DATA SECURITY & PRIVACY IN CLOUD COMPUTING:

C ONCEPTS AND E MERGING T RENDS

A P REPRINT

Rishabh Gupta∗ Deepika Saxena†

Department of Computer Applications Department of Computer Applications
arXiv:2108.09508v1 [cs.CR] 21 Aug 2021

National Institute of Technology National Institute of Technology

Kurukshetra, India Kurukshetra, India
[email protected] [email protected]

Ashutosh Kumar Singh

Department of Computer Applications
National Institute of Technology
Kurukshetra, India
[email protected]

August 24, 2021

A BSTRACT
Millions of users across the world leverages data processing and sharing benefits from cloud en-
vironment. Data security and privacy are inevitable requirement of cloud environment. Massive
usage and sharing of data among users opens door to security loopholes. This paper envisages a
disuscussion of cloud environment, its utilities, challenges, and emerging research trends confined to
secure processing and sharing of data.

Keywords Cloud computing · Security · User-Privacy · Data Protection

1 Introduction
In 2006, San Jose proposed cloud computing’s concepts in Search Engine Strategies (SES), and the National Institute
of Standards and Technology (NIST) gave a formal definition of cloud. Almost every type of organization aspires to
use cloud computing technology because of its features such as storage, computation, flexibility, scalability, etc [1].
At present, data is developing at a faster pace. Indeed, even excessive information is being created every second, and
likewise, the number of Internet-associated gadgets is increasing gradually [2], [3]. As the users may not have sufficient
space to store their information, they need to keep their data on the cloud server and access the services facilitated by
the cloud [4], [5]. Cloud storage gives an effect such that there is a vast space to store information and retrieve the data
easily. Users have to pay for each utilized cloud service [6], [7]. To make cloud computing possible and available to
end-users, some services and models function behind the scenes. Fig. 1 shows two kinds of cloud computing models:
deployment models and service models:
The deployment models pertain to where the cloud infrastructure is located and managed. There are three commonly
used cloud deployment models: public, private, and hybrid cloud. Another type of model is the community cloud which
is commonly less used. A public cloud is a collection of computing resources that third-party organizations provide. It
supports all users who want a computer resource, including subscription-based hardware (OS, CPU, memory, storage)
∗
This work is supported by the University Grant Commission, New Delhi, India, under the scheme of National Eligibility
Test-Junior Research Fellowship (NET-JRF) with reference id-3515/(NET-NOV 2017)
†
The authors would like to thank National Institute of Technology Kurukshetra, India for financially supporting this research
work.
 Data Security & Privacy in Cloud Computing A P REPRINT

Cloud Char- Deployment

Service Models
actertics Models

On-demand SaaS Public

self service
Private
Broad network
access
Hybrid
Resource PaaS
pooling Community
Rapid
elasticity or
expansion
IaaS
...

Measured
service

Figure 1: Cloud Computing (NIST)

or software (application server, database) to be used. A private cloud is typically one organization’s infrastructure. The
organization or a service provider can manage such infrastructure to help different customer groups. A hybrid cloud
is a combination of private as well as public cloud computing resources. An additional model is a community cloud
that shares in several organizations computing resources, and it is possible to manage either through organizational IT
resources or third-party vendors [8], [9], [10]. Cloud service models portray how customers can access the cloud [11].
A combination of IaaS (infrastructure as a service), PaaS (platform as a service), and SaaS (software as a service) are the
most basic service models. These service models can synergize; for instance, PaaS relies on IaaS because application
systems require physical facilities [12].
The model of IaaS offers customers parts of the infrastructure, including virtual machines, storage, networks, firewalls,
and load balancers components [13]. Customers have direct access to the lowest-level software like the virtual machine
operating system, firewall, and load balancer dashboard in the stack with IaaS. Amazon Web Services is one of the
largest IaaS providers. The PaaS model provides the customer with a pre-constructed application platform. Customers
do not need to spend time building their applications ’ underlying infrastructure. PaaS solutions usually offer an API
that contains several features for the management and creation of programmatic platforms. Google AppEngine is one of
the examples of a PaaS provider. SaaS offers prepared online software solutions. The SaaS software provider fully
controls the application software. Online mail, project management systems and social media platforms are examples
of SaaS applications [14].
In 2012, Cisco Bonomi coined the term “fog computing”. The idea of fog computing is to bring the cloud closer to the
devices. It is a distributed paradigm, and an extension of cloud computing. Fog computing’s primary goal is to address
the issues that cloud computing encounters during data processing. Fog is not a replacement for cloud computing; rather,
these two technology complements each other. Routers, switches, intelligent devices, access points, and gateways are
the essential parts of the fog layer. These devices can store data, compute, route, and forward packets. The fog layer
is the intermediate layer between user and cloud, as demonstrated in Fig. 2. This layer has rights of semi-permanent
capacity that permit the impermanent information and data computation capability. The use of this layer has many
advantages as low latency, higher real-time, broader geographical distributions. Moreover, it is observed that due to
neighborhood information, the routing and storage costs in the fog environment are lesser than in the cloud environment
[15]. All of these benefits make fog computing more suitable for applications that are sensitive to delay.

