CN115130121A - Method for training longitudinal logistic regression model under privacy calculation of third party - Google Patents
Method for training longitudinal logistic regression model under privacy calculation of third party Download PDFInfo
- Publication number
- CN115130121A CN115130121A CN202210650114.5A CN202210650114A CN115130121A CN 115130121 A CN115130121 A CN 115130121A CN 202210650114 A CN202210650114 A CN 202210650114A CN 115130121 A CN115130121 A CN 115130121A
- Authority
- CN
- China
- Prior art keywords
- value
- data
- data value
- ciphertext
- gradient
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/008—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols involving homomorphic encryption
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2211/00—Indexing scheme relating to details of data-processing equipment not covered by groups G06F3/00 - G06F13/00
- G06F2211/007—Encryption, En-/decode, En-/decipher, En-/decypher, Scramble, (De-)compress
- G06F2211/008—Public Key, Asymmetric Key, Asymmetric Encryption
Abstract
The invention discloses a longitudinal logistic regression model training method under privacy calculation of a third party, which comprises the steps of obtaining first user characteristic data in the privacy calculation and calculating a first data value; receiving a first encrypted data value sent by a data source side; calculating a target encrypted data value from the first data value and the first encrypted data value; calculating a target encryption data value by using a preset algorithm to obtain a first ciphertext prediction value; calculating a first ciphertext data value according to the first ciphertext prediction value and the tag data, and calculating a first ciphertext updating gradient value according to the first ciphertext data value; the first ciphertext updating gradient value is subjected to salting processing and then the gradient is updated; after the gradient is updated, the first ciphertext data value is encrypted according to the preset first homomorphic key to obtain a second encrypted data value, and the second encrypted data value is sent to the data source side, so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key, the data leakage risk is reduced, and the non-leakage property of the intermediate step is ensured.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a method for training a longitudinal logistic regression model under privacy calculation of a third party.
Background
In the current "artificial intelligence" and "data technology" era, data is in need of mass circulation and fusion as one of the most important production elements. At present, when a plurality of organizations develop data cooperation, plaintext data of each party needs to be collected at one place (the plaintext data can be collected in a certain organization in a unified way or can be collected in a certain independent third-party platform), and then centralized quantitative modeling or analysis work is carried out. The data has the requirement of opening and merging, but on the other side, the requirements on data security and protection are more and more strict, and the requirement of each organization for the protection of the business data of the organization makes the operation of 'ex-warehouse' of the clear data of the organization more and more infeasible.
In the prior art, in order to solve the contradiction between data circulation and data security protection, the following methods are generally used: a longitudinal logistic regression algorithm based on a 'trusted third party' under a federal learning framework is used, and a means adopted by a two-party logistic regression method which does not completely protect intermediate steps is also adopted. However, although these methods greatly simplify the difficulty of implementing the longitudinal logistic regression algorithm under federal learning, and if the "trusted third party" does not collude with any party, the security can be guaranteed, and manufacturers using both parties believe that the Z value of the intermediate step is already processed and will not expose too much information. However, these conditions are ideal, and in actual situations, both a scheme of a trusted third party and a scheme of two parties without protecting intermediate steps may have significant potential safety hazards such as data leakage.
Disclosure of Invention
In view of this, the embodiment of the present invention provides a method for training a longitudinal logistic regression model under privacy computation by a third party, so as to solve the technical problem that in the prior art, a longitudinal logistic regression algorithm under a federal learning framework based on a "trusted third party" and a two-party logistic regression method with incomplete protection of intermediate steps are used to solve the problem of serious potential safety hazards such as data leakage when the problems of data circulation and data safety protection are solved.
The technical scheme provided by the invention is as follows:
the embodiment of the invention provides a method for training a longitudinal logistic regression model under privacy computation of a third party, which comprises the following steps: acquiring first user characteristic data in privacy calculation and calculating to obtain a corresponding first data value according to the first user characteristic data; receiving a first encrypted data value sent by a data source party, wherein the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source party according to a data demand party; calculating according to the first data value and the first encrypted data value to obtain a corresponding target encrypted data value; calculating the target encrypted data value by using a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value; calculating according to the first ciphertext predicted value and a target tag variable to obtain a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm, and calculating the first ciphertext data value to obtain a corresponding first ciphertext updating gradient value; after the first ciphertext updating gradient value is subjected to salt adding treatment, updating a corresponding first gradient in the longitudinal logistic regression algorithm; and after the first gradient is updated, encrypting the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value, and sending the second encrypted data value to the data source side so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key.
Optionally, the updating the corresponding first gradient in the longitudinal logistic regression algorithm after the salt processing is performed on the first ciphertext updating gradient value includes: combining the first ciphertext updating gradient value with a first preset random number and then sending the first ciphertext updating gradient value and the first preset random number to the data source side so that the data source side decrypts the combination formed by the first ciphertext updating gradient value and the first preset random number to obtain a corresponding first decrypted data value; and processing the first decrypted data value to obtain a corresponding first gradient value, and updating a corresponding first gradient in the longitudinal logistic regression algorithm according to the first gradient value.
Optionally, the calculating the target encrypted data value by using a preset federal-learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value includes: fitting an activation function in the longitudinal logistic regression algorithm by using a preset polynomial function to obtain a corresponding preset federal learning-based longitudinal logistic regression algorithm; and calculating the target encrypted data value by using the preset federated learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value.
