CN104092686B - The method of secret protection and Data Access Security based on SVM classifier - Google Patents
The method of secret protection and Data Access Security based on SVM classifier Download PDFInfo
- Publication number
- CN104092686B CN104092686B CN201410332957.6A CN201410332957A CN104092686B CN 104092686 B CN104092686 B CN 104092686B CN 201410332957 A CN201410332957 A CN 201410332957A CN 104092686 B CN104092686 B CN 104092686B
- Authority
- CN
- China
- Prior art keywords
- svm classifier
- formula
- service provider
- key
- service
- 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.)
- Expired - Fee Related
Links
Landscapes
- Storage Device Security (AREA)
Abstract
The invention discloses a kind of secret protection based on SVM classifier and the method for Data Access Security; the method can not only realize the secret protection being related in SVM classifier; but also can realize safely outwards issuing SVM classifier; the method has security higher, and can protect the privacy of SVM classifier.The method can be widely applied for, and to SVM classifier secret protection and require related many application scenarios of safe issue SVM classifier parameter.
Description
Technical field
The present invention relates to information technology security fields, more particularly to a kind of secret protection sum based on SVM classifier
According to the method for secure access.
Background technology
At present, a main task of data mining is exactly to classify, training dataset of the sorting algorithm from known label
Acquistion of going to school is arrived, a disaggregated model, i.e. grader, to the data of later Unknown Label, is classified, and nowadays, is counted
Have been considered as being this privacy concern to one of personal sensitive information threat according to excavating, people is begun one's study number
According to the Privacy Protection of excavation applications, different sorting algorithms, there are different secret protection schemes, SVM is that classification is calculated at present
In method, a most popular sorting algorithm in terms of SVM secret protections, has there is many work, however, at present
The Privacy Protection of SVM, there is a problem, is not taken seriously also, in the grader that SVM study is obtained, comprising training data
Full instance, essentially, for some laws and the angle of business secret protection, the disaggregated model meeting of SVM
Destruction privacy concern, when grader is outwards announced, can show the personal sensitive data that training data is concentrated, such as, HIPAA
Legal requirement, medical data forbade externally issue before without suitable anonymization, therefore, in no secret protection
In the case of, which limits the grader that SVM study is obtained, the limitation externally issued.
It is assumed that a hospital or medical institutions, have collected substantial amounts of medical record data, the mechanism, it is desirable to utilize
These data collected learn to obtain a SVM classifier, using the SVM classifier, can predict a patient whether
This disease, comprising some medical record datas, if issuing this grader, can let out due in the grader that training is obtained
Leak the sensitive data of some patients, although, the identifier field of every record is removed, but every entity of record
Identity is likely to be identified from standard identifier, such as, sex, blood group, age, date of birth and postcode, by this
Individual example, is mainly described, if the SVM classifier that externally issue training is obtained, can leak the one of original training data concentration
A little data, can be related to Privacy Protection.
In 2011, Keng-Pei Lin et al. proposed a SVM approximate data.The SVM approximate data training in advance
Out SVM classifier, is processed the SVM classifier being previously obtained again afterwards, is protected with this in privacy of supporting vector
Hold.SVM can be externally issued in the case where supporting vector privacy is not leaked using the SVM approximate datas.Additionally, the SVM is near
There is close classification accuracy like algorithm and original SVM classifier.
By protecting the privacy content in grader, i.e. supporting vector, the grader of the protection privacy for as a result obtaining can be with
In the case where privacy is not leaked, externally announce.
SVM classifier with secret protection has many application scenarios, can be widely used, but, therewith
And come be exactly, safely external i.e. client requesting party, publisher server side trains the svm classifier with secret protection that obtains
Device.
It is assumed that be server B now with the provider of a service, and a requesting party for service, i.e. customer A, visitor
Family side A and server side B is, it is necessary to be communicated, server side B, first with the resource of oneself, is obtained with hidden by training
The SVM classifier of private protection, customer A, can to server side B, the training of request server side obtain with secret protection
SVM classifier.
Under an open and distributed environment, A and B will be communicated, certification and confidentiality, just be examined as needs
The two main problems considered, the key agreement protocol based on certification is the basis for building secure network environment, close by certification
Key agreement protocol, is between the participant that identity has been acknowledged in a communications system for the participant of communication provides authentication
Shared key is set up, for encrypting the message of transmission.
In addition, confidential information is not leaked in how ensureing communication process, the authenticity in authentication information source, it is ensured that information
Integrality and non repudiation, be exactly that network information security research needs the problem of research, the target of network security should expire
Foot:Identity reality, information confidentiality, information integrity, service availability, non-repudiation, system controllability and measurability, system are easy-to-use
Property, examination property etc., digital signature technology technology is one of important means of network security, it may ensure that information completely
Property, differentiate sender identity reality and non-repudiation.
The characteristic of digital signature includes:Identity differentiate, it is possible to authenticate the authenticity of wish and prevent from pretending to be;Non-repudiation
Property, undeniably preventing it from denying when information source;Data integrity protection, distorting or resetting less than data;Typically also make
With encryption technology protection information confidentiality, to prevent interception attack etc..
One of key technology as network security, digital signature also has very wide in the every field of social life
Application prospect.