2 Motivation
The paradigm of cloud computing has enabled various users to share resources [16], [17], [18]. This notion has achieved
broad popularity over the few years by constantly increasing the number of customers and supporting infrastructure
development [19]. For instance, Organizations like Cisco expect to have more than 24 billion gadgets connecting with

2
 Data Security & Privacy in Cloud Computing A P REPRINT

Figure 2: Fog Computing

the Internet by 2020. Further, Morgan Stanley has also predicted that almost 75 billion Internet devices will be in use by
2021. Cloud computing provides security to the customers, preventing their sensitive information and personal data
from unauthorized stakeholders and multiple parties. Furthermore, data providers can monitor their outsourced data
privacy at any moment. Organizations do not need to worry about data security because this facility is accessible as a
service nowadays.
Cloud computing is a growing technology that provides massive information without upfront investment to organizations
with a novel business model [20] [21], [22]. However, most organizations are still reluctant to explore their business
across the cloud because of security. When the data is stored in the cloud server, the consequences are data ownership
and management isolation [23]. Therefore, the Cloud service provider (CSP) can search and access data freely on the
cloud server without taking the permission of the user. Meanwhile, the cloud server might be assaulted by an aggressor
to get the user’s information. There are many security issues in the cloud environment, such as man-in-the-middle
attacks, data leaks, etc [24], [25], [26]. All the above problems are exceptionally perils for user’s information. Losses
of the information of users is hampered, and there may occur data leakage problems too. If the user directly uploads
data on the cloud, there may be issues like high bandwidth requirement, high latency, and a large volume of data. These
are the biggest obstacles limiting cloud development. Therefore, it is necessary to work on it and find the right solution.
Fig. 3 exhibits consumer fraud and identity theft complaints lodged during the period ranging from 2016 to 2020. It is
observed that 4.8 million identity theft and fraud complaints were received in the year 2020. Out of these, 1.4 million
complaints were identity theft, 25 percent of cases also reported money loss amounting, and remaining 2.2 million were
fraud complaints. The number of complaints in 2020 is increased up to 46 percent over the previous year.

To address the problems mentioned above, the state-of-art encryption schemes [27], [28] were used to protect the users’
data. These schemes enable the users to encrypt their data and store it on the cloud servers. Afterward, cloud servers can
carry out computations on it. But, Cloud servers are presently unable to provide innovative perspectives to users while
fully maintaining their privacy. Therefore, efforts must be made to answer the question: “How can data consumers
profit from cloud computing without compromising data privacy?”.