The embodiment of the invention provides a method for training a longitudinal logistic regression model under privacy computation of a third party, which comprises the following steps: when a data use request sent by a data demand party is acquired, preprocessing second user characteristic data corresponding to the data use request to obtain a corresponding second data value; encrypting the second data value according to a preset second homomorphic key to obtain a corresponding first encrypted data value and sending the first encrypted data value to a corresponding data demand side; receiving a second encrypted data value sent by the data demand party, wherein the second encrypted data value is obtained by encrypting a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm in the data demand party according to a preset first homomorphic encryption public key by the data demand party, and the first ciphertext data value is obtained by processing the first encrypted data value by the data demand party; decrypting the second encrypted data value by using the preset first homomorphic key to obtain a corresponding second decrypted data value; and calculating to obtain a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user characteristic data, and updating a corresponding second gradient in the longitudinal logistic regression algorithm after salt processing is performed on the second ciphertext updating gradient value.
Optionally, the updating the corresponding second gradient in the longitudinal logistic regression algorithm after the second ciphertext updating gradient value is salted includes: calculating to obtain a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user characteristic data; combining the second ciphertext updating gradient value with a second preset random number and then sending the combined second ciphertext updating gradient value and the second preset random number to the data demand side so that the data demand side decrypts the combination formed by the second ciphertext updating gradient value and the second preset random number to obtain a corresponding third decrypted data value; and processing the third decrypted data value to obtain a corresponding second gradient value, and updating a corresponding second gradient in the longitudinal logistic regression algorithm according to the second gradient value.
The third aspect of the embodiments of the present invention provides a device for training a longitudinal logistic regression model under privacy computation to a third party, where the device for training the longitudinal logistic regression model under privacy computation to the third party includes: the first acquisition module is used for acquiring first user characteristic data in privacy calculation and calculating to obtain a corresponding first data value according to the first user characteristic data; the first receiving module is used for receiving a first encrypted data value sent by a data source party, wherein the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source party according to a data demand party; the first calculation module is used for calculating to obtain a corresponding target encrypted data value according to the first data value and the first encrypted data value; the second calculation module is used for calculating the target encrypted data value by utilizing a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext prediction value; the third calculation module is used for calculating to obtain a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm according to the first ciphertext prediction value and a target tag variable, and calculating the first ciphertext data value to obtain a corresponding first ciphertext updating gradient value; the first processing module is used for updating a corresponding first gradient in the longitudinal logistic regression algorithm after the first ciphertext updating gradient value is subjected to salt adding processing; and the second processing module is used for encrypting the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value and sending the second encrypted data value to the data source side after the first gradient is updated, so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key.
Optionally, the first processing module includes: the first calculation submodule is used for calculating to obtain a corresponding first ciphertext updating gradient value according to the first ciphertext data value; the first decryption submodule is used for combining the first ciphertext updating gradient value and a first preset random number and then sending the first ciphertext updating gradient value and the first preset random number to the data source side so that the data source side decrypts the combination formed by the first ciphertext updating gradient value and the first preset random number to obtain a corresponding first decrypted data value; and the first updating submodule is used for processing the first decrypted data value to obtain a corresponding first gradient value and updating a corresponding first gradient in the longitudinal logistic regression algorithm according to the first gradient value.
The fourth aspect of the embodiments of the present invention provides a device for training a longitudinal logistic regression model under privacy computation to a third party, where the device for training a longitudinal logistic regression model under privacy computation to a third party includes: the third processing module is used for preprocessing second user characteristic data corresponding to a data use request to obtain a corresponding second data value when the data use request sent by a data demander is obtained; the first encryption module is used for encrypting the second data value according to a preset second homomorphic key to obtain a corresponding first encrypted data value and sending the first encrypted data value to a corresponding data demand party; the second receiving module is used for receiving a second encrypted data value sent by the data demander, wherein the second encrypted data value is obtained by encrypting a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm in the data demander by the data demander according to a preset first homomorphic encryption public key, and the first ciphertext data value is obtained by processing the first encrypted data value by the data demander; the second decryption module is used for decrypting the second encrypted data value by using the preset first homomorphic key to obtain a corresponding second decrypted data value; and the updating module is used for calculating to obtain a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user characteristic data, and updating a corresponding second gradient in the longitudinal logistic regression algorithm after salt processing is carried out on the second ciphertext updating gradient value.
A fifth aspect of the embodiments of the present invention provides a computer-readable storage medium, where computer instructions are stored, and the computer instructions are configured to enable the computer to execute the method for training a longitudinal logistic regression model under privacy computing to a third party according to any one of the first aspect and the first aspect of the embodiments of the present invention, or the method for training a longitudinal logistic regression model under privacy computing to a third party according to any one of the second aspect and the second aspect of the embodiments of the present invention.
A sixth aspect of an embodiment of the present invention provides an electronic device, including: the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the method for training the longitudinal logistic regression model under the privacy computation of the third party according to any one of the first aspect and the first aspect of the embodiment of the invention, or execute the method for training the longitudinal logistic regression model under the privacy computation of the third party according to any one of the second aspect and the second aspect of the embodiment of the invention.
