A Multi-Key Based Lightweight Additive

2021 International Conference on Artificial Intelligence for Cyber Security Systems and Privacy (AI-CSP) | 978-1-6654-6714-8/21/$31.00 ©2021 IEEE | DOI: 10.1109/AI-CSP52968.2021.9671216

Homomorphic Encryption Scheme

1st Khaled Chait 2nd Abdelkader Laouid
LIAP Laboratory, El Oued University, LIAP Laboratory, El Oued University,
PO Box 789, El Oued 39000, PO Box 789, El Oued 39000,
El Oued, Algeria El Oued, Algeria
[email protected] [email protected]

3rd Lamri Laouamer 4th Mostefa Kara

Department of Management Information Systems, LIAP Laboratory, El Oued University,
Qassim University, 51452, Al Melaydah, PO Box 789, El Oued 39000,
Al Qassim, Kingdom of Saudi Arabia El Oued, Algeria
[email protected] [email protected]

Abstract—Usually, IoT devices have low power. Merging IoT by that party only. Homomorphic encryption schemes include
and cloud computing provides the IoT environment with data three types: partially homomorphic (PHE), fully homomorphic
management flexibility, such as data scalability, enhanced de- level (LFHE), and fully homomorphic (FHE). PHE only allows
vices capacity, economies of scale, etc. But solutions based on
cloud computing can put privacy at risk when operating an one type of operation (only addition or only multiplication)
unreliable provider. The introduction of the homomorph allows with an unlimited number of times that can be performed
a third party to perform operations on the data without having on encrypted messages [9], [10]. LFHE schemes cover both
access to it. In this paper, we propose a partially homomorphic addition and multiplication but a limited number of operations,
lightweight asymmetric encryption technique, which allows to this type supports finite operations on encrypted messages
encrypt decimal digits of the message separately, by multiplying
each by a different key. There are many encryption techniques primarily from the circuit depth point of view. The first FHE
in the literature, most of them suffer from size and slowness schema is presented by Gentry 2009 [13]. This type of homo-
in terms of execution time. In our proposed technique, the morphic encryption allows both addition and multiplication
size of the ciphertext is close to the size of the original one. with an unlimited number of operations.
Moreover, we obtained a time complexity equal to O(n). Thus, The remainder of the paper is as following: Section II
the proposed scheme is suitable for devices and networks with
limited resources, while offering high security. presents the related work. In Section III, we show the proposed
Index Terms—Multi-Key Based technique, Lightweight tech- technique. In Section IV, we presents the robustness of our
nique, Encryption, Homomorphic scheme, Security, Privacy. cryptosystem. Section V shows the time execution results.
Finally, Section VI concludes the paper.
I. I NTRODUCTION
II. R ELATED WORK
Security is the aspect that worries data owners. Homo- In order to guarantee the security of user data, many
morphic encryption is a very promising solution to guarantee techniques are proposed. We will present some encryption
privacy in many sectors, this encryption makes it possible to schemes. For data mining preserving confidentiality in a multi-
effectively secure data. Among its applications, there is the key environment, the authors of [14] proposed an additive
security of data stored in the cloud in order to make operations homomorphic technique with a double decryption mechanism.
on it. Similarly, the cryptocurrency [1], which is a currency In [15], the authors presented a lightweight block cipher
issued from peer to peer without needing a central tier. There using a leveled fully homomorphic cipher scheme based on
are two commonly used types of encryption: symmetric and NTRU. Using a public key k, the one-bit encryption function
asymmetric encryption. In the first category basing on key is represented by a linear equation. To retrieve the plaintext
exchange protocol [2], the same key is used to encrypt and from the ciphertext using the private key f , the decryption
decrypt messages [3], [4], [5], [6], [7], [8]. The second allows function is: m = c × f mod 2. The multiplication in this
the user to distribute the encryption key [9], [10], [11], [12]. scheme introduces a significant increase in noise. The authors
Unfortunately, if the adversary finds the secret key of the of [3] presented a fully homomorphic encryption scheme based
symmetric cipher, he can decrypt all messages encrypted with on a symmetric key. Fully homomorphic targeted encryption
this key. In asymmetric encryption, if the adversary finds the has been proposed by [16]. Based on the modified El-Gamal
private key of one party, they can decrypt the data encrypted cryptosystem, the authors represented the ciphertext in three
978-1-6654-6714-8/21/$31.00 ©2021 IEEE parts. The third part has been added to ensure the addition