3 Privacy-Preserving Based on Cryptography Mechanism

Yuan and Yu [29] proposed a multiparty Back-Propagation Neural (BPN) network-based approach that is accurate,
efficient, and secure for collaborative learning over arbitrarily partitioned data. To conduct operations over ciphertexts,
they used a doubly homomorphic encryption technique. But they concentrated on enhancing data processing rather
than the algorithm’s efficiency. Zhang et al. [30] proposed a privacy-preserving deep computation model based on
homomorphic encryption. They used the offloading of the expensive operations to enhance the learning features of
the cloud. The exponential process required by the Sigmoid function was utilized using the Taylor theorem. However,

3
 Data Security & Privacy in Cloud Computing A P REPRINT

Figure 3: Identity Theft And Fraud Reports, 2016-2020

Source: Federal Trade Commission, Consumer Sentinel Network

the model only includes addition and multiplication operations. Yonetani et al. [31] introduced a privacy-preserving
mechanism based on a double-permitted homomorphic encryption (DPHE) scheme, which effectively learns visual
classifiers across distributed private data. This scheme provided multiparty protected scalar products while minimizing
the computational cost for high-dimensional classifiers. Nevertheless, either addition or multiplication operation can
only support at a time. A deep learning system based on additively homomorphic encryption was presented in [32]. The
proposed system protects gradients from the curious server. Asynchronous stochastic gradient descent (ASGD) trained
overall participants’ combined datasets obtained the same accuracy as the related deep learning system. However, the
modified parameters are decrypted by the owner’s secret key; thus, their model does not ensure parameter privacy. A
basic scheme based on multi-key fully homomorphic encryption (MK-FHE) mechanism was introduced in [33]. The
authors devised an advanced model for learning encrypted data in the cloud that uses the double decryption mechanism
and fully homomorphic encryption (FHE) mechanism. But this scheme has a high cost in terms of computation and
communication. To perform the deep neural network algorithms over encrypted data, a framework named CryptoDL
was proposed by Hesamifard et al. [34]. To address the existing limitations of homomorphic encryption schemes,
they designed neural networking techniques. However, the proposed algorithm protects the owner’s data by using
keys that is not practical. A privacy-preserving outsourced classification in cloud computing (POCC) framework was
presented in [35], which effectively enable an arbitrary number of multiplication and addition operations on ciphertexts.
The data and query were protected by providing a proxy fully homomorphic encryption based on Gentry’s scheme.
Nevertheless, the cost of calculation and communication was increased in the proposed framework. Ma et al. [36]
proposed a privacy-preserving deep learning model, namely PDLM, to train the model over the data encrypted by the
owners’ keys. A privacy-preserving calculation toolset based on stochastic gradient descent (SGD) was utilized to
accomplish the training task in a privacy-preserving way. Although the model minimized storage overhead, it has a
high computation cost and lower classification accuracy. A privacy-preserving outsourced classification scheme is
presented in [37], which provides the classification services over encrypted data for users. They also designed two
concrete secure classification protocols for the Naive Bayes classifier and the hyperplane decision-based classification,
respectively. But during the launch of a classification query, user interactions are often involved in this scheme. Gao
et al. [38] proposed a privacy-preserving Naive Bayes classifier scheme that prevents information leakage under the
substitution-then comparison (STC) attack. A double-blinding technique was adopted to protect the Naive Baye’s
privacy. Both the communication and processing overhead were decreased, but unable to discover the truth while
maintaining privacy. Phong and Phuong [39] proposed two systems, namely the Server-aided Network Topology (SNT)
system and the Fully-connected Network Topology (FNT) system based on the connection with SNT and FNT server
to protect the SGD privacy. The SNT and FNT systems achieved an accuracy corresponding to SGD using weight
parameters instead of gradient parameters. These systems are both effective and efficient in terms of computing and
communication. Table 1 shows the summary of privacy-preserving of data based on the cryptography mechanism.