The content of the invention
Visited the present invention solves the technical problem of a kind of secret protection and data safety based on SVM classifier is provided
The method asked, not only causes that there is SVM classifier protection privacy to be the purpose of supporting vector, so that being asked to client
When Fang Fabu has the SVM classifier of secret protection, the carrying out transmission SVM model parameters of safety.
In order to solve the above technical problems, one aspect of the present invention is:
A kind of method of secret protection and Data Access Security based on SVM classifier is provided, i.e. service requester A is to clothes
Business provider B requests obtain the SVM classifier with secret protection of service provider B training, and its specific steps includes:
One, key agreements:
A. establishment stage, i.e., generate the parameter and its own public private key pair of system in the PKG files of service provider B:
P ∈ G, G are set and represent prime number set, p is an element in the prime number set, i.e. p is a prime number;Using bilinear map
Obtain e:G×G→GT;Two maker g of selection random in PKG files (in step (1), i.e. key agreement step,
G represents generation unit) and maker h, and maker g, h ∈ G;It is α ∈ Z that master private key is set in PKG filesp, p represent element
Number, ZpIt is expressed as a residual class ring of mould p;Hash function is setService requester A kimonos is set
Shared session key between business provider B is sk, and sets k1=| sk |;Setting the public key of PKG is<G, g1, h, gT, H>, its
In, g1=gα∈ G, gT=e (g, g) ∈ GT;
B. key generation phase, i.e., be service requester A generation keys in the PKG files of service provider B:Clothes are set
The entitled user of user of business requesting party A, user identity is ID ∈ Zp(ID≠α);Generate a r at random in PKG filesID∈
Zp, and it is d to export a private keyID=<rID,hID>, whereinDuring service-oriented requesting party A, if
PutDuring service-oriented provider B, set
C. key agreement phase, i.e. service requester A and service provider B, by implicit key authentication, set up one
Shared shared session key, its step includes:
1.1) user identity for setting service requester A is IDA, the user identity for setting service provider B is IDB, service
Requesting party A selects an x ∈ Zp, variable x herein is mono- real variable of real number field R, and the variable is the residue class of mould p
A variable in ring Zp, and calculaterAIt is
One stochastic variable of service provider A generations, then TAIt is sent to service provider B;
1.2) service provider B selects a y ∈R Zp, variable y is a real number, and the variable belongs to Zp, and is counted
CalculaterBIt is a random change of service provider B generations
Amount, then TBIt is sent to service requester A;
1.3) service requester A calculates following formula:
1.4) service provider B calculates following formula:
From the step in 1.3 and 1.4:
And formula (1) and formula (2) are equal, so obtaining:
Obtain between service requester A and service provider B
Shared session key skA=skB;
Two, protect the SVM of privacy
Training obtains the SVM classifier with secret protection, i.e. service provider B and utilizes training dataset, and training is obtained
One initial SVM classifier, then enters line translation to this grader again, obtains a svm classifier with secret protection
Device:
2.1) service provider B obtains an initial svm classifier first with the training dataset of oneself by training
Device, the initial SVM classifier is based on gaussian kernel function, the form of its decision function:
WhereinNuclear parameter gh∈ R, nuclear parameter gh>0, deviation variables b ∈
R, R are any rational, { (SV1,y1),...,(SVm,ym) it is supporting vector, yiIt is corresponding SViLabel, { α1,...,αm}
It is support variable corresponding with supporting vector, m represents that training data concentrates the number of the supporting vector for including, yi∈{+1,-1},
SVi∈RΝ,X∈RN, N represents that the dimension of training data, i.e. training sample are the N-dimensional variables in real number field R, and C is cost parameter,
And C >=0, private data is SVi;
2.2) in decision function (1), | | SVi-X||2Represent a test sample X and supporting vector SViThe distance between
Quadratic power, set gaussian kernel function K (X, Y)=exp(-gh||X-Y||2), wherein X ∈ RN,Y∈RN, and according to Gaussian kernel letter
Number and decision function (1), obtain | | SVi-X||2=| | SVi||2-2(SVi.X)+||X||2;By decision function (1) change of equal value
It is changed to:
2.3) the exp (- g in formula (2)h||SVi||2) and exp (2ghSVi.X private data) is contained.According to formula
(2), by exp (- gh||SVi||2) be deformed into:ci=αiyiexp(-gh||SVi||2), wherein, i=(1, m), and by decision function
(1) it is deformed into:
2.4) in decision function (3), exp (2g are obtainedhSViThe representation of infinite series .X):
By exp (2ghSVi.X infinite series), obtainAnd will
In decision function (3)It is deformed into:
So that supporting vector only occurs in (SVi·X)dIn formula;
2.5) by using monomial Feature Mapping, (the SV in formula (5) is seti·X)d=Φd(SVi)·Φd(X),
Wherein, Φd() represents the monomial mapping of d- ranks, and formula (5) is deformed into:
2.6) in (6) formula, setIn each wdIn, all of spy
Levy vector and be mapped to a monomial feature space for d ranks;
SetFormula (7.1) and (7.2) formula are updated in formula (6), by formula
Sub (6) are deformed into:
Formula (8) formula is updated in decision function (3) again, decision function (3) is deformed into:
Obtain the SVM classifier with secret protection;
Decision function (9) is exactly the decision function for protecting privacy, and it is supporting vector that decision function needs data to be protected, certainly
The linear combination that plan function passes through supporting vector, can destroy wdPrivacy content, in original SVM classifier, in order to point
Class needs to know supporting vector that the supporting vector is exactly original private data, but in decision function (9), it is only necessary to know
Road wdJust can be so that supporting vector is mapped to monomial feature by decision function (9) first, afterwards, then monomial feature is entered
Row linear combination and then obtain wd, like this, as outwards issue grader SVM, the parameter of required issue includes wd(d
=1 ..., du), w0, b and gh.And the data of these issues will not relate to any private data, i.e. supporting vector { (SV1,
y1),...,(SVm,ym)}.Therefore, externally issue decision function will not reveal any private data;
Three, are encrypted, i.e., using the shared session key, symmetric encipherment algorithm and AES encryption algorithm obtained in key agreement,
Parameter w in the SVM classifier with secret protection obtained to step 2d(d=1 ..., du), d is an integer, and d is represented
Exponent number, w after Taylor expansion is launched in step 20, b and ghIt is encrypted, obtains ciphertext m ', ciphertext now represents these
The set constituted after variable encryption;
Four, generate digital signature, i.e., service provider B is to comprising parameter wd(d=1 ..., du)、w0, b and ghPlaintext
Ms, ms represent a set comprising above-mentioned variable, generate corresponding digital signature (s, e);
Five, are sent to client requesting party, i.e., obtained in the ciphertext m ' that service provider B will be obtained in step 3 and step 4
Plaintext ms digital signature (s, e), send jointly to service requester A;
Six, are decrypted, i.e. service requester A receives ciphertext m ' and the digital signature (s, e) that service provider B sends, and tests
Demonstrate,prove the correctness of digital signature;When digital signature is correct, the key sk obtained in obtaining step oneA, then ciphertext m ' is entered
Row decryption obtains plaintext ms.
In a preferred embodiment of the invention, in step 4, the Digital Signature Algorithm for using is that schnorr numerals are signed
Name algorithm.
In a preferred embodiment of the invention, the following institute of the specific method and step of schnorr Digital Signature Algorithms
Show:
(1) parameter selection:All users are allowed to can determine whether that, using a prime number set G, prime number set G has a Prime Orders q
The maker of generation unit g, i.e. G be g, and the rank q of maker g is a prime number, and all of user determine one plus
Close hash function, i.e. H:{0,1}*→Zq;
(2) mark:Exponentiation represents the repeated application of group operation;Side by side represent residue class on multiplying or
It is the application of group operation;Subtraction represents the subtraction in equivalence group group;M ∈ { 0,1 } is set*, it represents limited in the character string of position
Sequence;S, e, e are setv∈ZqWithZp is residue class mould set,It is the multiplicative group of mould integer, q is prime number,
(3) key generation:From in the set G for allowing, a private signature key x and publicly-owned authentication secret y=is selected
gx;
(4) a message M is signed:From in a set for permission, a random number k is selected;R=g is setkWith
E=H (M | | r), | | the effect of connection is represented, r represents a position character string;Set s=(k-xe), then digital signature be exactly (s,
Also, s, e ∈ Z e),q;
(5) verify:R is setv=gsye, ev=H (M | | rv), work as evDuring=e, then digital signature is regarded as by recognizing
Demonstrate,prove;
(6) proof of correctness:If the information of signature and the information of checking are equal, r is understoodv=gsge=gk -xegxe=gk=r, so ev=H (M | | rv)=H (M | | r)=e;Publicly-owned element is:G, g, q, y, s, e, r, privately owned element
It is:k,x.
The beneficial effects of the invention are as follows:Can not only realize the secret protection being related in SVM classifier, but also can be with
Realize safely outwards issuing SVM classifier, the method has security higher, and can protect the hidden of SVM classifier
It is private.
Brief description of the drawings
Technical scheme in order to illustrate more clearly the embodiments of the present invention, below will be to that will make needed for embodiment description
Accompanying drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing, wherein:
Fig. 1 is the schematic flow sheet of the method for the secret protection based on SVM classifier of the invention and Data Access Security;
Fig. 2 is the heretofore described schematic flow sheet trained and obtain the SVM classifier with secret protection.
Specific embodiment
The technical scheme in the embodiment of the present invention will be clearly and completely described below, it is clear that described implementation
Example is only a part of embodiment of the invention, rather than whole embodiments.Based on the embodiment in the present invention, this area is common
All other embodiment that technical staff is obtained under the premise of creative work is not made, belongs to the model of present invention protection
Enclose.