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on May 04,2024 at 14:21:23 UTC from IEEE Xplore. Restrictions apply.
 User Untrusted Provider
F
αοβ =?
α F α'
β β' α' ο β’ = λ’
ο untrusted channel
F -1
λ λ’

αοβ =λ

ο : operation, F : encryption, F -1 : decryption

Fig. 1. Homomorphic encryption with an unsecured provider

of two encrypted messages. A fully homomorphic encryption mi by a secret key ki . We consider the primitives M , C, K,
technique was presented by [17]. The encryption function Enc(), Dec() where:
consists in adding the clear message with a random number
• M denotes the plaintext space, with M << p.
multiplied by the secret key, the clear message is included in
• C denotes the ciphertext space.
the calculation of this random number. In order to preserve
• K denotes the public keys (pki ) and private keys (ski )
order, Yagoub et al. used a simple linear expression. A
space.
new fully homomorphic encryption technique based on error
• Enc() denotes the encryption function Encpk0 ,pk1 ,...pki :
learning (LWE) is proposed by [18], in order to avoid complex
P −→ C with pki ∈ K.
matrix operations, Ding et al. modified the re-linearization
• Dec() denotes the decryption function given by:
method developed by Brakerski et al. [19] and improved a
Decsk0 ,sk1 ,...ski : C −→ P with ski ∈ K.
new method called non-matrix key switching. Most techniques
suffer from the size or run time. For example, the Gentry and The proposed technique can be defined as follows :
Halevi FHE [20] scheme require a public key of 70 megabytes.
• KeyGen: Given a private and public keys ski and pki ,
The time required to perform a boot operation varies from 30
respectively, for any random variable ri , (ski , pki ) =
seconds. The technique took over 900 seconds to add two 32-
((ki , p), (ki + ri × p, n))
bit integers. This paper proposes an asymmetric multi-keys
• Enc(m) : The message m is encrypted using the public
PHE scheme (MK-PHE) on integers taking into account the
keys pki , c = m0 × pk0 + m1 × pk1 + . . . mi × pki , with
feasibility in an IoT environment.
pkx < pky if x < y
III. P ROPOSED SCHEME • Dec(c): The P0ciphertext is decrypted using the private keys
sk, m = i=l ( kci ) × 10i , c ← c − c × ki , where ( kci ) is
Against an untrusted cloud, the user tries to get the result of
the quotient of c ÷ ki .
αoβ (Figure 1). For this, we increase the security level of the
technique represented in Equation 1 where ψ is the ciphertext, This scheme is based on the separate processing of each
π is the original message, k denotes the private key, and r is decimal digit that makes up the message.
a random number. Key function We use s keys, where s is the number of
ψ = (π × k + r × p) mod n (1) digits that form the longest plaintext m. The natural case is
the optimal case of keys ki , where m is written in decimal.
where n = p × q, p and q are two large prime numbers. This The ki in the decimal writing are 1, 10, 100, 1000, 10000,
technique is leveled fully homomorphic and it is easily broken. etc. We notice that these ki check the condition: 100 > 9 ×
Our idea is to consider the message m as a concatenation of (1 + 10) = 99, 1000 > 9 × (1 + 10 + 100) = 999, 10000 >
decimal digits, and to treat each digit mi separately (0 <= 9 × (1 + 10 + 100 + 1000) = 9999, etc. We put the following
mi <= 9). Instead of multiplying m by k, we multiply each proposition :

