the first participant based on the gradient of the first node also includes: Update the intercept parameter using the following expression: ； In the formula, It is the element in the j-th row and i-th column of the current gradient intermediate parameter H.

6. A business execution device based on vertical federated logistic regression, characterized in that, It includes a memory and one or more processors, wherein the memory stores executable code, and the one or more processors execute the executable code to implement the business execution method based on vertical federated logistic regression as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the business execution method based on vertical federated logistic regression as described in any one of claims 1-5.

Images