The technical scheme provided by the invention has the following effects:
according to the method for training the longitudinal logistic regression model under the privacy computation of the third party, the first user characteristic data in the privacy computation are obtained, and the corresponding first data value is obtained through computation according to the first user characteristic data; receiving a first encrypted data value sent by a data source party, wherein the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source party according to a data use request; calculating according to the first data value and the first encrypted data value to obtain a corresponding target encrypted data value; calculating the target encrypted data value by using a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value; calculating according to the first ciphertext predicted value and a target tag variable to obtain a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm, and calculating the first ciphertext data value to obtain a corresponding first ciphertext updating gradient value; after the first ciphertext updating gradient value is subjected to salting processing, updating a corresponding first gradient in the longitudinal logistic regression algorithm; and after the first gradient is updated, encrypting the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value, and sending the second encrypted data value to the data source side so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key. The method utilizes an exchangeable homomorphic encryption means, successfully removes the requirement of a trusted third party, and solves the problems of potential security hidden danger and the like caused by the potential collusion risk of the third party; the two parties participating in calculation are both the two parties actually having data, so that the cost required by calculation can be better estimated, and the situations of calculation power waste and the like are avoided. Therefore, by the real-time method, calculation of intermediate steps and numerical value exchange are carried out by using an exchangeable homomorphic encryption means, so that the risk of data leakage is reduced; the salt addition also ensures the leak-free property of the intermediate step.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow diagram of a method for training a longitudinal logistic regression model under privacy computation to a third party according to an embodiment of the invention;
FIG. 2 is a flow diagram of a method for training a longitudinal logistic regression model under privacy computation to a third party, according to an embodiment of the invention;
FIG. 3 is a block diagram of a longitudinal logistic regression model training apparatus under privacy computation to a third party according to an embodiment of the present invention;
FIG. 4 is a block diagram of a longitudinal logistic regression model training apparatus under privacy computation to a third party according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a computer-readable storage medium provided according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The embodiment of the invention provides a method for training a longitudinal logistic regression model under privacy calculation of a third party, which comprises the following steps of:
step S101: the method comprises the steps of obtaining first user characteristic data in privacy calculation and obtaining a corresponding first data value according to the first user characteristic data in the privacy calculation. Particularly, the federal learning is a machine learning framework, and can effectively help a plurality of organizations to perform data use and machine learning modeling under the condition of meeting the requirements of user privacy protection, data safety and the like. In federal learning, the data source learning system comprises two participants, namely a data demand side and a data source side, wherein both the two participants have corresponding user characteristic data, and the data demand side (A) also has a corresponding label. And calculating to obtain a corresponding first data value according to the first user characteristic data owned by the data demander (A).
Specifically, the calculation is performed by the following formula:
Z A ＝np.dot(X A ,w A )+b…………………………………(1)
wherein Z is A Representing a first data value; x A The first user characteristic data can be numerical value type, character type and the like, and the inventionNot specifically limited as long as it can contribute to prediction of the corresponding tag; w is a A The initial value of the model coefficient is 0, the model training result has a plurality of rounds, each round of training obtains two gradient values,
and
then using a gradient to pair w a And w b Updating:
where alpha is a learning rate set by the user before model training, and may be set to 0.3, for example.
w a And w b After the update is completed, the training process is repeated once: to obtain new
And
thereby updating w a And w b Until the model training is finished; b represents the intercept;
step S102: receiving a first encrypted data value sent by a data source party, wherein the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source party according to a data need party. Specifically, the data source side (B) is a side without tag data, but has user feature data, and when the data source side receives a data use request initiated by the data demanding side, first, obtains corresponding second user feature data according to the data use request and pre-processes the user feature data:
Z B ＝np.dot(X B ,w B )…………………………………(2)
wherein Z is B Representing a processed value after preprocessing the user characteristic data; x B The second user characteristic data is represented, can be used as the supplementary content of the first user characteristic data in the data demand party, and can be in a numerical type, a character type and the like, and the invention is not particularly limited; w is a B Represents a parameter, and the initial value is 0;
secondly, the data source side generates a key (pk, sk) for homomorphic encryption (satisfying the commutative rule) B ；
Finally, the key pk is used B For the pre-processed processing value Z B Encrypting to obtain a first encrypted data value
And sending the first encrypted data value to the corresponding data consumer.
Step S103: and calculating according to the first data value and the first encrypted data value to obtain a corresponding target encrypted data value. In particular, due to the encryption of the data sent from the data source side
Is under a homomorphic encryption system, and therefore, the target encrypted data value under homomorphic encryption is directly obtained by the following formula:
step S104: and calculating the target encrypted data value by using a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value. Specifically, the activation function in the federal-learning-based longitudinal logistic regression algorithm cannot be calculated in a ciphertext form due to the nonlinear characteristic of the activation function, so that the preset federal-learning-based longitudinal logistic regression algorithm which can be calculated in the ciphertext is obtained through preprocessing, wherein the preprocessing method is not specifically limited in the invention, and only needs to meet the calculation requirement of the processing result.
And finally, calculating the target encrypted data value through the finally obtained algorithm to obtain a corresponding first ciphertext predicted value:
wherein, the first and the second end of the pipe are connected with each other,
representing a first ciphertext prediction value; sigmoid () represents a preset federal learning-based longitudinal logistic regression algorithm;
representing the target encrypted data value.