4
 Data Security & Privacy in Cloud Computing A P REPRINT

Table 1: Summary of Privacy-Preserving based on Cryptography

Literature Reference Approach Pros Cons
Wang et al. [15] Hash-Solomon code High degree High
algorithm privacy protection Computation cost
Yuan et al. [29] BGV fully and Doubly Encrypt data efficiently Low efficiency
homomorphic encryption secure scalar product
Yonetani et al. [31] Doubly homomorphic Supported multi-party Both addition and
encryption secure scalar product multiplication
Li et al. [33] Multi-key fully Preserve privacy of Low efficiency
homomorphic encryption sensitive data
Ma et al. [36] Distributed Two Trapdoors More efficiency Less accuracy
Public-Key Cryptosystem
Shokri and Shmatikov [40] Optimization algorithms Protect training data High complexity
Wang et al. [41] Homomorphic encryption Encrypted data by randomly Either addition or
splitting numerical multiplication
Chan et al. [42] Homomorphic and Partitioned Data Two party
ElGamal scheme vertically
Bansal et al. [43] Homomorphic and Partitioned Data Two party
ElGamal scheme arbitrarily
Samet et al. [44] BPNN and Extreme Partitioned Data horizontally, High Communication
learning machine scheme vertically with multi parties cost
Cao et al. [45] KNN,TF-IDF and Solve the problem of muti- Does not explore
Dynamic Data Operation Keyword ranked search checking integrity
Guo et al. [46] TF-IDF model Update the outsourced data Does not improve
at less cost computational efficiency
Fu et al. [47] Stemming algorithm Outsourced Data With Does not reflect
Accuracy Improvement at less cost the keyword weight
Fu et al. [48] Keyword based Use two cloud server High
Search Scheme for store and compute Compleaxity
Huang et al. [49] Public key authenticated Handle to resist Single
encryption inside keyword keyword
Qi et al. [50] Locality-Sensitive Handle the service Low
Hashing recommendation efficiency

4 Privacy-Preserving Based on Perturbation Mechanism

Dwork et al. [51] first proposed differential privacy and obtained complete background knowledge under the attacker’s
hypothesis. To protect data privacy, the randomly generated noise is disturbed according to a specially selected
distribution. Fletcher and Islam [52] proposed a differential privacy decision-making random forest algorithm to reduce
the query times and sensitivity. This scheme also minimizes the amount of noise that must be appended to protect the
privacy and improve data availability. However, there is no consideration for the distributed scenario where multiple
data owners conduct collaborative data mining. To perform privacy-preserving machine learning over cloud data from
different data providers, Li et al. [53] proposed a scheme that protects the data sets of various providers and the cloud.
They used the public-key encryption with a double decryption algorithm (DD-PKE) to encrypt the data sets of the
different providers with different public keys and -differential privacy to add statistical noises into data to protect the
privacy. Their scheme improved computational efficiency and data analysis accuracy. But such fully homomorphic
(FHE) encryption schemes are commonly low efficiency. A privacy-preserving Naive Bayes learning scheme with
various data sources is presented in [54]. The proposed scheme enabled a trainer to train a Naive Bayes classifier over
the dataset provided jointly by different data owners without the help of a trusted curator. However, collaboration is
permitted, or adversaries can forge and modify the data in this scheme. A distributed agent-based privacy-preserving
framework, namely DADP, was proposed by Wang et al. [55]. The proposed framework collects real-time spatial
statistics data and publishes it with an untrusted server. To achieve global w-event -differential privacy in a distributed
manner, they utilized a distributed budget allocation mechanism and an agent-based dynamic grouping mechanism.
The noise is added to crowd-sourced data using the Laplace technique in DAPM. It started a batch of reliable proxies
(Agents) and anonymous connection technology to safeguard users’ privacy under an untrusted server. Therefore, it was
regarded as a semi-centralized setting and resulted in a more complex system. An efficient privacy-preserving scheme
based on machine learning was proposed by Hassan et al. [56]. Authors adopted a partially homomorphic encryption
technique to encrypt data, and noised is added by applying a differential privacy mechanism. It allows all parties

5
 Data Security & Privacy in Cloud Computing A P REPRINT

to publicly check the ciphertext’s correctness via a low-cost unidirectional proxy re-encryption (UPRE) mechanism.
However, the proposed system shared fewer data. A private decision tree algorithm based on the noisy maximal vote
was introduced in [57]. To strike a balance between accurate counts and noise, an effective privacy budget allocation
approach was utilized. The main aim of constructing an ensemble model is to increase the accuracy and stability
by using differential privacy. The proposed algorithm performs the privacy analysis on each individual tree rather
than the ensemble as a whole. Gupta et al. [58] proposed a machine learning and probabilistic analysis-based model,
namely MLPAM. The authors used encryption, machine learning, and probabilistic approaches to share the multiple
participants’ data and minimize the risk affiliated with the leakage for prevention and detection. However, MLPAM is
not able to give protection to the classifier. To preserve the privacy of the data as well as query processing, Sharma et al.
[59] proposed a Differential Privacy Fuzzy Convolution Neural Network framework, namely DP-FCNN. The Laplace
mechanism was utilized to inject the noise and encrypt the data by applying the lightweight Piccolo algorithm. The key
properties were extracted using the BLAKE2s technique. But, DP-FCNN enhanced the computational overhead. Table
2 summarises the privacy-preserving data using the Perturbation Mechanism.