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on May 04,2024 at 14:21:23 UTC from IEEE Xplore. Restrictions apply.
 Decryption In a first step, we calculate the modulo c
j−1
X mod p to eliminate the last part (r × p). Then, we divide
k0 > 10, kj > 9 × ki ∀j > 0 (2)
successively (with decreasing i) c on ki where ki > ki−1 . In
i=0
Pj−1 a last step, we multiply successively (with decreasing i) the
According to proposition 2, we have kj = hj +9× i=0 ki digits obtained by 10i .
where hj are small random numbers, we put hi = 1 ∀i > 1. c
Pj−1
hj = 1 ⇒ kj = 1 + 9 × i=0 ki = (1 + 9 × k0 ) + 9 × k1 + Lemma 1. if ki = mi ⇒ Dec(c) = m
. . . 9 × kj−1 . Proof. It suffices to showthat m0 ×k0 +m1 ×k1 +. . . mj−1 ×
We put (1 + 9 × k0 ) = k1 ⇒ kj = k1 + 9 × k1 + . . . 9 × kj−1 . kj−1 < mj × kj ∀mi ∈ 0, .., 9 ,
We put k1 + 9 × k1 = k2 ⇒ kj = k2 + 9 × k2 + . . . 9 × kj−1 m ×k +m ×k +...mj−1 ×kj−1
i.e., 0 0 1 k1j = 0. In the worst case, mi =
and so on. 
9 ∀i ∈ 0, .., j − 1 and mi = 1 for i = j ⇒ 9 × k0 +
Finally, we will obtain kj = 10 × kj−1 . In general: 9 × k1 + . . . 9 × kj−1 < kj . 9 × (k0 + k1 + . . . kj−1 ) <
Pj−1
k0 > 10, k1 = 1 + 9 × k0 , kj = 10j−1 × k1 (3) kj ⇒ 9 × i=0 ki < kj . According to Proposition 2, kcj =
Let c = m0 × k0 + m1 × k1 + . . . ms−1 × ks−1 where mj . After that, we calculate the subtraction : c ← c − c ×
c
s = size(M ). According to proposition 2, we have c < ks = k j ⇒ c = m0 × k0 + . . . mj−1 × kj−1 ⇒ kj−1 = mj−1 .
10s × k0 . And so on, we will obtain m0 , m1 , . . . mj , then we calculate
Let d be the number of addition operation. For addition, C = m = m0 × 100 + m1 × 101 + . . . mj × 10j .
P d
ci , the coefficient of ki must be less than ki ⇒ k0 > To ensure the recovery of m, we starts the division on k j
Pi=1
d Pd
i=1 mi , we have max(mi = 9) ⇒ k0 > i=1 9 ⇒ k0 > where j = size(c). If i = size(m), we have j > i because
Pd
9 × d. We have i=1 ci < p and c < 10s × k0 ⇒ d × (10s × c > m. Likewise, the decryption is correct for mj < k0 .
k0 ) > d×c ⇒ p > d×(10s ×k0 ), where k0 > 9×d ⇒ p > The decryption function could be defined by the following
9 × d2 × 10s . algorithm :