Fig. 1 and Fig. 2 is referred to, the embodiment of the present invention includes:
A kind of method of secret protection and Data Access Security based on SVM classifier, i.e. service requester A is carried to service
Supplier B requests obtain the SVM classifier with secret protection of service provider B training, Mengbo Hou et al. and propose one
The key agreement protocol of the identity-based certification of individual enhanced two side, the agreement is higher in the security of entrusted management model, proposition
Agreement, there is provided almost all of known security attribute, especially preferable forward secrecy attribute and known interim spy
Determine the information privacy attribute of session, at this stage, A and B wants to set up a shared session key among each other, and
PKG, then be responsible for the generation and distribution of private key for user by using the master private key of oneself, and the protocol package contains three phases, builds
Vertical stage, key generation phase, key agreement phase;
Specific steps include:
One, key agreements:
A. establishment stage, i.e., generate the parameter and its own public private key pair of system in the PKG files of service provider B:
P=G (G represents prime number set, and p is an element in the prime number set, i.e. p is a prime number) is set;Reflected using bilinearity
Penetrate acquisition e:G×G→GT;Random two generation unit g of selection in PKG files (note, in the key agreement step, become
Amount g is expressed as generation unit, and the variable g in step 2 then represents the parameter of gaussian kernel function) and h, and g, h ∈ G;In PKG files
Middle setting master private key is α ∈ Zp(p represents prime number, and Zp is expressed as a residual class ring of mould p);Hash function is setThe shared session key set between service requester A and service provider B is sk, and sets k1
=| sk |;Setting the public key of PKG is<G, g1, h, gT, H>, wherein, g1=gα∈ G, gT=e (g, g) ∈ GT;
B. key generation phase, i.e., be service requester A generation keys in the PKG files of service provider B:Clothes are set
The entitled user of user of business requesting party A, user identity is ID ∈ Zp(ID≠α);Generate a r at random in PKG filesID∈
Zp, and it is d to export a private keyID=<rID,hID>, whereinDuring service-oriented requesting party A,
SetDuring service-oriented provider B, set
C. key agreement phase, i.e. service requester A and service provider B, by implicit key authentication, set up one
Shared shared session key, its step includes:
1.1) user identity for setting service requester A is IDA, the user identity for setting service provider B is IDB, service
Requesting party A selects an x ∈R Zp, (variable x herein is mono- real variable of real number field R, and the variable is the residue of mould p
A variable in class ring Zp) and calculate
(rAIt is a stochastic variable of service provider A generations), then TAIt is sent to service provider B;
1.2) service provider B selects a y ∈ R Zp, (variable y herein is a real number, and the variable belongs to
Zp) and calculate(rBIt is service provider B
One stochastic variable of generation), then TBIt is sent to service requester A;
1.3) service requester A calculates following formula:
1.4) service provider B calculates following formula:
From the step in 1.3 and 1.4:
And formula (1) and formula (2) are equal, so obtaining:
Obtain service requester A and service provider B it
Between shared session key skA=skB;
Two, training obtains the SVM classifier with secret protection, i.e. service provider B utilizes training dataset, trains
To an initial SVM classifier, line translation then is entered to this grader again, obtain a SVM with secret protection and divide
Class device.In 2011, Keng-Pei Lin et al. proposed a SVM approximate data.The SVM approximate datas training in advance is out
SVM classifier, is processed the SVM classifier being previously obtained again afterwards, and the privacy content of supporting vector is protected with this.Can
SVM is externally issued in the case where supporting vector privacy is not leaked with using the SVM approximate datas.Additionally, the SVM approximate datas
There is close classification accuracy with original SVM classifier.
2.1) service provider B obtains an initial svm classifier first with the training dataset of oneself by training
Device, the initial SVM classifier is based on gaussian kernel function, the form of its decision function:
WhereinNuclear parameter gh∈ R, nuclear parameter gh>0, deviation
Variable b ∈ R, R are any rational, { (SV1,y1),...,(SVm,ym) it is supporting vector, yiIt is corresponding SViLabel,
{α1,...,αmIt is support variable corresponding with supporting vector, m represents that training data concentrates the number of the supporting vector for including, yi
∈{+1,-1},SVi∈RΝ,X∈RN, N represents that the dimension of training data, i.e. training sample are the N-dimensional variables in real number field R, and C is
Cost parameter, and C >=0, private data are SVi.But their class label, yiGenerally it is not related to privacy.Our purpose
Exactly these private datas are irreversibly protected by one;
2.2) in decision function (1), | | SVi-X||2Represent a test sample X and supporting vector SViThe distance between
Quadratic power, gaussian kernel function K (X, Y)=exp (- g is seth||X-Y||2), wherein X ∈ RN,Y∈RN, and according to Gaussian kernel letter
Number and decision function (1), obtain | | SVi-X||2=| | SVi||2-2(SVi.X)+||X||2;By decision function (1) change of equal value
It is changed to:
2.3) the exp (- g in formula (2)h||SVi||2) and exp (2ghSVi.X private data) is contained.According to formula
(2), by exp (- gh||SVi||2) be deformed into:ci=αiyiexp(-gh||SVi||2), wherein, i=(1, m),
And be deformed into decision function (1):
2.4) in decision function (3), exp (2g are obtainedhSViThe representation of infinite series .X):
By exp (2ghSVi.X infinite series), obtainAnd will
In decision function (3)It is deformed into:
So that supporting vector only occurs in (SVi·X)dIn formula;
2.5) by using monomial Feature Mapping, (the SV in formula (5) is seti·X)d=Φd(SVi)·Φd(X),
Wherein, Φd() represents the monomial mapping of d- ranks, and formula (5) is deformed into:
2.6) in (6) formula, setIn each wdIn, all of spy
Levy vector and be mapped to a monomial feature space for d ranks;
SetFormula (7.1) and (7.2) formula are updated in formula (6), by formula
Sub (6) are deformed into:
Formula (8) formula is updated in decision function (3) again, decision function (3) is deformed into:
Obtain the SVM classifier with secret protection;
Decision function (9) is exactly the decision function for protecting privacy, and it is supporting vector that decision function needs data to be protected, certainly
The linear combination that plan function passes through supporting vector, can destroy wdPrivacy content, in original SVM classifier, in order to point
Class needs to know supporting vector that the supporting vector is exactly original private data, but in decision function (9), it is only necessary to know
Road wdJust can be so that supporting vector is mapped to monomial feature by decision function (9) first, afterwards, then monomial feature is entered
Row linear combination and then obtain wd, like this, as outwards issue grader SVM, the parameter of required issue includes wd(d=
1,…,du), w0, b and gh.And the data of these issues will not relate to any private data, i.e. supporting vector { (SV1,
y1),...,(SVm,ym)}.Therefore, externally issue decision function will not reveal any private data.
