Détails du magazine
Format
Magazine
eISSN
2444-8656
Première parution
01 Jan 2016
Périodicité
2 fois par an
Langues
Anglais
Accès libre

# Research on mobile Awareness service and data privacy Protection based on Linear Equations computing protocol

###### Accepté: 14 May 2022
Détails du magazine
Format
Magazine
eISSN
2444-8656
Première parution
01 Jan 2016
Périodicité
2 fois par an
Langues
Anglais
Analysis of linear equations

Contains variables x1, x2,... The linear equations of Xn are a1x1+a2x2+... +anxn, where b and A1, A2... The coefficients of and an are complex numbers or real numbers, generally known numbers, and the subscript n represents any positive integer.

The solution of a system of linear equations is a set of numbers (s1, s2..., sn), using this group of numbers to replace x1, x2,... When xn, both sides of all equations are equal. The set of all the solutions to the system is called the solution set of the linear system. Assuming that two systems of linear equations have the same class, they can be regarded as equivalent.

Matrix Notation
Definition 1

The main information contained in a system of linear equations can be represented by a matrix array. The following equations are obtained: ${x1−2x2+x3=02x2−8x3=85x1−5x3=10$ \left\{\matrix{{x_1} - 2{x_2} + {x_3} = 0 \hfill \cr 2{x_2} - 8{x_3} = 8 \hfill \cr 5{x_1} - 5{x_3} = 10 \hfill \cr} \right.

Write the coefficients of all variables in an aligned column. The matrix $[1−2102−850−5]$ \left[{\matrix{1 & {- 2} & 1 \cr 0 & 2 & {- 8} \cr 5 & 0 & {- 5} \cr}} \right] is the coefficient matrix of the system, and $[1−21002−8850−510]$ \left[{\matrix{1 & {- 2} & 1 & 0 \cr 0 & 2 & {- 8} & 8 \cr 5 & 0 & {- 5} & {10} \cr}} \right] represents the augmented matrix.

Solving linear equations
Theorem 1

There are four main ways to solve the above equations: First, keep the x1 in the equation and eliminate the x1 in the other equations, as follows: ${x1−2x2+x3=02x2−8x3=810x2−10x3=10 [1−21002−88010−1010]$ \left\{\matrix{{x_1} - 2{x_2} + {x_3} = 0 \hfill \cr 2{x_2} - 8{x_3} = 8 \hfill \cr 10{x_2} - 10{x_3} = 10\, \hfill \cr} \right.\left[{\matrix{1 & {- 2} & 1 & 0 \cr 0 & 2 & {- 8} & 8 \cr 0 & {10} & {- 10} & {10} \cr}} \right]

Second, multiply the equation by half so that the coefficient x2 becomes 1, as follows: ${x1−2x2+x3=0x2−4x3=410x2−10x3=10 [1−21001−44010−1010]$ \left\{\matrix{{x_1} - 2{x_2} + {x_3} = 0 \hfill \cr {x_2} - 4{x_3} = 4 \hfill \cr 10{x_2} - 10{x_3} = 10\, \hfill \cr} \right.\left[{\matrix{1 & {- 2} & 1 & 0 \cr 0 & 1 & {- 4} & 4 \cr 0 & {10} & {- 10} & {10} \cr}} \right]

Thirdly, use the X2 term in the equation to eliminate the coefficients in the equation, as shown below: ${x1−2x2+x3=0x2−4x3=4x3=−1 [1−21001−44001−1]$ \left\{\matrix{{x_1} - 2{x_2} + {x_3} = 0 \hfill \cr {x_2} - 4{x_3} = 4 \hfill \cr {x_3} = - 1\, \hfill \cr} \right.\left[{\matrix{1 & {- 2} & 1 & 0 \cr 0 & 1 & {- 4} & 4 \cr 0 & 0 & 1 & {- 1} \cr}} \right]

Fourth, use x3 to eliminate x 2 and −4x3 in the equation, as shown below: ${x1−2x2=1x2=0x3=−1 [1−2100100001−1]$ \left\{\matrix{{x_1} - 2{x_2} = 1 \hfill \cr {x_2} = 0 \hfill \cr {x_3} = - 1\, \hfill \cr} \right.\left[{\matrix{1 & {- 2} & 1 & 0 \cr 0 & 1 & 0 & 0 \cr 0 & 0 & 1 & {- 1} \cr}} \right]

Fifth, use the X2 term of the equation to eliminate the −2x2 term of the equation, as shown below: ${x1=1x2=0x3=−1 [10010100001−1]$ \left\{\matrix{{x_1} = 1 \hfill \cr {x_2} = 0 \hfill \cr {x_3} = - 1\, \hfill \cr} \right.\left[{\matrix{1 & 0 & 0 & 1 \cr 0 & 1 & 0 & 0 \cr 0 & 0 & 1 & {- 1} \cr}} \right]

Sixth, the only solution to the original system is (1, 0, −1).