Step S105: and calculating to obtain a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm according to the first ciphertext prediction value and the target tag variable, and calculating the first ciphertext data value to obtain a corresponding first ciphertext updating gradient value. Specifically, according to the longitudinal logistic regression algorithm, the data demander calculates by using the first ciphertext predicted value and the target tag variable through the following formula to obtain a first ciphertext data value for gradient calculation in the longitudinal logistic regression algorithm:
wherein the content of the first and second substances,
representing a first ciphertext data value; and Y represents a target label variable, represents a variable which is determined according to the first user characteristic data and is used for evaluating the label, and takes the value of 0 (bad label) or 1 (good label).
Calculating a first ciphertext update gradient value by:
wherein, the first and the second end of the pipe are connected with each other,
representing a first ciphertext update gradient value; m represents the sample size;
represents a pair X A And performing transposition processing.
Step S106: and after the first ciphertext updating gradient value is subjected to salt adding treatment, updating a corresponding first gradient in the longitudinal logistic regression algorithm. Specifically, the first ciphertext update gradient value is used for gradient update in the vertical logistic regression algorithm, the first ciphertext update gradient value obtained in step S105 is an encrypted numerical value and needs to be sent to the data source side for decryption, and in order to ensure that data is not acquired by the data source side, the first ciphertext update gradient value is subjected to salting processing, and then the processing result is used for updating the corresponding first gradient in the vertical logistic regression algorithm.
Step S107: and after the first gradient is updated, encrypting the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value, and sending the second encrypted data value to the data source side so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key. Specifically, after updating the corresponding gradient, the data demander generates a preset first homomorphic encryption key (pk, sk) belonging to the data demander A And using the key to pair the first ciphertext data value
Performing secondary encryption to obtain a second encrypted data value
This second encrypted data value is then sent to the data source side, due to the useThe homomorphic encryption methods are interchangeable, so that the data source side can decrypt the second encrypted data value to obtain a corresponding decrypted value
Wherein, the
The result is obtained by simply using the key of the data demand side.
According to the method for training the longitudinal logistic regression model under the privacy calculation of the third party, provided by the embodiment of the invention, the requirement of a credible third party is successfully removed by using an exchangeable homomorphic encryption means, and the problems of potential security hidden danger and the like caused by the potential collusion risk of the third party are solved; the two parties participating in calculation are both parties actually having data, so that the cost required by calculation can be better estimated, and the situations of calculation waste and the like are avoided. Therefore, by the real-time method, calculation of intermediate steps and numerical value exchange are carried out by using an exchangeable homomorphic encryption means, so that the risk of data leakage is reduced; the salt addition also ensures the leak-free property of the intermediate step.
As an optional implementation manner of the embodiment of the present invention, step S104 includes: fitting an activation function in the longitudinal logistic regression algorithm by using a preset polynomial function to obtain a corresponding preset federal learning-based longitudinal logistic regression algorithm; and calculating the target encrypted data value by using the preset federated learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value.
Specifically, an activation function (sigmoid function) in the federal-learning-based longitudinal logistic regression algorithm is a nonlinear function for logistic regression, and cannot be calculated in a ciphertext form. And the operation of polynomial level can be carried out on the ciphertext, so a polynomial function F is used to fit the sigmoid function, and at this time, the formula (4) is changed into the following form:
wherein e represents a constant; a is 1 ，a 2 ，a 3 Representing the coefficients before encryption.
As an optional implementation manner of the embodiment of the present invention, step S106 includes: the first ciphertext updating gradient value and a first preset random number are combined and then sent to the data source side, so that the data source side decrypts the combination formed by the first ciphertext updating gradient value and the first preset random number to obtain a corresponding first decrypted data value; and processing the first decrypted data value to obtain a corresponding first gradient value, and updating a corresponding first gradient in the longitudinal logistic regression algorithm according to the first gradient value.
Specifically, the first ciphertext update gradient value is subjected to salting:
where r represents a random number known only to the data-requiring party, will be
Sending the data to a corresponding data source side for decryption to obtain dw A + r, then the data consumer receives the dw A After + r, the random number r is removed, and the corresponding gradient value dw can be obtained A And simultaneously using the gradient value dw A Updating the corresponding first gradient w in the longitudinal logistic regression algorithm A 。
The embodiment of the invention also provides a method for training a longitudinal logistic regression model under privacy computation of a third party, which comprises the following steps of:
step S201: when a data use request sent by a data demand party is acquired, preprocessing second user characteristic data corresponding to the data use request to obtain a corresponding second data value. For a specific implementation process, reference is made to the process of preprocessing the second user feature data in step S102, which is not described herein again. Wherein the second data value is Z in formula (2) B 。
Step S202: and encrypting the second data value according to a preset second homomorphic key to obtain a corresponding first encrypted data value and sending the first encrypted data value to a corresponding data demand side. The specific implementation process refers to the encryption process in step S102, and is not described herein again. Wherein, the preset second homomorphic key is (pk, sk) B 。
Step S203: and receiving a second encrypted data value sent by the data demand party, wherein the second encrypted data value is obtained by encrypting a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm in the data demand party according to a preset first homomorphic encryption public key by the data demand party, and the first ciphertext data value is obtained by processing the first encrypted data value by the data demand party. Specifically, the process of obtaining the first ciphertext data value according to the first encryption data value refers to the implementation processes of step S103 to step S105; the process of obtaining a second encrypted data value from the first ciphertext data value refers to the encryption process of step S107; and will not be described in detail herein.