Three, are encrypted, i.e., using the shared session key, symmetric encipherment algorithm and AES encryption algorithm obtained in key agreement,
Parameter w in the SVM classifier with secret protection obtained to step 2d(d=1 ..., du) (d is an integer, represents step
In rapid two Taylor expansion launch after exponent number), w0, b and ghIt is encrypted, (ciphertext herein represents that these become to obtain ciphertext m '
The set constituted after amount encryption);
Four, generate digital signature, i.e., service provider B is to comprising parameter wd(d=1 ..., du)、w0, b and ghPlaintext ms
(plaintext ms herein represents a set comprising these variables), generates corresponding digital signature (s, e);
Five, are sent to client requesting party, i.e., obtained in the ciphertext m ' that service provider B will be obtained in step 3 and step 4
Plaintext ms digital signature (s, e), send jointly to service requester A.
Six, are decrypted, i.e. service requester A receives ciphertext m ' and the digital signature (s, e) that service provider B sends, and tests
Demonstrate,prove the correctness of digital signature;When digital signature is correct, the key sk obtained in obtaining step oneA, then ciphertext m ' is entered
Row decryption obtains plaintext ms.
In step 4, the Digital Signature Algorithm for using is schnorr Digital Signature Algorithms.Its security be based on from
The intractability of logarithm is dissipated, schnorr Digital Signature Algorithms are considered as simplest digital signature scheme.Random oracle model is demonstrate,proved
Bright schnorr Digital Signature Algorithms are safe.Additionally, the efficiency of schnorr Digital Signature Algorithms is very high, can generate very short
Signature.
In a preferred embodiment of the present invention, the following institute of the specific method and step of schnorr Digital Signature Algorithms
Show:(1) parameter selection:All users are allowed to can determine whether that, using a prime number set G, prime number set G has a life of Prime Orders q
Cheng Yuan (generation unit can obtain all elements in set G by certain computing), and all of user determines that one adds
Close hash function, i.e. H:{0,1}*→Zq;(2) mark:Exponentiation represents the repeated application of group operation;Residue class is represented side by side
On multiplying or group operation application;Subtraction represents the subtraction in equivalence group group;M ∈ { 0,1 } is set*, its expression
Finite sequence in the character string of position;S, e, e are setv∈ZqWithQ is residue class mould set,It is multiplying for mould integer
Method group, q is prime number,Key is generated:Selection one private signature key x and publicly-owned
Authentication secret y=gx;(4) a message M is signed:From in a set for permission, a random number k is selected;R is set
=gkWith e=H (M | | r), | | represent the effect of connection, r represents a position character string;S=(k-xe) is set, then digital signature
It is exactly (s, e), also, s, e ∈ ZqIf, q<2160, then the length of digital signature may be adjusted to 40 bytes;(5) verify:
R is setv=gsye, ev=H (M | | rv), work as evDuring=e, then digital signature is regarded as by have authenticated;(6) correctness
Prove:If the information of signature and the information of checking are equal, r is understoodv=gsge=gk-xegxe=gk=r, so ev=H
(M||rv)=H (M | | r)=e;Publicly-owned element is:G, g, q, y, s, e, r, privately owned element is:k,x.
The purpose of the present embodiment has at 2 points:One is the SVM classifier obtained to initial training, is converted again, is not being lost
On the premise of classifier performance, secret protection is carried out to SVM classifier.The SVM classifier for finally obtaining, can effectively protect
The privacy content of SVM.Two is to set up safely communication protocol between the two parties.
The conventional Privacy Protection on SVM is all to carry out secret protection to SVM on training dataset, and is closed
Parameter in svm classifier model, i.e. supporting vector, carry out the problem of secret protection, are not studied also.So as to close
Also it is not suggested in the agreement of customer issue protection privacy, i.e. supporting vector.The purpose of the present invention is first to train
To the SVM classifier with secret protection, then, then the agreement for passing through safety, to client requesting party, safely transmit SVM's
Model parameter.
Specifically, the agreement can be divided into four-stage, and first stage includes step 1, be customer and service side it
Between, the key agreement based on certification is carried out, second stage includes step 2, is request of the service side according to client, and training is obtained
The corresponding SVM classifier with secret protection, three phases include step 3,4, and four-stage includes step 5,6,7,
In third and fourth stage, safe communication is carried out between both sides.
Key agreement protocol (the bag of the new opportunity certification between first stage, both sides by being proposed in the 1st step
Containing three phases, establishment stage, key generation phase, key agreement phase), it is that one is obtained between both sides the agreement is completed
Individual shared session key, i.e. skA=skB, in second stage, in this process, server side can utilize the training of oneself
Data set, training obtains an initial SVM classifier, then enters line translation to this grader again, obtains one with hidden
The SVM classifier of private protection.In three phases, service side generates digital signature, then uses the first rank to plaintext m first
The private key that section is produced is encrypted to plaintext m, generates ciphertext m ', afterwards again rises the digital signature one of ciphertext m ' and plaintext m
Customer is given, four-stage, customer is decrypted to ciphertext first, then whether checking information is from service hair hair again
Bring, if so, just receiving, otherwise just refuse.