We can summarize the parameter conditions of the proposed
cryptosystem as follows: Algorithm 2 Decryption algorithm
• Let : m ∈ M, s = size(M ), d is addition depth and h is Require: c, (k0 , k1 , . . . kj ), p
a small random number. Ensure: m = Dec(c)
Pj−1
• Kyes : k0 > 10, kj = hj + 9 × i=0 ki .
′ s−1 function D EC
• For addition : k0 = h + 9 × d, p = h + d × (10 × k1 ). 1:
Encryption For the encryption function, let 2: c ← c mod p
m = m j mj−1 . . . m0 where m0 is the coefficient of 3: l ← size(c)
100 , mi ∈ 0, .., 9 ∀i ∈ 0, .., j and mj ̸= 0. 4: m←0
We multiply each mi by pki where pkx < pky if x < y. 5: for i ← 0 to l do
c
c = m0 × pk0 + m1 × pk1 + . . . mj × pkj ⇒ 6: x ← kl−i−1
c = m0 ×(k0 +r0 ×p)+m1 ×(k1 +r1 ×p)+. . . mi ×(kj +rj ×p) 7: c ← c − x × kl−i−1
8: m ← m + x × 10l−i−1
We obtain : 9: end for
10: return m
c = (m0 × k0 + m1 × k1 + . . . mj × kj + r × p) mod n (4)
11: end function
where r is random, r = m0 × r0 + m1 × r1 + . . . mi × rj
and : m0 × k0 + m1 × k1 + . . . mj × kj < p. The encryption Homomorphic property To ensure the additive property of
function could be defined by the following algorithm: the homomorphic which is : Dec(m1 + m2 ) = Dec(m1 ) +
Dec(m2 ), let x < y, c1 = m01 ×k0 +m11 ×k1 +. . . mx1 ×kx +
Algorithm 1 Encryption algorithm
r1 × p and c2 = m02 × k0 + m12 × k1 + . . . my2 × ky + r2 × p
Require: m, (pk0 , pk1 , . . . pkj ), n that implies Enc(m1 ) + Enc(m2 ) = (m01 + m02 ) × k0 +
Ensure: c = Enc(m) (m11 + m12 ) × k1 + . . . (mx1 + mx2 ) × kx +. . . my2 ) × ky .
Knowing that : (mi1 + mj2 ) < ks ∀i ∈ 0, .., x , j ∈
function E NC   Enc(m1 )+Enc(m2 )
2: l ← size(m) 0, .., y , s ∈ 0, .., y , kv = mv ∀v ∈
c←0 0, .., y ⇒ Dec(c1 + c2 ) = m1 + m2 .
4: for i ← 0 to l do IV. P ERFORMANCE AND SECURITY ANALYSIS
x ← digit(m, [l − i − 1 : l − i]) The proposed asymmetric technique contains two types of
6: c ← (c + x × (pki )) mod n secret keys, the first is the ki , the second is the trapdoor p
end for which is hidden in n with n = p × q (p and q are two large
8: return c prime numbers). If an adversary discovered p, he will be able
end function to extract the all private keys ki from the public keys where

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on May 04,2024 at 14:21:23 UTC from IEEE Xplore. Restrictions apply.
 pki = ki +ri ×p. However, to get the trapdoor p, the adversary In the proposed technique, where we have s secret keys, there
has to solve a factorization problem. The integer factorization are at least (2k0 )s operations that the adversary must do them,
problem is to find p and q, Q more generally, to find primes where k0 is the smallest size of keys and s = size(M ). So,
j
p1 , p2 , . . . , pj such that n = i=1 pi , this particular problem in the BFA with symmetric encryption we have two layers of
cannot be solved in polynomial time until this day. Let λ is the security, one: find the peer (m, c), two: find the sk. In the BFA
security parameter and 2λ < p < 2λ+1 , a security parameter with the asymmetric encryption, there is not an exhaustive
λ asserts that cryptanalysis requires 2λ operations to find the search directly on the secret keys ki because they are hidden
trapdoor p. in r × p (c = F (m, k) + r × p ), so there is only one layer,
Known-plaintext attack: where the adversary has where we have 2p operations to find the private key p.
access to both the plaintext and its encrypted version Computational complexity: Let j be the size of the
(the ciphertext). These can be used to reveal the private message m, the algorithm 1 shows that the complexity
keys. If this known-plaintext m0 is the size of 1 digit, the C(j + 1) = C(j) + χ where χ is a constant, we have
adversary cannot reveal anything because c = m0 × pk0 max(j) = M , this means that the time complexity is a
with c, m0 and pk0 are known. If the adversary tries linear T (M ) = O(M ). Table I shows the performance of
to multiply by the inverse of the public key, then the proposed technique in terms of complexity compared with
c × pk0−1 = m0 × pk0 × pk0−1 ⇒ α = m0 + β × n ⇒ α other techniques under the same parameters (security level,
mod n = (m0 + β × n) mod n ⇒ m0 = m0 where key length).
β is independent of k0 ; so, the adversary cannot reveal
Technique Enc Dec
anything. Assuming now the size of m is j > 1, MK-PHE O(λ) O(λ)
c = m0 × pk0 + m1 × pk1 , if the adversary multiplies [21] O(λ4 ) O(λ4 )
by pk0−1 or pk1−1 , the same thing, he will get c × pk0−1 = [3] O(λ5 ) O(λ4.8 )
m0 +m1 ×pk1 ×pk0−1 ⇒ α+β1 ×n = α+β2 ×n ⇒ α = α [22] O(λ6 ) O(λ5 )
[23] O(λ6 ) O(λ5 )
where β1 and β2 are independent of k0 and k1 ; so, [24] O(λ13 ) O(λ12 )
the adversary cannot reveal anything. Assuming now
TABLE I
that the adversary knows two plaintext m and m′ , if E NCRYPTION AND DECRYPTION COMPLEXITY COMPARISON
size(m) = size(m′ ) = 1, then it is not useful. If
size(m) > 1 and size(m′ ) > 1, then let pk11 = k1 + r1 × p.
 Reduced size: In the symmetric version, the proposed
c = m0 × pk0 + m1 × pk1 simplified formula shown in Equation 3 provides a fairly small
⇒
c′ = m′0 × pk0 + m′1 × pk1 size of the ciphertext.
 ′
m0 × c = m′0 × m0 × pk0 + m′0 × m1 × pk1 Lemma 2. size(Enc(m)) = size(m) + χ where χ is a
⇒
m0 × c′ = m′0 × m0 × pk0 + m0 × m′1 × pk1 constant.
To eliminate one of the public keys by multiplication and  size(m) = j, 0
Proof. Let
substraction, the coefficients of the public key pk1 in the two m = m0 × 10 + m1 × 101 + . . . mj × 10j
we have
equations must be different. In our example, m′0 × m1 must Enc(m) = m0 × k0 + m1 × k1 + . . . mj × kj
be different to m0 × m′1 . The adversary will get another value
According to proposition 2, we can put k1 = 10 × k0 , we
of pk1 (pk1′ ) by substraction, pk1 − pk1′ = (k1 + r1 × p) −
will obtain :
(k1 + r1′ × p) = r1′′ × p, it suffices factorize or to do the
Enc(m) = m0 × 100 × k0 + m1 × 101 × k0 + . . . mj × 10j ×
modular division with n (n = p × q) to obtain the private key
k0 ⇒ Enc(m) = k0 ×(m0 ×100 +m1 ×101 +. . . mj ×10j ) =
p. To reveal secret information, it is not sufficient to have two
k0 × m ⇒ Enc(m) = k0 × m
message m and m′ and its ciphertext. Moreover, m and m′
must verify the following equality m′0 ×m1 ̸= m0 ×m′1 (where We can deduce that χ = size(k0 ) − 1, for example, if
m = m0 m1 and m′ = m′0 m′1 ). For size of known-plaintext size(k0 ) = 11, then size(Enc(m)) = size(m) + 1.
equal to 3 digits, the adversary must know three known-
plaintext to reveal secret information, and so on. Brut force V. I MPLEMENTATION RESULTS
attack (BFA): In many types of attacks, there is one main The implementation of the proposed technique has been
difference that influences the hardness between asymmetric done in Python program language running on a PC with
and symmetric techniques. In the first family, it is easy to get Processor Intel(R) Core(TM) i3-3110M CPU 2.40 GHz, 2
m and c because the adversary uses the public key to create Core(s), 4 Logical Processor(s), 4 Go RAM. Encrypting a
a peer (m, c), then he starts the search (F −1 (c, sk) = m). On message of 16 bits. The results in Table II shows how much the
the other hand, in symmetrical schemes, the adversary must execution time in the proposed technique is reduced compared
first capture the peer (m, c). In the BFA with the symmetric to four techniques.
encryption, the adversary will do an exhaustive search to We conducted several experiments to determine the time
retrieve the private key. In a simple example, comparing by spent in the encryption and decryption operations. Putting
RSA, the adversary must make 2d operations with cd = m. size(m) = 16 bits and size(n) = 360 bits (n = p × q), these

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on May 04,2024 at 14:21:23 UTC from IEEE Xplore. Restrictions apply.
 Technique Enc Dec
[6] N. Aggarwal, C. Gupta, and I. Sharma, “Fully homomorphic symmetric
MK-PHE 0.036 0.041
scheme without bootstrapping,” in Proceedings of 2014 International
[11] 0.07 11.95
Conference on Cloud Computing and Internet of Things. IEEE, 2014,
[14] 11.91 17.67
pp. 14–17.
[25] 47 15
[7] J. Daemen and V. Rijmen, “The block cipher rijndael,” in Interna-
[26] 50 10 tional Conference on Smart Card Research and Advanced Applications.
[27] 255 493 Springer, 1998, pp. 277–284.
[28] 899 785 [8] P. Ekdahl and T. Johansson, “A new version of the stream cipher snow,”
TABLE II in International Workshop on Selected Areas in Cryptography. Springer,
E XECUTION TIME ( MS ) 2002, pp. 47–61.
[9] R. Rivest, A. Shamir, and L. Adleman, “A method for obtaining digital
signatures and public-key cryptosystems,” Communications of the ACM,
vol. 21, pp. 120–126, 1978.
[10] T. Elgamal, “A public key cryptosystem and a signature scheme based on
experiments has shown that the decryption time increases at a discrete logarithms,” IEEE Transactions on Information Theory, vol. 31,
much higher speed than the encryption time. This is due to two no. 4, pp. 469–472, 1985.
[11] M. Kara, A. Laouid, R. Euler, M. A. Yagoub, A. Bounceur, M. Ham-
reasons, the first is that the decryption contains five processes moudeh, and S. Medileh, “A homomorphic digit fragmentation encryp-
(Alg 2), while the encryption contains three processes (Alg 1). tion scheme based on the polynomial reconstruction problem,” in The 4th
The second reason, which is the most important, the size of International Conference on Future Networks and Distributed Systems
(ICFNDS), 2020, pp. 1–6.
the ciphertext increases faster than the size of the plaintext.
[12] M. Kara, A. Laouid, M. A. Yagoub, R. Euler, S. Medileh, M. Ham-
The decryption process is based on processing each digit moudeh, A. Eleyan, and A. Bounceur, “A fully homomorphic encryption
separately, this will create a difference influenced by the size based on magic number fragmentation and el-gamal encryption: Smart
of ciphertext. healthcare use case,” Expert Systems, pp. 1–10, 2021.
[13] C. Gentry, “Fully homomorphic encryption using ideal lattices,” in
Proceedings of the forty-first annual ACM symposium on Theory of
VI. C ONCLUSION computing, 2009, pp. 169–178.
In this paper, we have proposed a novel lightweight tech- [14] H. Pang and B. Wang, “Privacy-preserving association rule mining using
homomorphic encryption in a multikey environment,” IEEE Systems
nique with additive homomorphic encryption property. Our Journal, 2020.
technique is applied to integer numbers, their hardness is [15] Y. Doröz, A. Shahverdi, T. Eisenbarth, and B. Sunar, “Toward practical
based on factorization problems and polynomial reconstruction homomorphic evaluation of block ciphers using prince,” vol. 8438, 03
2014, pp. 208–220.
problems. In which, we obtained a time complexity T (n) [16] Y. Yang, S. Zhang, J. Yang, J. Li, and Z. Li, “Targeted fully ho-
equal to O(n). For each digit mi of the message m to momorphic encryption based on a double decryption algorithm for
be encrypted, we applied the simplest encryption technique, polynomials,” Tsinghua science and technology, vol. 19, no. 5, pp. 478–
485, 2014.
which is c = m × k. We have defined the formulas of the [17] M. A. Yagoub, L. Abdelkader, O. Kazar, A. Bounceur, R. Euler, and
private key k0 and p which determine the depth of the addition. M. AlShaikh, “An adaptive and efficient fully homomorphic encryption
The symmetric encryption of our technique is compared with technique,” 06 2018, pp. 1–6.
the asymmetric encryption, and we demonstrated their order- [18] Y. Ding, X. Li, H. Lü, and X. Li, “A novel fully homomorphic encryption
scheme bsed on lwe,” Wuhan University Journal of Natural Sciences,
preserving. The security analysis of the proposed MK-PHE vol. 21, no. 1, pp. 84–92, 2016.
scheme showed a high level of hardness. The implementation [19] Z. Brakerski and V. Vaikuntanathan, “Efficient fully homomorphic
has further proved their performance particularly in terms of encryption from (standard) lwe,” SIAM Journal on Computing, vol. 43,
no. 2, pp. 831–871, 2014.
data processing rapidity compared to other techniques. The [20] C. Gentry and S. Halevi, “Implementing gentry’s fully-homomorphic
obtained results show that the operations can be performed by encryption scheme,” in Annual international conference on the theory
low computing capacity devices for real applications. and applications of cryptographic techniques. Springer, 2011, pp. 129–
148.
[21] K. Gai, M. Qiu, Y. Li, and X.-Y. Liu, “Advanced fully homomorphic
R EFERENCES encryption scheme over real numbers,” in 2017 IEEE 4th international
[1] M. Kara, A. Laouid, M. AlShaikh, M. Hammoudeh, A. Bounceur, conference on cyber security and cloud computing (CSCloud). IEEE,
R. Euler, A. Amamra, and B. Laouid, “A compute and wait in pow (cw- 2017, pp. 64–69.
pow) consensus algorithm for preserving energy consumption,” Applied [22] H.-M. Yang, Q. Xia, X.-f. Wang, and D.-h. Tang, “A new somewhat
Sciences, vol. 11, no. 15, p. 6750, 2021. homomorphic encryption scheme over integers,” in 2012 International
[2] M. Kara, A. Laouid, M. AlShaikh, A. Bounceur, and M. Hammoudeh, Conference on Computer Distributed Control and Intelligent Environ-
“Secure key exchange against man-in-the-middle attack: Modified mental Monitoring. IEEE, 2012, pp. 61–64.
diffie-hellman protocol,” Jurnal Ilmiah Teknik Elektro Komputer dan [23] Y. G. Ramaiah and G. V. Kumari, “Towards practical homomorphic
Informatika, vol. 7, no. 3, pp. 380–387, 2021. [Online]. Available: encryption with efficient public key generation,” International Journal
http://journal.uad.ac.id/index.php/JITEKI/article/view/22210 on Network Security, vol. 3, no. 4, p. 10, 2012.
[3] V. Biksham and D. Vasumathi, “A lightweight fully homomorphic en- [24] J. H. Cheon, J.-S. Coron, J. Kim, M. S. Lee, T. Lepoint, M. Tibouchi,
cryption scheme for cloud security,” International Journal of Information and A. Yun, “Batch fully homomorphic encryption over the integers,”
and Computer Security, vol. 13, no. 3-4, pp. 357–371, 2020. in Annual International Conference on the Theory and Applications of
[4] M. van Dijk, C. Gentry, S. Halevi, and V. Vaikuntanathan, “Fully Cryptographic Techniques. Springer, 2013, pp. 315–335.
homomorphic encryption over the integers,” in Advances in Cryptology [25] M. Thangavel and P. Varalakshmi, “Enhanced dna and elgamal cryp-
– EUROCRYPT 2010, H. Gilbert, Ed. Berlin, Heidelberg: Springer tosystem for secure data storage and retrieval in cloud,” Cluster Com-
Berlin Heidelberg, 2010, pp. 24–43. puting, vol. 21, no. 2, pp. 1411–1437, 2018.
[5] S. Medileh, A. Laouid, E. Nagoudi, R. Euler, A. Bounceur, M. Ham- [26] J.-S. Coron, A. Mandal, D. Naccache, and M. Tibouchi, “Fully ho-
moudeh, M. AlShaikh, A. Eleyan, and O. A. Khashan, “A flexible momorphic encryption over the integers with shorter public keys,” in
encryption technique for the internet of things environment,” Ad Hoc Advances in Cryptology – CRYPTO 2011, P. Rogaway, Ed. Berlin,
Networks, p. 102240, 2020. Heidelberg: Springer Berlin Heidelberg, 2011, pp. 487–504.

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on May 04,2024 at 14:21:23 UTC from IEEE Xplore. Restrictions apply.
 [27] S. Dasgupta and S. Pal, “Design of a polynomial ring based symmetric
homomorphic encryption scheme,” Perspectives in Science, vol. 8, 07
2016.
[28] D. Boer and S. Kramer, “Secure sum outperforms homomorphic encryp-
tion in (current) collaborative deep learning,” 06 2020.

Authorized licensed use limited to: Birla Inst of Technology and Science Pilani Dubai. Downloaded on May 04,2024 at 14:21:23 UTC from IEEE Xplore. Restrictions apply.