Step S204: and decrypting the second encrypted data value by using the preset first homomorphic key to obtain a corresponding second decrypted data value. The specific decryption process refers to the decryption process in step S107, and is not described herein again. Wherein, the result obtained in step S107
I.e. the second decrypted data value.
Step S205: and calculating to obtain a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user characteristic data, and updating a corresponding second gradient in the longitudinal logistic regression algorithm after salt processing is performed on the second ciphertext updating gradient value.
Specifically, referring to equation (6), the corresponding second ciphertext update gradient value is calculated by the following equation:
wherein，
Representing a second ciphertext update gradient value; m represents the sample size;
represents a pair X B And performing transposition processing.
According to the method for training the longitudinal logistic regression model under the privacy computation without the third party, provided by the embodiment of the invention, the requirement of a credible third party is successfully removed by using an exchangeable homomorphic encryption means, and the problems of potential safety hazards and the like caused by the potential collusion risk of the third party are solved; the two parties participating in calculation are both the two parties actually having data, so that the cost required by calculation can be better estimated, and the situations of calculation power waste and the like are avoided. Therefore, by the real-time method and the system, the calculation of the intermediate step and the exchange of the numerical value are carried out by using an exchangeable homomorphic encryption means, and the risk of data leakage is reduced.
As an optional implementation manner of the embodiment of the present invention, step S205 includes: combining the second ciphertext updating gradient value with a second preset random number and then sending the combined second ciphertext updating gradient value and the second preset random number to the data demand side so that the data demand side decrypts the combination formed by the second ciphertext updating gradient value and the second preset random number to obtain a corresponding third decrypted data value; and processing the third decrypted data value to obtain a corresponding second gradient value, and updating a corresponding second gradient in the longitudinal logistic regression algorithm according to the second gradient value.
Specifically, the second ciphertext update gradient value is subjected to salting:
wherein r is 1 Representing a random number known only to the data source, will be
Sending the data to a corresponding data demand side for decryption to obtain dw B +r 1 Then, the data is requestedThe party receives the dw B +r 1 Then, the random number r is removed 1 Not only can obtain the corresponding gradient value dw B And simultaneously using the gradient value dw B Updating the corresponding second gradient w in the vertical logistic regression algorithm B . The non-leakage property of the intermediate step is ensured by utilizing the means of adding salt.
An embodiment of the present invention further provides a device for training a longitudinal logistic regression model under privacy computation by a third party, and as shown in fig. 3, the device includes:
the first obtaining module 301 is configured to obtain first user characteristic data in privacy calculation and calculate a corresponding first data value according to the first user characteristic data; for details, refer to the related description of step S101 in the above method embodiment.
A first receiving module 302, configured to receive a first encrypted data value sent by a data source, where the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source according to the data use request; for details, refer to the related description of step S102 in the above method embodiment.
A first calculating module 303, configured to calculate a corresponding target encrypted data value according to the first data value and the first encrypted data value; for details, refer to the related description of step S103 in the above method embodiment.
The second calculation module 304 is configured to calculate the target encrypted data value by using a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext prediction value; for details, refer to the related description of step S104 in the above method embodiment.
A third calculating module 305, configured to calculate, according to the first ciphertext prediction value and the target tag variable, a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm, and calculate the first ciphertext data value to obtain a corresponding first ciphertext update gradient value; for details, refer to the related description of step S105 in the above method embodiment.
A first processing module 306, configured to update a corresponding first gradient in the longitudinal logistic regression algorithm after performing salt processing on the first ciphertext update gradient value; for details, refer to the related description of step S106 in the above method embodiment.
A second processing module 307, configured to encrypt the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value and send the second encrypted data value to the data source side after the first gradient is updated, so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key; for details, refer to the related description of step S107 in the above method embodiment.
The longitudinal logistic regression model training device under the privacy computation without the third party provided by the embodiment of the invention successfully removes the requirement of a credible third party by using an exchangeable homomorphic encryption means, and solves the problems of potential safety hidden danger and the like caused by the potential collusion risk of the third party; the two parties participating in calculation are both the two parties actually having data, so that the cost required by calculation can be better estimated, and the situations of calculation power waste and the like are avoided. Therefore, by the real-time method, the calculation of the intermediate step and the exchange of the numerical value are carried out by using an exchangeable homomorphic encryption means, so that the risk of data leakage is reduced; the salt adding method also ensures the non-leakage property of the intermediate step.
As an optional implementation manner of the embodiment of the present invention, the first processing module includes: the first decryption submodule is used for combining the first ciphertext updating gradient value and a first preset random number and then sending the first ciphertext updating gradient value and the first preset random number to the data source side so that the data source side decrypts the combination formed by the first ciphertext updating gradient value and the first preset random number to obtain a corresponding first decrypted data value; and the first updating submodule is used for processing the first decrypted data value to obtain a corresponding first gradient value and updating a corresponding first gradient in the longitudinal logistic regression algorithm according to the first gradient value.
