Guest Editorial Introduction for the Special Section on Deep Learning Algorithms and Systems for Enhancing Security in Cloud Services

Deep learning is one of the major disciplines within Artificial Intelligence (AI) that uses complex algorithms to make computers understand any task that a human does. Neural networks are the base platform on top of which deep learning algorithms are developed. It can be easily said that most of the sophisticated and AI-based systems use deep learning algorithms for feature extraction, understanding, classification, and prediction. With the growing volumes of digital data which is available courtesy of several social media platforms, deep learning can serve as a vital catalyst in discovering new patterns and trends from these digital data. Deep learning concepts have been applied across a wide range of business and commercial activities such as medicine, e-commerce, stock market analysis and predictions, e-banking transactions, retail promotions, etc. Deep learning can be applied in any space of digital finance and marketing. However, a sufficient amount of “training” data should be available.

The collection of articles included in this special section address the challenges arising when applying deep learning principles for cloud services and also provide deep learning-based solutions for enhancing security in Cloud Service.

In this article “Machine Learning and Soil Humidity Sensing: Signal Strength Approach”, the author explains the IoT vision of ubiquitous and pervasive computing for future smart irrigation systems comprising the physical and digital worlds. Further, the vision includes implementing a smart irrigation ecosystem combined with machine learning and successfully solving the soil humidity sensing task for optimal water usage. The author explores a concept of a novel, low-power, LoRa-based, cost-effective system which achieves humidity sensing using deep learning techniques that can be employed to sense soil humidity with high accuracy simply by measuring the signal strength of the given underground beacon device. The results show that use of long short-term memory (LSTM) neural network as a deep learning approach provided significant results in terms of accuracy and estimation.

In the article titled “DANCE: Distributed Generative Adversarial Networks with Communication Compression”, the author focuses on deploying and training general adversarial networks (GANs) at the edge, rather than converging edge data to the central node. In order to address this problem, the author has designed a novel distributed learning architecture for GANs, called DANCE. DANCE can adaptively perform communication compression based on the available bandwidth, while supporting both data and model parallelism training of GANs. In addition, inspired by the gossip mechanism and Stackelberg game, a compatible algorithm, AC-GAN, is proposed. The theoretical analysis and results guarantee the convergence of the model and the existence of approximate equilibrium in AC-GAN and improves both simulation and prototype system.

In the article named “Multi-Tier Stack of Blockchain with Proxy Re-encryption Method Scheme on the IoT Platform”, the researcher proposes a block chain-based proxy re-encryption program to resolve both the trust and scalability problems, and to simplify the transactions. After encryption, the system saves the internet of things (IoT) data in a distributed cloud. The framework offers dynamic, smart Contacts between the sensor and the device user without the intervention of a trustworthy third party to exchange the captured IoT data. It uses an efficient proxy re-encryption system, which provides the owner and the person existing in the smart contract to see the data. Here, the splitting of proxy re-encryption method (Split-PRE) has been suggested based on the IoT to improve security and privacy in a private block chain. The experimental outcomes show that the proposed approach enhances the efficiency, security, privacy, and feasibility of the system.

In the article “Deep-Confidentiality: IoT-enabled Privacy Preserving Framework for Unstructured Big Biomedical Data”, the contributor has detailed a novel framework for textual medical data privacy. The proposed framework improves medical entity recognition using deep neural networks and sanitization compared to the current state-of-the-art techniques. Moreover, a new and generic utility metric is also proposed, which overcomes the shortcomings of the existing utility metric. It provides the true representation of sanitized documents as compared to the original documents. The results show that this method improves the confidentiality of patient information in cloud making it fully accessible.

In the research paper “A Flexible and Privacy-Preserving Collaborative Filtering Scheme in Cloud Computing for Vehicular Ad Hoc Network (VANETs)”, the author describes the development of an efficient machine learning algorithm for VANETs. Based on homomorphic encryption and secure multiparty computing technology, a flexible and privacy-preserving collaborative filtering scheme is proposed to accomplish a personalized recommendation for users, which is based on users’ interests and locations. On the one hand, the data can be updated by users flexibly to ensure the freshness and accuracy of the dataset of interest. On the other hand, the weighted values of user interests can be safely sorted to effectively improve the accuracy of collaborative filtering. Moreover, a novel collaborative filtering algorithm based on the homomorphic encryption technology is designed, which can guarantee the decryption result by machine learning. Both theoretical and experimental analyses demonstrate that the proposed scheme is secure and efficient for collaborative filtering in cloud computing in VANETs.

In the article named “Privacy-Preserving Distributed Multi-Task Learning Against Inference Attack in Cloud Computing”, the author proposes a novel privacy-preserving mechanism for distributed multitask learning (MTL), namely NOInfer, to allow several task nodes to train the model locally and transfer their shared knowledge privately. Specifically, the author constructs a single-server architecture to achieve the private MTL, which protects task nodes’ local data even if n−1 out of n nodes colluded. Then, a new protocol for alternating direction method of multipliers (ADMM) is designed to perform the privacy-preserving model training, which resists the inference attack through the intermediate results and ensures that the training efficiency is independent of the number of training samples. The results confirm that, over two testing datasets and evaluation results, it is demonstrated that NOInfer efficiently and effectively achieves the distributed MTL.

In “IoT-based Cloud Service for Secured Android Markets Using Program Dependency Graph (PDG)-based Deep Learning Classification”, the author addresses the software in commercial software piracy, as pirated applications on Android app stores are harming developers and their users by clone scammers using an IoT-based cloud architecture for clone detection using program dependency analysis. The dependency graphs of Java files are generated to extract a set of weighted features. The stacked-long short-term memory (S-LSTM) deep learning model is designed to predict possible clones. Also, the researcher has extracted the linear features from PDG and processes them for deep learning classification which changes the topology of the graphical structures and has added embedding methods to input the PDG graphical features directly to the deep learning model. The proposed approach is a successful method of classifying cloned applications in various Android stores.

In the research paper entitled “On the Neural Backdoor of Federated Generative Models in Edge Computing”, the author discusses how, in an edge computing environment, generative models (GMs) have been found to be valuable and useful in machine learning tasks such as data augmentation and data pre-processing. federated learning (FL) and distributed learning (DL) refer to training GMs in the edge computing network. However, FL and DL also bring additional risks to GMs since all peers in the network have access to the model under training. To enhance the required poisonous data samples and cope with dynamic network environments. The experimental results show that the neural backdoors can be successfully embedded by including poisonous samples in the local training dataset of an attacker.

The author in the article “InFeMo: Flexible Big Data Management through a Federated Cloud System”, describes a novel architecture scenario based on cloud computing and count on the innovative model of federated learning. The proposed model named Integrated Federated Model, has the acronym InFeMo. InFeMo incorporates all the existing cloud models with a federated learning scenario, as well as other related technologies that may have integrated use with each other, offering a novel integrated scenario. In addition to this, a proposed model is motivated to deliver a more energy-efficient system architecture and environment for the users. Proposed systems are built on the resources by cloud service providers (CSPs), and the PaaS (Platform as a Service) model, to handle user requests better and faster.

In the paper entitled “A Shared Two-way Cybersecurity Model for Cloud Service Sharing for Distributed User Applications”, the author explains a shared two-way security model (STSM) to provide adaptable service security for the end-users. In this security model, a cooperative closed access session for information sharing between the cloud and end-user is designed with the help of cybersecurity features. This closed access provides less complex authentication for users and data that is capable of matching the verifications of the cloud services. A deep belief learning algorithm is used to differentiate the cooperative and non-cooperative secure sessions between the users and the cloud to ensure closed access throughout the data sharing time. The output of the network decides the actual session time between the user and the cloud, improving the span of the sharing session. Besides, the proposed model reduces false alarm and communication failures, under controlled complexity.

In the research paper “Mobile Crowd-Sensing Applications: Data Redundancies, Challenges, and Solutions”, the author gives details about the mobile crowd-sensing system which is a cyber-physical system (CPS) that allows people with mobile devices to collect and supply data to CPSs’ centers. In practical mobile crowd-sensing applications, due to limited budgets for the different expenditure categories in the system, it is necessary to minimize the collection of redundant information to save more resources for the investor. Here, researchers have studied the problem of selecting participants in mobile crowd-sensing systems without redundant information such that the number of users is minimized, and the number of records (events) reported by users is maximized, also known as the participant-report-incident redundant avoidance (PRIRA) problem. The author has proposed a new approximation algorithm, called the maximum-participant-report algorithm (MPRA) to solve the PRIRA problem. The experimental results show that proposed method performs well within reasonable bounds of computational complexity.

In the article “Enhancing Security Problem Based-Deep learning in Mobile Edge Computing”, the author proposes new mobile edge computing (MEC) to perform all or part of the mobile device (MD’s) task, which greatly reduces the energy consumption of the MD and improves the quality of service (QoS) for application. However, offloading tasks to the edge server is vulnerable to attacks such as tampering and snooping, resulting in a deep learning (DL) security feature developed by major cloud service providers. Here, the author uses an effective security strategy method to minimize ongoing attacks in the MEC setting. The algorithm is based on the synthetic principle of a special set of strategies, and it can quickly construct suboptimal solutions even if the number of targets reaches hundreds of millions. In addition, for a given structure and a given number of patrollers, the upper bound of the protection level can be obtained, and the lower bound required for a given protection level can also be inferred. These bounds apply to universal strategies. The experiments show that the algorithm is better in performance.

The ease and scalable aspects of deep learning principles make it an instantaneous choice for adapting into any domain of decision making. This special issue potentially investigates articles that enhance security in cloud services. It also gives an overview of a multidisciplinary field of study concentrating on all types of innovative data services in the cloud.

We would like to thank Editor in Chief Liu Ling and the editorial staff of ACM and ACM TOIT for their support in difficult times, and all authors for their valuable contributions.

Dr. Gunasekaran Manogaran

Howard University, Washington D.C., USA

Dr. Hassan Qudrat-Ullah

York University, Toronto, Canada

Dr. Qin Xin

University of the Faroe Islands, Faroe Islands

Dr. Latifur Khan

The University of Texas at Dallas, Texas, USA