Table 2: Summary of Privacy-Preserving based on Perturbation

Literature Reference Approach Pros Cons
Fletcher and Islam [52] -Differential Privacy Reduce query time Less accuracy
Li et al. [53] Homomorphic encryption More efficiency High Computation
Differential Privacy cost
Li et al. [54] -Differential Privacy Preserve privacy of data Forge data
Wang et al. [55] w event Data protection Complex system
-Differential Privacy
Hassan et al. [56] Homomorphic encryption Low cost Limited data sharing
-Differential Privacy
Liu et al. [57] Laplace mechanism Balance between Individual privacy
accuracy and noise
Gupta et al. [58] Gaussian mechanism High accuracy No classifier
protection
Sharma et al. [59] Laplace mechanism Data protection Computation overhead
protection

5 Research Gaps
On the basis of the literature review, the following research gaps are identified.
1. Less protection, and privacy of the outsourced data.
2. The efficiency of methods to protect privacy must be increased.
3. Multiple user-based protection techniques are required.
4. Computational and communication costs during the data transfer must be reduced.
5. Minimization of the threats of data leakage during transmission.

6 Research Objectives
To fill the identified study gaps, the goal described below is developed.
1. To optimize the computation and communication costs among different entities.
2. To reduce data leakage by using advanced encryption techniques.
3. To solve the security issue of cloud computing.
4. To formulate the approach to reduce the latency while preserving data privacy during transmission.

7 Justification for the objectives

In an age of quickly growing information generation, it requires a computing infrastructure that can store and process
enormous amounts of information. Developing these new cloud-based methods defines current safety and privacy

6
 Data Security & Privacy in Cloud Computing A P REPRINT

demands and addresses consumer issues facing quality, effectiveness, and the best use of consumer needs. We provide
complete awareness of secure cloud and efficient use of resources to validate our methodologies. Because security helps
prevent data leakage and the disposal of data. With the following benefit, the advances to be carried out in research
work will open an era for enhanced cloud facilities:

1. More quality of service (QoS)

2. Better resource utilization
3. Improves confidence in cryptography services
4. Low the cost of computation of the process
5. Less service level agreement (SLA) violation
6. Better use of the cloud for securely sharing of data among the organizations.
7. Better efficiency
8. Saving the cost
9. Access the file universally
10. Increase the security

References
[1] Deepika Saxena, Ishu Gupta, Jitendra Kumar, Ashutosh Kumar Singh, and Xiaoqing Wen. A secure and
multiobjective virtual machine placement framework for cloud data center. IEEE Systems Journal, 2021.
[2] Jitendra Kumar, Deepika Saxena, Ashutosh Kumar Singh, and Anand Mohan. Biphase adaptive learning-based
neural network model for cloud datacenter workload forecasting. Soft Computing, pages 1–18, 2020.
[3] Ashutosh Kumar Singh, Deepika Saxena, Jitendra Kumar, and Vrinda Gupta. A quantum approach towards the
adaptive prediction of cloud workloads. IEEE Transactions on Parallel and Distributed Systems, 2021.
[4] Ishu Gupta and Ashutosh Kumar Singh. A probabilistic approach for guilty agent detection using bigraph after
distribution of sample data. Procedia Computer Science, 125:662–668, 2018.
[5] Deepika Saxena and Ashutosh Kumar Singh. Communication cost aware resource efficient load balancing (care-lb)
framework for cloud datacenter. Recent Advances in Computer Science and Communications, 12:1–00, 2020.
[6] Jitendra Kumar and Ashutosh Kumar Singh. Dynamic resource scaling in cloud using neural network and black
hole algorithm. In 2016 Fifth International Conference on Eco-friendly Computing and Communication Systems
(ICECCS), pages 63–67. IEEE, 2016.
[7] Deepika Saxena and Ashutosh Kumar Singh. Auto-adaptive learning-based workload forecasting in dynamic
cloud environment. International Journal of Computers and Applications, pages 1–11, 2020.
[8] D Saxena and AK Singh. Security embedded dynamic resource allocation model for cloud data centre. Electronics
Letters, 56(20):1062–1065, 2020.
[9] Saxena Deepika and Singh Ashutosh Kumar. A proactive autoscaling and energy-efficient vm allocation framework
using online multi-resource neural network for cloud data center. Neurocomputing, 426:248–264, 2021.
[10] Deepika Saxena and Ashutosh Kumar Singh. OSC-MC: Online secure communication model for cloud environ-
ment. IEEE Communications Letters, 2021.
[11] Ishu Gupta, Niharika Singh, and Ashutosh Kumar Singh. Layer-based privacy and security architecture for cloud
data sharing. Journal of Communications Software and Systems, 15(2):173–185, 2019.
[12] Niharika Singh and Ashutosh Kumar Singh. Data privacy protection mechanisms in cloud. Data Science and
Engineering, 3(1):24–39, 2018.
[13] Sakshi Chhabra and Ashutosh Kumar Singh. A probabilistic model for finding an optimal host framework and
load distribution in cloud environment. Procedia Computer Science, 125:683–690, 2018.
[14] Niharika Singh and Ashutosh Kumar Singh. Sql-injection vulnerabilities resolving using valid security tool in
cloud. Pertanika Journal of Science & Technology, 27(1), 2019.
[15] Tian Wang, Jiyuan Zhou, Xinlei Chen, Guojun Wang, Anfeng Liu, and Yang Liu. A three-layer privacy preserving
cloud storage scheme based on computational intelligence in fog computing. IEEE Transactions on Emerging
Topics in Computational Intelligence, 2(1):3–12, 2018.

7
 Data Security & Privacy in Cloud Computing A P REPRINT

[16] Deepika Saxena, Ashutosh Kumar Singh, and Rajkumar Buyya. Op-mlb: An online vm prediction based multi-
objective load balancing framework for resource management at cloud datacenter. IEEE Transactions on Cloud
Computing, 2021.
[17] Ashutosh Kumar Singh and Deepika Saxena. A cryptography and machine learning based authentication for
secure data-sharing in federated cloud services environment. Journal of Applied Security Research, pages 1–24,
2021.
[18] Jitendra Kumar and Ashutosh Kumar Singh. Performance assessment of time series forecasting models for cloud
datacenter networks’ workload prediction. Wireless Personal Communications, 116(3):1949–1969, 2021.
[19] Deepika Saxena and Ashutosh Kumar Singh. workload forecasting and resource management models based on
machine learning for cloud computing environments. arXiv preprint arXiv:2106.15112, 2021.
[20] Jitendra Kumar, Ashutosh Kumar Singh, and Rajkumar Buyya. Ensemble learning based predictive framework
for virtual machine resource request prediction. Neurocomputing, 397:20–30, 2020.
[21] Sakshi Chhabra and Ashutosh Kumar Singh. Dynamic data leakage detection model based approach for mapre-
duce computational security in cloud. In 2016 Fifth International Conference on Eco-friendly Computing and
Communication Systems (ICECCS), pages 13–19. IEEE, 2016.
[22] AK Singh and Jitendra Kumar. Secure and energy aware load balancing framework for cloud data centre networks.
Electronics Letters, 55(9):540–541, 2019.
[23] Aman Singh Chauhan, Dikshika Rani, Akash Kumar, Rishabh Gupta, and Ashutosh Kumar Singh. A survey on
privacy-preserving outsourced data on cloud with multiple data providers. In Proceedings of the International
Conference on Innovative Computing & Communications (ICICC), 2020.
[24] Ashutosh Kumar Singh and Ishu Gupta. Online information leaker identification scheme for secure data sharing.
Multimedia Tools and Applications, 79(41):31165–31182, 2020.
[25] Deepika Deepika, Rajnesh Malik, Saurabh Kumar, Rishabh Gupta, and Ashutosh Kumar Singh. A review on data
privacy using attribute-based encryption. In Proceedings of the International Conference on Innovative Computing
& Communications (ICICC), 2020.
[26] Sakshi Chhabra and Ashutosh Kumar Singh. A secure vm allocation scheme to preserve against co-resident threat.
International Journal of Web Engineering and Technology, 15(1):96–115, 2020.
[27] Kamal Nayan Kaur, Ishu Gupta, Ashutosh Kumar Singh, et al. Digital image watermarking using (2, 2) visual
cryptography with dwt-svd based watermarking. In Computational intelligence in data mining, pages 77–86.
Springer, 2019.
[28] Kamaljeet Kaur, Ishu Gupta, and Ashutosh Kumar Singh. Data leakage prevention: e-mail protection via gateway.
In Journal of Physics: Conference Series, volume 933, page 012013. IOP Publishing, 2017.
[29] Jiawei Yuan and Shucheng Yu. Privacy preserving back-propagation neural network learning made practical with
cloud computing. IEEE Transactions on Parallel and Distributed Systems, 25(1):212–221, 2013.
[30] Qingchen Zhang, Laurence T Yang, and Zhikui Chen. Privacy preserving deep computation model on cloud for
big data feature learning. IEEE Transactions on Computers, 65(5):1351–1362, 2015.
[31] Ryo Yonetani, Vishnu Naresh Boddeti, Kris M Kitani, and Yoichi Sato. Privacy-preserving visual learning using
doubly permuted homomorphic encryption. In Proceedings of the IEEE International Conference on Computer
Vision, pages 2040–2050, 2017.
[32] Yoshinori Aono, Takuya Hayashi, Lihua Wang, Shiho Moriai, et al. Privacy-preserving deep learning via additively
homomorphic encryption. IEEE Transactions on Information Forensics and Security, 13(5):1333–1345, 2017.
[33] Ping Li, Jin Li, Zhengan Huang, Tong Li, Chong-Zhi Gao, Siu-Ming Yiu, and Kai Chen. Multi-key privacy-
preserving deep learning in cloud computing. Future Generation Computer Systems, 74:76–85, 2017.
[34] Ehsan Hesamifard, Hassan Takabi, Mehdi Ghasemi, and Rebecca N Wright. Privacy-preserving machine learning
as a service. Proc. Priv. Enhancing Technol., 2018(3):123–142, 2018.
[35] Ping Li, Jin Li, Zhengan Huang, Chong-Zhi Gao, Wen-Bin Chen, and Kai Chen. Privacy-preserving outsourced
classification in cloud computing. Cluster Computing, 21(1):277–286, 2018.
[36] Xindi Ma, Jianfeng Ma, Hui Li, Qi Jiang, and Sheng Gao. Pdlm: Privacy-preserving deep learning model on
cloud with multiple keys. IEEE Transactions on Services Computing, 2018.
[37] Tong Li, Zhengan Huang, Ping Li, Zheli Liu, and Chunfu Jia. Outsourced privacy-preserving classification service
over encrypted data. Journal of Network and Computer Applications, 106:100–110, 2018.