As an optional implementation manner of the embodiment of the present invention, the second calculating module includes: the fitting module is used for fitting an activation function in the longitudinal logistic regression algorithm by using a preset polynomial function to obtain a corresponding preset federal learning-based longitudinal logistic regression algorithm; and the first calculation submodule is used for calculating the target encrypted data value by utilizing the preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value.
The functional description of the device for training the longitudinal logistic regression model under the privacy computation to the third party provided by the embodiment of the invention is described in detail with reference to the method for training the longitudinal logistic regression model under the privacy computation to the third party in the embodiment.
An embodiment of the present invention further provides a device for training a longitudinal logistic regression model under privacy computation by a third party, and as shown in fig. 4, the device includes:
the third processing module 401 is configured to, when a data usage request sent by a data demander is obtained, pre-process second user characteristic data corresponding to the data usage request to obtain a corresponding second data value; for details, refer to the related description of step S201 in the above method embodiment.
A first encryption module 402, configured to encrypt the second data value according to a preset second homomorphic key to obtain a corresponding first encrypted data value, and send the first encrypted data value to a corresponding data demand side; for details, refer to the related description of step S202 in the above method embodiment.
A second receiving module 403, configured to receive a second encrypted data value sent by the data demander, where the second encrypted data value is obtained by the data demander encrypting, according to a preset first homomorphic encryption public key, a first ciphertext data value used for gradient calculation in a vertical logistic regression algorithm in the data demander, and the first ciphertext data value is obtained by the data demander processing the first encrypted data value; for details, refer to the related description of step S203 in the above method embodiment.
A second decryption module 404, configured to decrypt the second encrypted data value by using the preset first homomorphic key to obtain a corresponding second decrypted data value; for details, refer to the related description of step S204 in the above method embodiment.
An updating module 405, configured to calculate a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user feature data, and update a corresponding second gradient in the longitudinal logistic regression algorithm after performing salt processing on the second ciphertext updating gradient value; for details, refer to the related description of step S205 in the above method embodiment.
The longitudinal logistic regression model training device under privacy calculation for the third party provided by the embodiment of the invention successfully removes the requirement of a credible third party by using an exchangeable homomorphic encryption means, and solves the problems of potential safety hazards and the like caused by the potential collusion risk of the third party; the two parties participating in calculation are both the two parties actually having data, so that the cost required by calculation can be better estimated, and the situations of calculation power waste and the like are avoided. Therefore, by the real-time method and the system, the calculation of the intermediate step and the exchange of the numerical value are carried out by using an exchangeable homomorphic encryption means, and the risk of data leakage is reduced.
As an optional implementation manner of the embodiment of the present invention, the update module includes: the second decryption submodule is used for combining the second ciphertext updating gradient value and a second preset random number and then sending the combined value to the data demand side so that the data demand side decrypts the combination formed by the second ciphertext updating gradient value and the second preset random number to obtain a corresponding third decrypted data value; and the second updating submodule is used for processing the third decrypted data value to obtain a corresponding second gradient value and updating a corresponding second gradient in the longitudinal logistic regression algorithm according to the second gradient value.
The functional description of the device for training the longitudinal logistic regression model under the privacy computation to the third party provided by the embodiment of the invention is described in detail with reference to the method for training the longitudinal logistic regression model under the privacy computation to the third party in the embodiment.
An embodiment of the present invention further provides a storage medium, as shown in fig. 5, on which a computer program 501 is stored, where the instructions, when executed by a processor, implement the steps of the method for training a longitudinal logistic regression model under privacy computation by a third party in the foregoing embodiment. The storage medium is also stored with audio and video stream data, characteristic frame data, an interactive request signaling, encrypted data, preset data size and the like. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a Flash Memory (Flash Memory), a Hard Disk Drive (Hard Disk Drive, abbreviated as HDD), or a Solid State Drive (SSD); the storage medium may also comprise a combination of memories of the kind described above.
Those skilled in the art will appreciate that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium and can include the processes of the embodiments of the methods described above when executed. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a Flash Memory (Flash Memory), a Hard Disk (Hard Disk Drive, abbreviated as HDD), a Solid State Drive (SSD), or the like; the storage medium may also comprise a combination of memories of the kind described above.
An embodiment of the present invention further provides an electronic device, as shown in fig. 6, the electronic device may include a processor 61 and a memory 62, where the processor 61 and the memory 62 may be connected by a bus or in another manner, and fig. 6 illustrates the connection by the bus as an example.
The memory 62, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as the corresponding program instructions/modules in embodiments of the present invention. The processor 61 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory 62, namely, implementing the method for training the longitudinal logistic regression model under the privacy computation of the third party in the above method embodiment.
The memory 62 may include a storage program area and a storage data area, wherein the storage program area may store an operating device, an application program required for at least one function; the storage data area may store data created by the processor 61, and the like. Further, the memory 62 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 62 may optionally include memory located remotely from the processor 61, and these remote memories may be connected to the processor 61 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The one or more modules are stored in the memory 62 and, when executed by the processor 61, perform a method of training a longitudinal logistic regression model under privacy calculations to third parties as in the embodiment of fig. 1-2.