The correctness and security to protocol frame are analyzed below.
It is of the invention comprehensively to have used an improved key agreement protocol based on certification, improved SVM secret protections
Algorithm and an enhanced digital signature protocol.The correctness of the calculating on protocol frame of the invention, and the safety for communicating
Property problem analysis, just do not list one by one, specifically may be referred to it is above-mentioned correlation bibliography, herein simply
Introduce, the analysis of the correctness of the improved key agreement protocol based on certification that the present invention is used, in step 1,
A can calculate following formula:
B can calculate following formula:
When the agreement performs end, A and B can obtain an identical session key, because:
Obviously, (1) and (2) be it is equal, similarly,
So, we can be obtained by skA=skB, analysis above
Illustrate, in the end of the key agreement protocol, A and B can obtain the shared session key of identical, so as to illustrate that the key is assisted
Business's agreement is correct.
The beneficial effect of method of secret protection and Data Access Security of the present invention based on SVM classifier is:Can not only
The secret protection being related in SVM classifier is enough realized, but also can realize safely outwards issuing SVM classifier, the party
Method has security higher, and can protect the privacy of SVM classifier.The method can be widely applied for, with svm classifier
Device secret protection and require related many application scenarios of safe issue SVM classifier parameter.
Embodiments of the invention are the foregoing is only, the scope of the claims of the invention is not thereby limited, it is every to utilize this hair
Equivalent structure or equivalent flow conversion that bright description is made, or directly or indirectly it is used in other related technology necks
Domain, is included within the scope of the present invention.
Claims (3)
1. a kind of method of secret protection and Data Access Security based on SVM classifier, i.e. service requester A is provided to service
Square B requests obtain the SVM classifier with secret protection of service provider B training, it is characterised in that specific steps include:
One, key agreements:
A. establishment stage, i.e., generate the parameter and its own public private key pair of system in the PKG files of service provider B:P is set
∈ G, G represent prime number set, and p is an element in the prime number set, i.e. p is a prime number;Obtained using bilinear map
e:G×G→GT;Random selection two maker g and maker h in PKG files, and maker g, h ∈ G;In PKG files
Middle setting master private key is α ∈ Zp, p represents prime number, ZpIt is expressed as a residual class ring of mould p;Hash function is setThe shared session key set between service requester A and service provider B is sk, and sets k1
=| sk |;Setting the public key of PKG is<G, g1, h, gT, H>, wherein, g1=gα∈ G, gT=e (g, g) ∈ GT;
B. key generation phase, i.e., be service requester A generation keys in the PKG files of service provider B:The service of setting please
The entitled user of user of the side of asking A, user identity is ID ∈ Zp(ID≠α);Generate a r at random in PKG filesID∈Zp, and
And one private key of output is dID=<rID,hID>, whereinDuring service-oriented requesting party A, setDuring service-oriented provider B, set
C. key agreement phase, i.e. service requester A and service provider B, by implicit key authentication, set up one altogether
The shared session key enjoyed, its step includes:
1.1) user identity for setting service requester A is IDA, the user identity for setting service provider B is IDB, service request
Square A selects an x ∈ Zp, variable x herein is mono- real variable of real number field R, and the variable is the residual class ring Z of mould pp
In a variable, and calculaterAIt is that service is carried
One stochastic variable of supplier A generations, then TAIt is sent to service provider B;
1.2) service provider B selects a y ∈ Zp, variable y is a real number, and the variable belongs to Zp, and is calculatedrBIt is a stochastic variable of service provider B generations,
Then TBIt is sent to service requester A;
1.3) service requester A calculates following formula:
1.4) service provider B calculates following formula:
From the step in 1.3 and 1.4:
And formula (1) and formula (2) are equal, so obtaining: The shared session obtained between service requester A and service provider B is close
Key skA=skB;
Two, protect the SVM of privacy
Training obtains the SVM classifier with secret protection, i.e. service provider B and utilizes training dataset, and training obtains one
Initial SVM classifier, then enters line translation to this grader again, obtains a SVM classifier with secret protection:
2.1) service provider B obtains an initial SVM classifier first with the training dataset of oneself by training,
The initial SVM classifier is based on gaussian kernel function, the form of its decision function:
WhereinI=1 ..., m, nuclear parameter gh∈ R, nuclear parameter gh>0, deviation variables b ∈ R, R
It is any rational, { (SV1,y1),...,(SVm,ym) it is supporting vector, yiIt is corresponding SViLabel, { α1,...,αmBe with
The corresponding support variable of supporting vector, m represents that training data concentrates the number of the supporting vector for including, yi∈{+1,-1},SVi∈
RΝ,X∈RN, N represents that the dimension of training data, i.e. training sample are the N-dimensional variables in real number field R, and C is cost parameter, and C >=
0, private data is SVi;
2.2) in decision function (1), | | SVi-X||2Represent a test sample X and supporting vector SViThe distance between two
Power, sets gaussian kernel function K (X, Y)=exp (- gh||X-Y||2), wherein X ∈ RN,Y∈RN, and according to gaussian kernel function and
Decision function (1), obtains | | SVi-X||2=| | SVi||2-2(SVi.X)+||X||2;By being transformed to for decision function (1) equivalence:
2.3) the exp (- g in formula (2)h||SVi||2) and exp (2ghSVi.X private data) is contained, according to formula (2), will
exp(-gh||SVi||2) be deformed into:ci=αiyi exp(-gh||SVi||2), wherein, i=(1, m), and decision function (1) is become
Shape is:
2.4) in decision function (3), exp (2g are obtainedhSViThe representation of infinite series .X):
By exp (2ghSVi.X infinite series), obtainAnd will be certainly
In plan function (3)It is deformed into:
So that supporting vector only occurs in (SVi·X)dIn formula;
2.5) by using monomial Feature Mapping, (the SV in formula (5) is seti·X)d=Φd(SVi)·Φd(X), wherein,
Φd() represents the monomial mapping of d- ranks, and formula (5) is deformed into:
2.6) in (6) formula, setIn each wdIn, all of characteristic vector
It has been mapped to a monomial feature space for d ranks;
SetFormula (7.1) and (7.2) formula are updated in formula (6), by formula (6)
It is deformed into:
Formula (8) formula is updated in decision function (3) again, decision function (3) is deformed into:
Obtain the SVM classifier with secret protection;
Decision function (9) is exactly the decision function for protecting privacy, and it is supporting vector, decision-making letter that decision function needs data to be protected
The linear combination that number passes through supporting vector, can destroy wdPrivacy content, in original SVM classifier, in order to classify need
It is to be understood that supporting vector, the supporting vector is exactly original private data, but in decision function (9), it is only necessary to know wd
Just can be so that supporting vector is mapped to monomial feature by decision function (9) first, afterwards, then enters line to monomial feature
Property combination and then obtain wd, like this, as outwards issue grader SVM, the parameter of required issue includes wd(d=
1,…,du), w0, b and gh, and the data of these issues will not relate to any private data, i.e. supporting vector { (SV1,
y1),...,(SVm,ym), therefore, externally issue decision function will not reveal any private data;
Three, are encrypted, i.e., using the shared session key, symmetric encipherment algorithm and AES encryption algorithm obtained in key agreement, to step
Parameter w in rapid two SVM classifiers with secret protection for obtainingd(d=1 ..., du), d is an integer, and d represents step
Exponent number, w after Taylor expansion is launched in two0, b and ghIt is encrypted, obtains ciphertext m ', ciphertext now represents these variables
The set constituted after encryption;
Four, generate digital signature, i.e., service provider B is to comprising parameter wd(d=1 ..., du)、w0, b and ghPlaintext ms, ms table
Show a set comprising above-mentioned variable, generate corresponding digital signature (s, e);
Five, are sent to client requesting party, i.e., service provider B will be obtained in step 3 ciphertext m' and obtained in step 4 it is bright
The digital signature (s, e) of literary ms, sends jointly to service requester A;
Six, are decrypted, i.e. service requester A receives ciphertext m' and the digital signature (s, e) that service provider B sends, and verifies number
The correctness of word signature;When digital signature is correct, the key sk obtained in obtaining step oneA, then ciphertext m' is solved
It is close to obtain plaintext ms.
2. the method for secret protection and Data Access Security based on SVM classifier according to claim 1, its feature exists
In in step 4, the Digital Signature Algorithm for using is schnorr Digital Signature Algorithms.
3. the method for secret protection and Data Access Security based on SVM classifier according to claim 2, its feature exists
In the specific method and step of schnorr Digital Signature Algorithms is as follows:
(1) parameter selection:All users are allowed to can determine whether that, using a prime number set G, prime number set G has a life of Prime Orders q
Maker into first g, i.e. G is g, and the rank q of maker g is a prime number, and all of user determines that an encryption dissipates
Array function, i.e. H:{0,1}*→Zq;
(2) mark:Exponentiation represents the repeated application of group operation;The multiplying in residue class or group behaviour are represented side by side
The application of work;Subtraction represents the subtraction in equivalence group group;M ∈ { 0,1 } is set*, it represents the finite sequence in the character string of position;If
Put s, e, ev∈ZqWithZp is residue class mould set,It is the multiplicative group of mould integer, q is prime number,Y,
R, rv∈G;
(3) key generation:From in the set G for allowing, a private signature key x and publicly-owned authentication secret y=g is selectedx;
(4) a message M is signed:From in a set for permission, a random number k is selected;R=g is setkAnd e=H
(M | | r), | | the effect of connection is represented, r represents a position character string;S=(k-xe) is set, then digital signature is exactly (s, e),
Also, s, e ∈ Zq;(5) verify:R is setv=gsye, ev=H (M | | rv), work as evDuring=e, then digital signature is regarded as
By have authenticated;
(6) proof of correctness:If the information of signature and the information of checking are equal, r is understoodv=gsge=gk-xegxe
=gk=r, so ev=H (M | | rv)=H (M | | r)=e;Publicly-owned element is:G, g, q, y, s, e, r, privately owned element is:
k,x。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410332957.6A CN104092686B (en) | 2014-07-14 | 2014-07-14 | The method of secret protection and Data Access Security based on SVM classifier |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410332957.6A CN104092686B (en) | 2014-07-14 | 2014-07-14 | The method of secret protection and Data Access Security based on SVM classifier |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104092686A CN104092686A (en) | 2014-10-08 |
CN104092686B true CN104092686B (en) | 2017-06-09 |
Family
ID=51640367
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410332957.6A Expired - Fee Related CN104092686B (en) | 2014-07-14 | 2014-07-14 | The method of secret protection and Data Access Security based on SVM classifier |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104092686B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105447526A (en) * | 2015-12-15 | 2016-03-30 | 国网智能电网研究院 | Support vector machine based power grid big data privacy protection classification mining method |
CN105718948A (en) * | 2016-01-20 | 2016-06-29 | 江南大学 | Privacy protection SVM classification method based on information concentration |
CN106411955B (en) * | 2016-12-01 | 2019-07-23 | 微鲸科技有限公司 | A kind of account generation method based on mould m congruence class ring |
CN108520181B (en) * | 2018-03-26 | 2022-04-22 | 联想(北京)有限公司 | Data model training method and device |
CN109360611B (en) * | 2018-09-25 | 2021-05-18 | 湖北工业大学 | Privacy protection data mining system and method based on medical big data |
CN111753885B (en) * | 2020-06-09 | 2023-09-01 | 华侨大学 | Privacy enhanced data processing method and system based on deep learning |
CN115801449B (en) * | 2023-01-09 | 2023-05-26 | 深圳市迪博企业风险管理技术有限公司 | Privacy protection method, system and readable storage medium for risk assessment data |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004046849A2 (en) * | 2002-08-09 | 2004-06-03 | Rsa Security, Inc. | Cryptographic methods and apparatus for secure authentication |
CN102983971A (en) * | 2012-10-10 | 2013-03-20 | 中国科学技术大学苏州研究院 | Certificateless signature algorithm for user identity authentication in network environment |
CN103117858A (en) * | 2013-01-22 | 2013-05-22 | 河海大学 | Signature ring signature method provided with specified revocability |
-
2014
- 2014-07-14 CN CN201410332957.6A patent/CN104092686B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004046849A2 (en) * | 2002-08-09 | 2004-06-03 | Rsa Security, Inc. | Cryptographic methods and apparatus for secure authentication |
CN102983971A (en) * | 2012-10-10 | 2013-03-20 | 中国科学技术大学苏州研究院 | Certificateless signature algorithm for user identity authentication in network environment |
CN103117858A (en) * | 2013-01-22 | 2013-05-22 | 河海大学 | Signature ring signature method provided with specified revocability |
Non-Patent Citations (3)
Title |
---|
On the Design and Analysis of the Privacy-Preserving SVM Classifier;Keng-Pei Lin et al;《IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING》;20111130(第11(2011)期);全文 * |
基于身份的认证密钥协商协议的安全分析与改进;汪小芬等;《通信学报》;20081231(第12(2008)期);全文 * |
身份基认证密钥协商协议的分析与改进;侯孟波;《计算机工程与应用》;20100731(第07(2010)期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN104092686A (en) | 2014-10-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104092686B (en) | The method of secret protection and Data Access Security based on SVM classifier | |
ES2687182T3 (en) | Determine a common secret for the secure exchange of information and hierarchical and deterministic cryptographic keys | |
CN105262843B (en) | A kind of anti-data-leakage guard method for cloud storage environment | |
Nagaraju et al. | Trusted framework for online banking in public cloud using multi-factor authentication and privacy protection gateway | |
CN109583219A (en) | A kind of data signature, encryption and preservation method, apparatus and equipment | |
CN106850656B (en) | Multi-user's file-sharing control method under a kind of cloud environment | |
CN104038493B (en) | Bilinear pairing-free cloud storage data security audit method | |
CN110069946A (en) | A kind of Security Index system based on SGX | |
Li et al. | Healthfort: A cloud-based ehealth system with conditional forward transparency and secure provenance via blockchain | |
Xin et al. | Identity-based quantum designated verifier signature | |
Tian et al. | A provably secure and public auditing protocol based on the bell triangle for cloud data | |
Habib et al. | A Blockchain-based Technique to Prevent Grade Tampering: A University Perspective | |
CN110519040B (en) | Anti-quantum computation digital signature method and system based on identity | |
CN116827670A (en) | Intelligent medical data security sharing method based on national cryptographic algorithm | |
Singamaneni et al. | [Retracted] An Enhanced Dynamic Nonlinear Polynomial Integrity‐Based QHCP‐ABE Framework for Big Data Privacy and Security | |
Salvakkam et al. | An improved lattice based certificateless data integrity verification techniques for cloud computing | |
Zhu et al. | A secure data sharing scheme with designated server | |
Ding et al. | Secure Multi‐Keyword Search and Access Control over Electronic Health Records in Wireless Body Area Networks | |
Hasan et al. | Efficient mutual authentication using Kerberos for resource constraint smart meter in advanced metering infrastructure | |
Li et al. | Privacy protection for medical image management based on blockchain | |
An et al. | Examchain: A privacy-preserving onscreen marking system based on consortium blockchain | |
Guan et al. | Privacy‐Preserving Health Data Collection for Preschool Children | |
CN106357379B (en) | Health data polymerization based on difference privacy | |
Iwasokun et al. | An RSA algorithm for securing financial data on the cloud | |
Saritha | Block chain authentication using elliptic curve digital signature algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170609 Termination date: 20200714 |
|
CF01 | Termination of patent right due to non-payment of annual fee |