8
 Data Security & Privacy in Cloud Computing A P REPRINT

[38] Chong-zhi Gao, Qiong Cheng, Pei He, Willy Susilo, and Jin Li. Privacy-preserving naive bayes classifiers secure
against the substitution-then-comparison attack. Information Sciences, 444:72–88, 2018.
[39] Tran Thi Phuong et al. Privacy-preserving deep learning via weight transmission. IEEE Transactions on
Information Forensics and Security, 14(11):3003–3015, 2019.
[40] Reza Shokri and Vitaly Shmatikov. Privacy-preserving deep learning. In Proceedings of the 22nd ACM SIGSAC
conference on computer and communications security, pages 1310–1321, 2015.
[41] Qian Wang, Shengshan Hu, Minxin Du, Jingjun Wang, and Kui Ren. Learning privately: Privacy-preserving
canonical correlation analysis for cross-media retrieval. In IEEE INFOCOM 2017-IEEE Conference on Computer
Communications, pages 1–9. IEEE, 2017.
[42] Tingting Chen and Sheng Zhong. Privacy-preserving backpropagation neural network learning. IEEE Transactions
on Neural Networks, 20(10):1554–1564, 2009.
[43] Ankur Bansal, Tingting Chen, and Sheng Zhong. Privacy preserving back-propagation neural network learning
over arbitrarily partitioned data. Neural Computing and Applications, 20(1):143–150, 2011.
[44] Saeed Samet and Ali Miri. Privacy-preserving back-propagation and extreme learning machine algorithms. Data
& Knowledge Engineering, 79:40–61, 2012.
[45] Ning Cao, Cong Wang, Ming Li, Kui Ren, and Wenjing Lou. Privacy-preserving multi-keyword ranked search
over encrypted cloud data. IEEE Transactions on parallel and distributed systems, 25(1):222–233, 2013.
[46] Cheng Guo, Xue Chen, Yingmo Jie, Fu Zhangjie, Mingchu Li, and Bin Feng. Dynamic multi-phrase ranked
search over encrypted data with symmetric searchable encryption. IEEE Transactions on Services Computing,
2017.
[47] Zhangjie Fu, Xinle Wu, Chaowen Guan, Xingming Sun, and Kui Ren. Toward efficient multi-keyword fuzzy
search over encrypted outsourced data with accuracy improvement. IEEE Transactions on Information Forensics
and Security, 11(12):2706–2716, 2016.
[48] Zhangjie Fu, Lili Xia, Xingming Sun, Alex X Liu, and Guowu Xie. Semantic-aware searching over encrypted
data for cloud computing. IEEE Transactions on Information Forensics and Security, 13(9):2359–2371, 2018.
[49] Qiong Huang and Hongbo Li. An efficient public-key searchable encryption scheme secure against inside keyword
guessing attacks. Information Sciences, 403:1–14, 2017.
[50] Lianyong Qi, Xuyun Zhang, Wanchun Dou, and Qiang Ni. A distributed locality-sensitive hashing-based approach
for cloud service recommendation from multi-source data. IEEE Journal on Selected Areas in Communications,
35(11):2616–2624, 2017.
[51] Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to sensitivity in private data
analysis. In Theory of cryptography conference, pages 265–284. Springer, 2006.
[52] Sam Fletcher and Md Zahidul Islam. Differentially private random decision forests using smooth sensitivity.
Expert systems with applications, 78:16–31, 2017.
[53] Ping Li, Tong Li, Heng Ye, Jin Li, Xiaofeng Chen, and Yang Xiang. Privacy-preserving machine learning with
multiple data providers. Future Generation Computer Systems, 87:341–350, 2018.
[54] Tong Li, Jin Li, Zheli Liu, Ping Li, and Chunfu Jia. Differentially private naive bayes learning over multiple data
sources. Information Sciences, 444:89–104, 2018.
[55] Zhibo Wang, Xiaoyi Pang, Yahong Chen, Huajie Shao, Qian Wang, Libing Wu, Honglong Chen, and Hairong
Qi. Privacy-preserving crowd-sourced statistical data publishing with an untrusted server. IEEE Transactions on
Mobile Computing, 18(6):1356–1367, 2018.
[56] Alzubair Hassan, Rafik Hamza, Hongyang Yan, and Ping Li. An efficient outsourced privacy preserving machine
learning scheme with public verifiability. IEEE Access, 7:146322–146330, 2019.
[57] Xiaoqian Liu, Qianmu Li, Tao Li, and Dong Chen. Differentially private classification with decision tree ensemble.
Applied Soft Computing, 62:807–816, 2018.
[58] Ishu Gupta, Rishabh Gupta, Ashutosh Kumar Singh, and Rajkumar Buyya. Mlpam: A machine learning and
probabilistic analysis based model for preserving security and privacy in cloud environment. IEEE Systems
Journal, 2020.
[59] Jhilakshi Sharma, Donghyun Kim, Ahyoung Lee, and Daehee Seo. On differential privacy-based framework for
enhancing user data privacy in mobile edge computing environment. IEEE Access, 9:38107–38118, 2021.