The details of the electronic device may be understood by referring to the corresponding descriptions and effects in the embodiments shown in fig. 1 to fig. 2, and are not described herein again.
Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope defined by the appended claims.
Claims (10)
1. A longitudinal logistic regression model training method under privacy calculation of a third party is characterized by comprising the following steps:
acquiring first user characteristic data in privacy calculation and calculating to obtain a corresponding first data value according to the first user characteristic data;
receiving a first encrypted data value sent by a data source party, wherein the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source party according to a data demand party;
calculating according to the first data value and the first encrypted data value to obtain a corresponding target encrypted data value;
calculating the target encrypted data value by using a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value;
calculating according to the first ciphertext predicted value and a target tag variable to obtain a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm, and calculating the first ciphertext data value to obtain a corresponding first ciphertext updating gradient value;
after the first ciphertext updating gradient value is subjected to salt adding treatment, updating a corresponding first gradient in the longitudinal logistic regression algorithm;
and after the first gradient is updated, encrypting the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value, and sending the second encrypted data value to the data source side so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key.
2. The method of claim 1, wherein updating the corresponding first gradient in the vertical logistic regression algorithm after salting the first ciphertext update gradient value comprises:
combining the first ciphertext updating gradient value with a first preset random number and then sending the first ciphertext updating gradient value and the first preset random number to the data source side so that the data source side decrypts the combination formed by the first ciphertext updating gradient value and the first preset random number to obtain a corresponding first decrypted data value;
and processing the first decrypted data value to obtain a corresponding first gradient value, and updating a corresponding first gradient in the longitudinal logistic regression algorithm according to the first gradient value.
3. The method according to claim 1, wherein the calculating the target encrypted data value by using a preset federated learning-based vertical logistic regression algorithm to obtain a corresponding first ciphertext predicted value comprises:
fitting an activation function in the longitudinal logistic regression algorithm by using a preset polynomial function to obtain a corresponding preset federal learning-based longitudinal logistic regression algorithm;
and calculating the target encrypted data value by using the preset federated learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext predicted value.
4. A longitudinal logistic regression model training method under privacy computation of a third party is characterized by comprising the following steps:
when a data use request sent by a data demand party is acquired, preprocessing second user characteristic data corresponding to the data use request to obtain a corresponding second data value;
encrypting the second data value according to a preset second homomorphic key to obtain a corresponding first encrypted data value and sending the first encrypted data value to a corresponding data demand side;
receiving a second encrypted data value sent by the data demand party, wherein the second encrypted data value is obtained by encrypting a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm in the data demand party according to a preset first homomorphic encryption public key by the data demand party, and the first ciphertext data value is obtained by processing the first encrypted data value by the data demand party;
decrypting the second encrypted data value by using the preset first homomorphic key to obtain a corresponding second decrypted data value;
and calculating to obtain a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user characteristic data, and updating a corresponding second gradient in the longitudinal logistic regression algorithm after salt processing is performed on the second ciphertext updating gradient value.
5. The method of claim 4, wherein updating the corresponding second gradient in the vertical logistic regression algorithm after salting the second ciphertext update gradient value comprises:
combining the second ciphertext updating gradient value with a second preset random number and then sending the combined second ciphertext updating gradient value and the second preset random number to the data demand side so that the data demand side decrypts the combination formed by the second ciphertext updating gradient value and the second preset random number to obtain a corresponding third decrypted data value;
and processing the third decrypted data value to obtain a corresponding second gradient value, and updating a corresponding second gradient in the longitudinal logistic regression algorithm according to the second gradient value.
6. A device for training a longitudinal logistic regression model under privacy computation of a third party comprises:
the first acquisition module is used for acquiring first user characteristic data in privacy calculation and calculating to obtain a corresponding first data value according to the first user characteristic data;
the first receiving module is used for receiving a first encrypted data value sent by a data source party, wherein the first encrypted data value is obtained by preprocessing and encrypting second user characteristic data corresponding to a data use request initiated by the data source party according to a data demand party;
the first calculation module is used for calculating to obtain a corresponding target encrypted data value according to the first data value and the first encrypted data value;
the second calculation module is used for calculating the target encrypted data value by utilizing a preset federal learning-based longitudinal logistic regression algorithm to obtain a corresponding first ciphertext prediction value;
the third calculation module is used for calculating to obtain a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm according to the first ciphertext prediction value and a target tag variable, and calculating the first ciphertext data value to obtain a corresponding first ciphertext updating gradient value;
the first processing module is used for updating a corresponding first gradient in the longitudinal logistic regression algorithm after the first ciphertext updating gradient value is subjected to salt adding processing;
and the second processing module is used for encrypting the first ciphertext data value according to a preset first homomorphic key to obtain a corresponding second encrypted data value and sending the second encrypted data value to the data source side after the first gradient is updated, so that the data source side decrypts the second encrypted data value by using the preset first homomorphic key.
7. The apparatus of claim 6, wherein the first processing module comprises:
the first calculation submodule is used for calculating to obtain a corresponding first ciphertext updating gradient value according to the first ciphertext data value;
the first decryption submodule is used for combining the first ciphertext updating gradient value and a first preset random number and then sending the first ciphertext updating gradient value and the first preset random number to the data source side so that the data source side decrypts the combination formed by the first ciphertext updating gradient value and the first preset random number to obtain a corresponding first decrypted data value;
and the first updating submodule is used for processing the first decrypted data value to obtain a corresponding first gradient value and updating a corresponding first gradient in the longitudinal logistic regression algorithm according to the first gradient value.