Mobile awareness service and data privacy protection analysis
Service Framework

The structure of mobile awareness service model with Internet of Things technology as the core is shown in Figure 1 below, which includes mobile awareness, network interaction, resource scheduling, intelligent processing, awareness services and other contents.

From the perspective of practical application, mobile perception layer mainly uses static perception nodes and mobile perception nodes to obtain or transmit attribute information of entity objects in the real physical world.[1.2] The network interconnection layer mainly uses mobile communication, wireless communication and other technologies to complete the transmission and exchange of perception data and services. The resource scheduling layer mainly uses service-aware middleware or API to obtain the request command transmitted by network interconnection, correctly handles the scheduling and monitoring of resources such as computing nodes with unified resource management strategy, and finally returns the result data to target users.

Privacy Security

Depending on the type of data privacy are shown in table 1 below and content analysis found that mobile service perception model this paper mainly discusses the static, dynamic, derived the three contents, and privacy protection in the scientific control collection, transmission, storage and other operation process, to ensure the user access to target based on the results of perceived service to enjoy, with integrity and controllability for private data[3.4].

Data privacy analysis of mobile awareness service

The data type Content of the privacy
Static data Identity information, location information, user information
Dynamic data Activity trajectory, user request, time, temperature, speed
Derived data Perceiving service content and intelligent information processing results
Model Design

The mobile awareness service model with Internet of Things technology as the core is shown in Figure 2 below, in which the first server receives the service request and transmits it to the end user after effective processing. This system model includes traditional users, perceptive nodes, perceptive services and so on. The objective of privacy protection is divided into two aspects: on the one hand, it refers to the perception data sent by the end user, and on the other hand, it refers to the query request and the control information of the end user.

In order to ensure that mobile awareness services can provide required information for users, Slope One algorithm can be used to select services in practical design. This content was first proposed by Daniel Lemire and Anna Maclachlan. [5.6]It is a service selection algorithm with user evaluation as the core. The practical application is very simple and easy to expand. For example, if there are two service items A and B, and three users X, Y and Z give scores respectively, but user Z does not share the scores of service item A, the average deviation of service items can be calculated by using the scores of the other two users, and finally the score of user Z for service item A can be calculated by combining the deviation. The actual operation process is shown in the figure 3 below:

First, calculate the average deviation of A and B, look for users who evaluate them at the same time, analyze the difference of all users' scores on service items, and then get the average deviation; Second, calculate the score of user Z on service item A. Add the score of user Z on service item B and the average deviation between service item A and B to determine the specific score of user Z on service item A.

Proposition 2

Through extended analysis of this algorithm, for two service items A and B, users' scores will be presented in the form of vectors, as shown below: $A={U1a,U2a,…,Una} and B={U1b,U2b,…,Unb}$ A = \left\{{{U_{1a}},{U_{2a}}, \ldots,{U_{na}}} \right\}\,\,\,\,\,{and}\,\,\,\,\,\,\,\,\,\,B = \left\{{{U_{1b}},{U_{2b}}, \ldots,{U_{nb}}} \right\}

In the above formula, U1a represents user U1 to score service item A. Combined with the above calculation process, the average deviation of dev is determined first, and the specific formula is as follows: $dev〈A,B〉=∑U∈(A∩B)Ub−Uanum(A∩B)$ {{dev}_{\left\langle {A,B} \right\rangle}} = \sum\limits_{U \in \left({A \cap B} \right)} {{{{U_b} - {U_a}} \over {{num}\left({A \cap B} \right)}}}

Lemma 3

In the above formula, Ua and Ub represent user U's ratings of service items A and B, and num(AB) represents the number of users who evaluate service items at the same time. Note that num(AB) does not equal zero.

The average deviation dev can be used to calculate the user's score of service item B, the specific formula is as follows: $P(Ub)=Ua+dev〈A.B〉$ P\left({{U_b}} \right) = {U_a} + {{dev}_{\left\langle {A.B} \right\rangle}}

In the above formula, P (Ub) represents user U's score of service item B.

The improvement and discussion of the Slope One algorithm mentioned above can be divided into the following steps: First, let all service projects complete effective grouping, and group the service projects with similar or similar contents into One group; Second, ensure that each group chooses a benchmark service project, and follow the standards for users to score it comprehensively and accurately; Third, calculate the average deviation accurately using the benchmark service items; Fourthly, calculate the user's service item score with average deviation.[6.7]

Corollary 4

According to the calculation and analysis, the formula for calculating the average deviation of the improved Slope One algorithm is shown as follows: $dev〈Ag,B〉=∑U∈(Ag∩B)Ub−Uagnum(Ag∩B)$ {{dev}_{\left\langle {{A_g},B} \right\rangle}} = \sum\limits_{U \in \left({{A_g} \cap B} \right)} {{{{U_b} - {U_{{a_g}}}} \over {{num}\left({{A_g} \cap B} \right)}}}

If you have service items A, B, C, D, and E, select three users to score, then divide the service items into multiple groups. Suppose the grouping results are A and B, C, D and E, and the final results are A and C strictly according to the benchmark selection. The results of the first set of A and B are no different from those of the improved application algorithm. However, for the second group, it needs to be calculated according to the standard of C. Assuming that the score of user U3 on service item D is studied, the calculation formula of the actual average deviation of the score is as follows: $dev=(U1d−U1c)+(U2d−U2c)2$ {{dev}_{< {C_g},D >}} = {{\left({{U_{1d}} - {U_{1c}}} \right) + \left({{U_{2d}} - {U_{2c}}} \right)} \over 2}

The scoring formula of user U3 for service item D is as follows: $U3d=U3c+dev$ {U_{3d}} = {U_{3c}} + de{v_{< {C_g},D >}}

The calculation formula of average deviation of user U3's score for service item E is as follows: $dev=(U1c−U1c)+(U2c−U2c)2$ {{dev}_{< {C_g},D >}} = {{\left({{U_{1c}} - {U_{1c}}} \right) + \left({{U_{2c}} - {U_{2c}}} \right)} \over 2}

Conjecture 5

The scoring formula of user U3 for service item E is as follows: $U3c=U3c+dev$ {U_{3c}} = {U_{3c}} + de{v_{< {C_g},E >}}

Compared with the application algorithms before and after the improvement, it is found that the operation steps of the improved algorithm are simpler, and only need to consider the benchmark and computing service items.

Data privacy algorithm
Example 6

First, the relevant text encryption algorithm. The most common encryption algorithm is symmetric encryption algorithm. Based on the analysis of the algorithm flow chart shown in Figure 4 below, it can be seen that this algorithm assumes that the plaintext is P, the ciphertext is C, the key is K, and the encryption function is D. Then the actual encryption process is as follows: $C=E(K,P)$ {\rm{C}} = {\rm{E}}\left({{\rm{K}},{\rm{P}}} \right)

The decryption process is as follows: $P=D(K,G)$ {\rm{P}} = {\rm{D}}\left({{\rm{K}},{\rm{G}}} \right)

Second, differential privacy technology. This content is put forward by Dwork and others in the study of practice can be deduced to calculate model number of privacy protection, do not need other additional information can be privacy protection, and all of the content to the data collection is mainly used for sensitive data analysis and mining, and can be combined with machine learning to provide mobile service perception of data privacy protection.[8.9]

Suppose you have two data sets D and D′ with the same properties, then they can be called adjacent data sets. All function mappings in data set D are treated as queries on the data set, one set of queries being F = [f1, f2,...]. Assuming that a random algorithm M appears, and PM represents the set of all outputs of M, then all adjacent data sets conform to the following formula: $Pr[M(D)∈SM]≤eε×Pr[M(D′)∈SM]+δ$ {P_r}\left[{M\left(D \right) \in {S_M}} \right] \le {e^\varepsilon} \times {P_r}\left[{M\left({{D^{'}}} \right) \in {S_M}} \right] + \delta

Under this condition, algorithm M can provide (ε, δ) differential privacy protection, SM represents any subset of PM, parameter ε represents the privacy protection budget, parameter δ represents the tolerance of budget failure.

Difference of privacy protection plus noise mechanism is divided into two kinds: a mechanism for Laplace, on the basis of given data set D, assuming function accord with f: DRd this condition, the actual sensitivity is Δf, so random algorithm can provide ε difference privacy protection, Y ~ Lapf/ε) represents the random noise, and obey the distribution of the scale parameter is Δf / ε conditions; The other is an exponential mechanism. It is assumed that the random algorithm M inputs the data set D, and the output target is rR, q(D, r), which is the availability function. Δq represents the sensitivity of the availability function. Algorithm M can provide differential privacy protection if algorithm M selects and outputs R from R with a probability proportional to exp (εq(D, r) / 2Δq)

Privacy at the same time, the difference algorithm and machine learning, according to analysis in three ways: the first refers to the objective function disturbance mechanism, assuming that the original loss function of machine learning is L (f, D), f represents the model parameters, D on behalf of the training data set, combined with the noise mechanism for noise vector b, then the corresponding loss function is as follows: $Ldf(f,D)=L(f,D)+1nbTf$ {L_{df}}\left({f,D} \right) = L\left({f,D} \right) + {1 \over n}{b^T}f

It is assumed that GT (xi) represents the model in the t iteration and the gradient data on data XI gt(xi) ← ∇θt L (θt,xi), the disturbance intermediate parameter is on a batch data of size B, the noise vector is B, and the gradient value of adding noise is shown as follows: $g^t←1B(∑igt(xi)+b)$ {\hat g_t} \leftarrow {1 \over B}\left({\sum\nolimits_i {{g_t}\left({{x_i}} \right) + b}} \right)

Finally, it refers to the disturbance output mechanism. After the model parameter F is obtained from the noise-free data set by combining the loss function L (f, D), the mechanism of adding noise disturbance is directly used in the model parameters. If the noise vector is B, the parameters after adding noise are shown in the following formula: $fdf=f+b$ {f_{df}} = f + b

Direct use of FAF instead of F can carry out differential privacy protection for data.

Third, network representation learning algorithm. Combined with the analysis of the algorithm model structure shown in Figure 5 below, it can be seen that to protect data privacy security in the mobile awareness service model, cloud service mode with multi-party training should be used to prevent attackers with large background knowledge from conducting differential attacks on shared resources, resulting in data leakage.

After constructing the original structure diagram of the network, the random walk strategy with the original structure diagram as the core should be used to construct the corresponding node relation sequence of all entity nodes. During the random walk, the node at step I +1 should be selected according to the following calculation formula: $Pr(vi|vi−1;TG)=1|TG(vi−1)|•1|{u|(vi−1,u)∈E,ϕ(vi)=ϕ(u)|}|$ \Pr \left({{v_i}|{v_{i - 1}};{T_G}} \right) = {1 \over {\left| {{T_G}\left({{v_{i - 1}}} \right)} \right|}} \bullet {1 \over {\left| {\left\{{u\left| {\left({{v_{i - 1}},u} \right) \in E,\phi \left({{v_i}} \right) = \phi \left(u \right)} \right|} \right\}} \right|}}

In the above formula, |TG(vi−1)| represents the number of node types connected to node VI-1 in the attribute network. If the vi-1 node is an entity node, then |TG(vi−1)| represents the number of types of attribute nodes; If the node vi-1 represents the attribute node, then |TG(vi−1)| equals 1 {u|(vi−1,u)∈ E,ϕ(vi) = ϕ(u)} represents the total number of class-like nodes adjacent to node VI-1. Meanwhile, the loss function as shown below is designed to make the self-coding model converge: $La=∑i=1n‖H^(vi)−xi‖$ {L_a} = \sum\nolimits_{i = 1}^n {\left\| {\hat H\left({{v_i}} \right) - {x_i}} \right\|}

In the above formula, |N (i)| represents the number of entity nodes having proximity relationship with entity node VI in the network, and PIj represents whether there is proximity relationship between entity node VI and entity node Vj. The specific formula is as follows: $H(vi)=1|N(i)|∑j∈N(i)pijxj$ H\left({{v_i}} \right) = {1 \over {\left| {N\left(i \right)} \right|}}\sum\nolimits_{j \in N(i)} {{p_{ij}}{x_j}}

The mapping function between layers is as follows: $hi(1)=σ(Wxi(1)+b(1))hi(k)=σ(W(k)hik−1+b(k)),k=2,…,k$ \eqalign{& h_i^{\left(1 \right)} = \sigma \left({W_{xi}^{\left(1 \right)} + {b^{\left(1 \right)}}} \right) \cr & h_i^{\left(k \right)} = \sigma \left({{W^{\left(k \right)}}h_i^{k - 1} + {b^{\left(k \right)}}} \right),k = 2, \ldots,k \cr}

In the above formula, W represents the weight matrix between neural network layers, B represents the offset vector during the calculation of neural network data, σ represents the activation function, K represents the number of neural network layers in the global depth self-coding model, and K represents the KTH layer of the network corresponding to the mapping function.[10]

Performance analysis

Combined with the above analysis of the mobile awareness service model with linear equations computing protocol as the core, it is found that its data privacy protection is stronger during time operation. Even if the attacker knows that all data content has been cleared in the service request data set, the privacy security of relevant data records can still be guaranteed. At the same time, this paper studies design of a variety of privacy data privacy protection model algorithm, such as difference of privacy protection, said learning algorithm, etc., can not only help to improve efficiency of mobile service perception to enjoy, you can also directly in the data, at the request of the sensing layer, mobile sensing data privacy of Internet will not be restricted by the end user equipment, It has strong application performance during practical operation. Therefore, in future technological exploration, researchers from various countries should study the application effects of different application algorithms in data privacy protection of mobile awareness services from multiple perspectives, so as to grasp more technical achievements.

Conclusion

To sum up, in the development of modern technology and social economy innovation steadily, iot mobile perception service framework and data privacy protection put forward higher requirements, so all research scholars in guarantee the quality of the service at the same time, more attention should be paid to improve the service response speed, deep to explore how to use diverse applications algorithm to ensure the safety of mobile service perception of data privacy. According to the empirical analysis of this paper, it is found that both differential privacy protection and network representation learning algorithm can meet the operation requirements of the mobile awareness service model of the Internet of Things, in which information control, query request, perceptual data and other operational processes can effectively control the probability of data loss and other problems. Therefore, on the basis of comprehensively mastering the structure of mobile awareness service model, researchers in various countries should carry out targeted technical exploration and empirical analysis according to the emerging problems in data privacy protection, and only in this way can they master more valuable contents.

#### Data privacy analysis of mobile awareness service

The data type Content of the privacy
Static data Identity information, location information, user information
Dynamic data Activity trajectory, user request, time, temperature, speed
Derived data Perceiving service content and intelligent information processing results

Yong ge. Design of Several Secure Multi-Party Numerical Computation Basic Protocols [D]. Anhui University, 2017. Yongge Design of Several Secure Multi-Party Numerical Computation Basic Protocols [D] Anhui University 2017 Search in Google Scholar

Li wang, Qingwen wang. Computational protocol for general linear equations protecting private information [J]. Journal of applied mathematics and computational mathematics, 2014,28(03):300–307. Liwang Qingwenwang Computational protocol for general linear equations protecting private information [J] Journal of applied mathematics and computational mathematics 2014 28 03 300 307 Search in Google Scholar

Li chen. Research on the application of secure multi-party computing in Internet of Things [D]. Fuzhou University, 2014. Lichen Research on the application of secure multi-party computing in Internet of Things [D] Fuzhou University 2014 Search in Google Scholar

Li chen, Baigang lin. Secure Multi-Party Computing Protocol Based on Distributed Linear Equations [J]. Information Network Security, 2013(09):2–5. Lichen Baiganglin Secure Multi-Party Computing Protocol Based on Distributed Linear Equations [J] Information Network Security 2013 09 2 5 Search in Google Scholar

Li wang. Design and Analysis of several matrix cryptographic protocols [D]. Shanghai University, 2013. (in Chinese) Liwang Design and Analysis of several matrix cryptographic protocols [D] Shanghai University 2013 (in Chinese) Search in Google Scholar

Yang yang. Simple secure multi-party computing protocol [J]. Silicon Valley, 2011(10):161+163. Yangyang Simple secure multi-party computing protocol [J] Silicon Valley 2011 10 161 163 Search in Google Scholar

Hengyue jia, Huanping liu. Computer Engineering and Applications, 2008(33):112–114. Hengyuejia Huanpingliu Computer Engineering and Applications 2008 33 112 114 Search in Google Scholar

Zhifang zhang. Multiplicative monotone Stretching scheme [J]. Journal of Graduate School of Chinese Academy of Sciences, 2006(06):827–832. Zhifangzhang Multiplicative monotone Stretching scheme [J] Journal of Graduate School of Chinese Academy of Sciences 2006 06 827 832 Search in Google Scholar

Wenjun luo, Xiang li. Multi-secure Matrix Product Protocol and Its Application [J]. Chinese Journal of Computers, 2005(07):1230–1235. Wenjunluo Xiangli Multi-secure Matrix Product Protocol and Its Application [J] Chinese Journal of Computers 2005 07 1230 1235 Search in Google Scholar

Hua zhang. Research on Secure Multi-party Computing and its Application [D]. Xidian University, 2005. Huazhang Research on Secure Multi-party Computing and its Application [D] Xidian University 2005 Search in Google Scholar

Articles recommandés par Trend MD