CN113343277B - Safe and efficient entrusted privacy data category prediction method - Google Patents

Safe and efficient entrusted privacy data category prediction method Download PDF

Info

Publication number
CN113343277B
CN113343277B CN202110752010.0A CN202110752010A CN113343277B CN 113343277 B CN113343277 B CN 113343277B CN 202110752010 A CN202110752010 A CN 202110752010A CN 113343277 B CN113343277 B CN 113343277B
Authority
CN
China
Prior art keywords
key
matrix
privacy data
ciphertext
sigma
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.)
Active
Application number
CN202110752010.0A
Other languages
Chinese (zh)
Other versions
CN113343277A (en
Inventor
刘静
张良峰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
ShanghaiTech University
Original Assignee
ShanghaiTech University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by ShanghaiTech University filed Critical ShanghaiTech University
Priority to CN202110752010.0A priority Critical patent/CN113343277B/en
Publication of CN113343277A publication Critical patent/CN113343277A/en
Application granted granted Critical
Publication of CN113343277B publication Critical patent/CN113343277B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • 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
    • G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
    • G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
    • G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
    • 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
    • G06F21/64—Protecting data integrity, e.g. using checksums, certificates or signatures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Security & Cryptography (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Bioethics (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a safe and efficient method for entrusting privacy data category prediction, which utilizes a given classifier to conduct category prediction on privacy data to obtain category prediction results Y of a plurality of classifiers on a plurality of pieces of privacy data, sets model parameters of the plurality of classifiers to form a matrix F, and forms a matrix X of the plurality of pieces of privacy data of an entruster. The invention provides a entrusted calculation method for privacy data category prediction, which can simultaneously meet the data privacy, public verifiability and experiment-based high efficiency. When the matrix size is greater than 10000, it can be observed in experimental data that the run time of the principal after the principal calculation is less than the time required for direct calculation.

Description

Safe and efficient entrusted privacy data category prediction method
Technical Field
The invention relates to a method for entrusting privacy data category prediction.
Background
Machine learning classifiers have a wide range of applications in real life, such as medical or genomic prediction, spam detection, facial recognition, and financial computing, where large amounts of private data are involved and need to be protected at the time of prediction. Most class prediction processes of classifiers (e.g., perceptrons, support vector machines) can be roughly seen as matrix multiplication operations. The prediction process is therefore highly complex, and is a heavy computational burden for devices with limited computational resources (e.g., internet of things clients). Delegation of computing solves this problem as a new computing model that allows computing resource-limited devices to delegate computing to resource-rich devices (e.g., cloud servers).
Entrusting calculation brings challenges while avoiding the participation of entrusters in complex calculation and improving the calculation efficiency. Principal challenges faced by delegated computing are: the privacy information of the delegator may be known by a malicious attacker; the delegate may return erroneous calculation results in order to reduce the workload or collusion with a malicious attacker; the runtime of the delegate after delegating the computation may be greater than the time required for direct computation. The design of a matrix multiplication entrusted calculation scheme capable of coping with all challenges has important application value.
The existing delegation scheme for privacy data class prediction cannot simultaneously cope with all challenges, namely, the following important characteristics are provided: 1) Data privacy: semantic security is achieved in the aspect of protecting the privacy information of the entrusters; 2) Public verifiability: the verification algorithm can verify the correctness of the entrusted calculation result, and can be run by any entity; 3) Efficiency based on experiments: from the experimental data it can be observed that the runtime of the delegate after the delegate calculation is less than the time required for the direct calculation.
Mohassel protects data privacy by using existing homomorphic encryption schemes (e.g., goldwasser-Micali encryption schemes) and verifies the correctness of the delegated computation results by a random algorithm. The scheme provides data privacy and public verifiability, but the calculation process involves a large number of modular exponentiations, and experimental-based efficiency is not realized. The patent et al propose an operation of delegated matrix vector multiplication, which implements public verifiability. When the data privacy is satisfied, it is found through theoretical analysis that when the matrix size is very large (greater than 250000), the runtime of the principal in the scheme after the principal calculation is smaller than the time required for direct calculation. This approach does not achieve experimental-based efficiency.
Disclosure of Invention
The invention aims to solve the technical problems that: the existing delegation scheme for privacy data class prediction cannot simultaneously provide data privacy, public verifiability and experiment-based high efficiency.
In order to solve the technical problems, the technical scheme of the invention provides a safe and efficient method for entrusting privacy data category prediction, which utilizes a given classifier to conduct category prediction on privacy data to obtain category prediction results Y of a plurality of classifiers on a plurality of pieces of privacy data, sets model parameters of the plurality of classifiers to form a matrix F, and forms a matrix X of the plurality of pieces of privacy data of an entruster, and is characterized in that FX is calculated to obtain the category prediction results of the plurality of classifiers on the plurality of pieces of privacy data at the same time, and at the moment, the matrix F and the matrix X are regarded as functions and inputs in matrix multiplication operation, and the method comprises the following steps:
step 1, selecting a safety parameter lambda, and according to the safety parameter lambda and a function setThe principal gets the public key pk=lhep and initializes the private key SK to +.>The common parameter lhep= (p, q, n, f (x), χ) is input 1 by the linear homomorphic encryption LHE's parameter generation algorithm λ Obtained, wherein q is a prime number; />Is a prime number; f (x) =x n +1 is a cyclic polynomial; /> Is distributed in the ring->The standard deviation is the discrete gaussian distribution of r;
step 2, according to the public key PK, the private key SK and the functionThe principal gets the key EK for the calculation F =f and key VK for authentication F ＝F；
Step 3, according to the public key PK, the private key SK and the inputThe entruster obtains the secret corresponding to the inputText (A)Key for authentication VK X And a key DK for decryption X ；
Step 4, according to the key used for calculationAnd ciphertext->The trusted party can calculate the result ciphertext +.>
Step 5, according to the key for verificationVK X ＝σ X And result ciphertext->The principal can obtain 1, representing the result ciphertext sigma Y Is correct, or 0, represents the result ciphertext sigma Y Is wrong;
step 6, according to the key DK for decryption X Sum result ciphertextThe delegate can actually calculate the result y=fx.
Preferably, the step 3 specifically includes the following steps:
step 301, obtaining an encryption key sk by a key generation algorithm of the linear homomorphic encryption LHE according to the public key pk=lhep;
step 302, dividing tn columns of input X into t blocks X 1 ,X 2 ,…,X t Each block X i Containing n columns, then input X write X= [ X ] 1 … X t ]Each blockIs a d x n matrix;
step 303, for each vector obtained in step 302Performing an encryption algorithm of a linear homomorphic encryption LHE with sk as an encryption key to obtain ciphertext ++>And each block->Is encrypted as->The input X is encrypted as +.>
Step 304, performing the following assignment: VK (vK) X ＝σ X ，DK X =sk, wherein decryption key DK X Is private to the principal.
Preferably, the step 5 specifically includes the following steps:
step 501, slave setUniformly and randomly selecting a vector r;
step 501, calculate F (σ) X r) and sigma Y r, if F (sigma) X r) and sigma Y r is equal, the principal obtains 1, if F (σ X r) and sigma Y r is not equal, the principal gets 0.
Preferably, the step 6 specifically includes the following steps:
step 601, inputting sigma Y Divided into t blocks by 2tn columnsEach block->Containing 2n columns, then input sigma Y WritingEach block->Is an m x 2n matrix, i.epsilon.t]；
Step 601, for each vectorWith the encryption key sk as the decryption key, performing a decryption algorithm of the linear homomorphic encryption LHE to obtain ciphertext ++>And each block->Decrypted as +.>Ciphertext->Decrypted as +.>
The invention provides a entrusted calculation method for privacy data category prediction, which can simultaneously meet the data privacy, public verifiability and experiment-based high efficiency. When the matrix size is greater than 10000, it can be observed in experimental data that the run time of the principal after the principal calculation is less than the time required for direct calculation.
Drawings
Fig. 1 is a schematic diagram of the partitioning of an input X.
Detailed Description
The invention will be further illustrated with reference to specific examples. It is to be understood that these examples are illustrative of the present invention and are not intended to limit the scope of the present invention. Further, it is understood that various changes and modifications may be made by those skilled in the art after reading the teachings of the present invention, and such equivalents are intended to fall within the scope of the claims appended hereto.
In the delegated privacy data class prediction problem, the parameters in the classifier have been obtained through a training process. The present invention only discusses how class predictions are made on private data using a given classifier. The model parameters of one classifier can be regarded as a row vector and the model parameters of a plurality of classifiers can form a matrix F. A piece of privacy data of the principal can be regarded as a column vector, and a plurality of pieces of privacy data can constitute a matrix X. The FX is calculated to obtain the category prediction results of a plurality of classifiers on a plurality of pieces of private data at the same time. At this time, the matrix F and the matrix X can be regarded as functions and inputs in the matrix multiplication operation.
Selecting a security parameter lambda, and summing the security parameter lambda and the function set Representing a set of m rows and d columns of the matrix, wherein the elements in the set come from the group +.>Representing the integer set, the principal can obtain the public key pk=lhep, and let the private key +.>The common parameter lhep= (p, q, n, f (x), χ) is input 1 by the linear homomorphic encryption LHE's parameter generation algorithm λ Obtained, wherein q is a prime number; />Is prime number, & lt & gt>Superscript x of (1) denotes the multiplicative group; f (x) =x n +1 is a cyclic polynomial; is distributed in the ring->The standard deviation is a discrete gaussian distribution of r,representing the coefficients at the ∈ ->Is a polynomial of (a).
Based on public key PK, private key SK and functionThe principal can obtain the key EK for the calculation F =f and key VK for authentication F ＝F。
Based on public key PK, private key SK and inputThe entruster can obtain the ciphertext corresponding to the input>Key for authentication VK X And a key DK for decryption X . The specific process is as follows:
the encryption key sk is obtained from a key generation algorithm that linearly homomorphic encrypts the LHE according to the public key pk=lhep. As shown in FIG. 1, tn columns of input X are divided into t blocks X 1 ,X 2 ,…,X t Each block X i Contains n columns. The input may be written as x= [ X ] 1 … X t ]Each blockIs a d x n matrix.
For each vectorPerforming an encryption algorithm of a linear homomorphic encryption LHE with sk as an encryption key to obtain ciphertext ++>Thus, each block ∈ ->Is encrypted as->The input X is encrypted as +.>The following assignments were made: VK (vK) X ＝σ X ，Dk X =sk. Decryption key DK X Is private to the principal.
Based on the key used for calculationAnd ciphertext->The trusted party can calculate the result ciphertext +.>
Based on keys for authenticationVK X ＝σ X And result ciphertext->The principal can obtain 1 (representing the result ciphertext sigma) Y Is correct) or 0 (representing the result ciphertext sigma) Y Is erroneous). The verification process is as follows:
from a collection(in order to match the specifications of the matrix multiplication, a matrix with 2tn elements is chosen here; the elements involved in the matrix multiplication are all from the group +.>) The vector r is chosen uniformly and randomly. Calculation of F (sigma) X r) and sigma Y r. If the two are equal, the commissioner obtains 1; if not, the delegate gets a 0.
Based on key DK for decryption X Sum result ciphertextThe delegate can actually calculate the result y=fx. The decryption process is as follows:
will input sigma Y Divided into t blocks by 2tn columnsEach block->Containing 2 columns. The input may be writtenEach block->Is an m x 2n matrix. For each vectorPerforming a decryption algorithm of the linear homomorphic encryption LHE with sk as decryption key to obtain ciphertext ++>Thus, each block ∈ ->Decrypted as +.>Ciphertext->Decrypted as +.>The matrix Y is the class prediction result of multiple classifiers on multiple pieces of private data.
The invention will be further described with the example of medical data classification prediction. The internet of things device (e.g., smart bracelet, smart appliance, etc.) may detect a plurality of pieces of private health data. A plurality of trained disease classifiers can be used to determine whether the disease is present. In this problem, the model parameters of one disease classifier can be regarded as one row vector, and a plurality of classifiers can constitute a matrix F. A private piece of health data of the principal can be regarded as a column vector, and a plurality of private pieces of data can be formed into a matrix X. The delegated computation FX can obtain the category prediction results of a plurality of disease classifiers on a plurality of pieces of private data at the same time.

Claims (4)

1. A safe and efficient method for entrusting the prediction of the class of the privacy data utilizes a given classifier to predict the class on the privacy data, obtain the prediction result Y of the class of a plurality of classifiers on a plurality of pieces of privacy data, set up model parameter of a plurality of classifiers to form matrix F, the privacy data of the entruster form matrix X, characterized by that, calculate FX and get the prediction result of class of a plurality of classifiers on a plurality of pieces of privacy data at the same time, at this moment, matrix F and matrix X are regarded as function and input in the multiplication operation of the matrix, comprising the following steps:
step 1, selecting a safety parameter lambda, and according to the safety parameter lambda and a function set Representing a set of m rows and d columns of the matrix, wherein the elements in the set come from the group +.> Representing an integer set, the principal gets the public key pk=lhep and initializes the private key SK to +.>The common parameter lhep= (p, q, n, f (x), χ) is input 1 by the linear homomorphic encryption LHE's parameter generation algorithm λ Obtained, wherein q is a prime number; />Is prime number->Is a multiplicative group of (a); f (x) =x n +1 is a cyclic polynomial; n=2 [logλ]-1 ；/>Is distributed in the ring->The standard deviation is a discrete gaussian distribution of r,representing the coefficients at the ∈ ->Is a polynomial of (2);
step 2, according to the public key PK, the private key SK and the functionThe principal gets the key EK for the calculation F =f and key VK for authentication F ＝F；
Step 3, according to the public key PK, the private key SK and the inputThe entruster obtains the ciphertext corresponding to the input>Key for authentication VK X And a key DK for decryption X ；
Step 4, according to the key used for calculationAnd ciphertext->The trusted party can calculate the result ciphertext +.>
Step 5, according to the key for verificationVK X ＝σ X And result ciphertext->The principal can get 1, representing the knotFruit ciphertext sigma Y Is correct, or 0, represents the result ciphertext sigma Y Is wrong;
step 6, according to the key DK for decryption X Sum result ciphertextThe delegate can actually calculate the result y=fx.
2. The method for delegating privacy data class prediction in a safe and efficient manner as claimed in claim 1, wherein step 3 comprises the steps of:
step 301, obtaining an encryption key sk by a key generation algorithm of the linear homomorphic encryption LHE according to the public key pk=lhep;
step 302, dividing tn columns of input X into t blocks X 1 ，X 2 ，...，X t Each block X i Containing n columns, then input X write X= [ X ] 1 … X t ]Each blockIs a d x n matrix, i.e. [ t ]]；
Step 303, for each vector obtained in step 302Performing an encryption algorithm of a linear homomorphic encryption LHE with sk as an encryption key to obtain ciphertext ++>And each block->Is encrypted asThe input X is encrypted as +.>
Step 304, performing the following assignment: VK (vK) X ＝σ X ，DK X =sk, wherein decryption key DK X Is private to the principal.
3. The method for delegating privacy data class prediction in a safe and efficient manner as recited in claim 1, wherein step 5 comprises the steps of:
step 501, slave setUniformly and randomly selecting a vector r;
step 501, calculate F (σ) x r) and sigma Y r, if F (sigma) X r) and sigma Y r is equal, the principal obtains 1, if F (σ X r) and sigma Y r is not equal, the principal gets 0.
4. The method for delegating privacy data class prediction in a safe and efficient manner as recited in claim 1, wherein step 6 comprises the steps of:
step 601, inputting sigma Y Divided into t blocks by 2tn columnsEach block->Containing 2n columns, then input sigma Y WritingEach block->Is an m x 2n matrix, i.epsilon.t]；
Step 601, for each vectorWith the encryption key sk as the decryption key, performing a decryption algorithm of the linear homomorphic encryption LHE to obtain ciphertext ++>And each block->Decrypted asCiphertext->Decrypted as +.>
CN202110752010.0A 2021-07-02 2021-07-02 Safe and efficient entrusted privacy data category prediction method Active CN113343277B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110752010.0A CN113343277B (en) 2021-07-02 2021-07-02 Safe and efficient entrusted privacy data category prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110752010.0A CN113343277B (en) 2021-07-02 2021-07-02 Safe and efficient entrusted privacy data category prediction method

Publications (2)

Publication Number Publication Date
CN113343277A CN113343277A (en) 2021-09-03
CN113343277B true CN113343277B (en) 2023-12-29

Family

ID=77482332

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110752010.0A Active CN113343277B (en) 2021-07-02 2021-07-02 Safe and efficient entrusted privacy data category prediction method

Country Status (1)

Country Link
CN (1) CN113343277B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101475747B1 (en) * 2014-01-22 2014-12-23 고려대학교 산학협력단 Method for an outsourcing multi-party computation using homomorphic encryption
CN109787743A (en) * 2019-01-17 2019-05-21 广西大学 A kind of full homomorphic cryptography method that can verify that based on matrix operation
CN110008717A (en) * 2019-02-26 2019-07-12 东北大学 Support the decision tree classification service system and method for secret protection
CN110826089A (en) * 2019-12-12 2020-02-21 四川大学 Verifiable outsourcing computing method, client and cloud computing system for realizing safety and high efficiency of large-scale matrix multiplication
CN112995189A (en) * 2021-03-08 2021-06-18 北京理工大学 Method for publicly verifying matrix multiplication correctness based on privacy protection

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101475747B1 (en) * 2014-01-22 2014-12-23 고려대학교 산학협력단 Method for an outsourcing multi-party computation using homomorphic encryption
CN109787743A (en) * 2019-01-17 2019-05-21 广西大学 A kind of full homomorphic cryptography method that can verify that based on matrix operation
CN110008717A (en) * 2019-02-26 2019-07-12 东北大学 Support the decision tree classification service system and method for secret protection
CN110826089A (en) * 2019-12-12 2020-02-21 四川大学 Verifiable outsourcing computing method, client and cloud computing system for realizing safety and high efficiency of large-scale matrix multiplication
CN112995189A (en) * 2021-03-08 2021-06-18 北京理工大学 Method for publicly verifying matrix multiplication correctness based on privacy protection

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于全同态加密的数据检验和可验证委托计算方案;李家;;北京电子科技学院学报(第04期);全文 *

Also Published As

Publication number Publication date
CN113343277A (en) 2021-09-03

Similar Documents

Publication Publication Date Title
Shan et al. Practical secure computation outsourcing: A survey
Yu et al. Enabling cloud storage auditing with verifiable outsourcing of key updates
Liu et al. Privacy-preserving outsourced calculation toolkit in the cloud
Lagendijk et al. Encrypted signal processing for privacy protection: Conveying the utility of homomorphic encryption and multiparty computation
Liang et al. Practical and secure SVM classification for cloud-based remote clinical decision services
Samanta et al. Cipher block chaining support vector machine for secured decentralized cloud enabled intelligent IoT architecture
CN114521319B (en) Lattice-based signature with uniform secret
JP2010039890A (en) Authentication terminal, authentication server, authentication system, authentication method and authentication program
Niu et al. Toward verifiable and privacy preserving machine learning prediction
Alberto Torres et al. Privacy-preserving biometrics authentication systems using fully homomorphic encryption
Tong et al. VFIRM: Verifiable fine-grained encrypted image retrieval in multi-owner multi-user settings
Huang et al. Secure and flexible cloud-assisted association rule mining over horizontally partitioned databases
Chen et al. Esb-fl: Efficient and secure blockchain-based federated learning with fair payment
Alharbi et al. Survey on homomorphic encryption and address of new trend
Ahmad et al. A secure network communication protocol based on text to barcode encryption algorithm
Ye et al. Anonymous biometric access control
CN112380404B (en) Data filtering method, device and system
Aljuhani et al. A deep learning integrated blockchain framework for securing industrial iot
Ren et al. How to extract image features based on co-occurrence matrix securely and efficiently in cloud computing
Sharma et al. Text security using 2d cellular automata rules
Morampudi et al. Multi-instance iris remote authentication using private multi-class perceptron on malicious cloud server
Rahulamathavan et al. Scalar product lattice computation for efficient privacy-preserving systems
CN113343277B (en) Safe and efficient entrusted privacy data category prediction method
Hamian et al. Blockchain-based User Re-enrollment for Biometric Authentication Systems
Arulananth et al. Multi party secure data access management in cloud using user centric block chain data encryption

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
GR01 Patent grant
GR01 Patent grant