8. A device for training a longitudinal logistic regression model under privacy computation of a third party comprises:
the third processing module is used for preprocessing second user characteristic data corresponding to a data use request to obtain a corresponding second data value when the data use request sent by a data demander is obtained;
the first encryption module is used for encrypting the second data value according to a preset second homomorphic key to obtain a corresponding first encrypted data value and sending the first encrypted data value to a corresponding data demand party;
the second receiving module is used for receiving a second encrypted data value sent by the data demander, wherein the second encrypted data value is obtained by encrypting a first ciphertext data value used for gradient calculation in a longitudinal logistic regression algorithm in the data demander by the data demander according to a preset first homomorphic encryption public key, and the first ciphertext data value is obtained by processing the first encrypted data value by the data demander;
the second decryption module is used for decrypting the second encrypted data value by using the preset first homomorphic key to obtain a corresponding second decrypted data value;
and the updating module is used for calculating to obtain a corresponding second ciphertext updating gradient value according to the second decrypted data value and the user characteristic data, and updating a corresponding second gradient in the longitudinal logistic regression algorithm after salt processing is carried out on the second ciphertext updating gradient value.
9. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the method for training a longitudinal logistic regression model under privacy computation to a third party according to any one of claims 1 to 5.
10. An electronic device, comprising: a memory and a processor, the memory and the processor being communicatively coupled, the memory storing computer instructions, and the processor executing the computer instructions to perform the method of training a longitudinal logistic regression model under privacy calculations to a third party according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210650114.5A CN115130121A (en) | 2022-06-09 | 2022-06-09 | Method for training longitudinal logistic regression model under privacy calculation of third party |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210650114.5A CN115130121A (en) | 2022-06-09 | 2022-06-09 | Method for training longitudinal logistic regression model under privacy calculation of third party |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115130121A true CN115130121A (en) | 2022-09-30 |
Family
ID=83378310
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210650114.5A Pending CN115130121A (en) | 2022-06-09 | 2022-06-09 | Method for training longitudinal logistic regression model under privacy calculation of third party |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115130121A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115292739A (en) * | 2022-10-08 | 2022-11-04 | 江苏浚荣升新材料科技有限公司 | Data management method of metal mold design system |
| CN115580496A (en) * | 2022-12-09 | 2023-01-06 | 北京融数联智科技有限公司 | Logistic regression training method, system and device under privacy calculation without third party |
-
2022
- 2022-06-09 CN CN202210650114.5A patent/CN115130121A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115292739A (en) * | 2022-10-08 | 2022-11-04 | 江苏浚荣升新材料科技有限公司 | Data management method of metal mold design system |
| CN115580496A (en) * | 2022-12-09 | 2023-01-06 | 北京融数联智科技有限公司 | Logistic regression training method, system and device under privacy calculation without third party |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109002861B (en) | Federal modeling method, device and storage medium | |
| CN110189192B (en) | Information recommendation model generation method and device | |
| US20210248244A1 (en) | Model parameter training method, terminal, and system based on federation learning, and medium | |
| CN110276210B (en) | Method and device for determining model parameters based on federal learning | |
| WO2020248538A1 (en) | Federated learning-based model parameter training method and device | |
| CN115130121A (en) | Method for training longitudinal logistic regression model under privacy calculation of third party | |
| CN112347495B (en) | Trusted privacy intelligent service computing system and method based on blockchain | |
| CN111898145A (en) | Neural network model training method, device, equipment and medium | |
| CN112182644A (en) | Data processing method and device and electronic equipment | |
| CN114696990B (en) | Multi-party computing method, system and related equipment based on fully homomorphic encryption | |
| CN111104968B (en) | Safety SVM training method based on block chain | |
| CN113836556A (en) | Federal learning-oriented decentralized function encryption privacy protection method and system | |
| CN111027981A (en) | Method and device for multi-party joint training of risk assessment model for IoT (Internet of things) machine | |
| CN112818369A (en) | Combined modeling method and device | |
| CN114301677B (en) | Key negotiation method, device, electronic equipment and storage medium | |
| CN113472805B (en) | Model training method and device, storage medium and electronic equipment | |
| CN113055153B (en) | Data encryption method, system and medium based on fully homomorphic encryption algorithm | |
| CN113221153B (en) | Graph neural network training method and device, computing equipment and storage medium | |
| CN115021985B (en) | Logistic regression model training method and system without third party participation | |
| CN115580443A (en) | Graph data processing method, device, equipment and medium | |
| CN113537516B (en) | Training method, device, equipment and medium for distributed machine learning model | |
| CN115758402A (en) | Artificial intelligence model federal learning method combining homomorphic encryption and model watermarking | |
| CN112054891B (en) | Block chain-based common user determination method, electronic device and storage medium | |
| CN113672954A (en) | Feature extraction method and device and electronic equipment | |
| CN113886868A (en) | Logistic regression safety training method and system for multi-source vertical distribution data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |