As the IoT expands exponentially, the amount of data generated by individuals has increased. To process big data efficiently, machine learning (especially deep learning) has emerged. However, existing machine learning has the disadvantage of being vulnerable to data privacy because it sends raw data to the center. Therefore, federated learning (FL) was introduced to address this privacy problem, in which only learning parameters are sent to the center after training the user’s own local model with their own raw data. However, FL remains vulnerable to various attacks. In this paper, we propose an efficient and safe FL framework using the Interplanetary File System (IPFS) that minimizes the effect of data poisoning attacks on FL. In this system, the roles of nodes are divided into three: leader node, A-node (Aggregation-node), and T-node (Training-node). In this way, the A-node and T-node cannot manipulate the learning information, allowing the sharing of information and data safely through IPFS while protecting raw data with a similarity-based data shuffling scheme used by the A-node. Moreover, nodes with high accuracy receive more incentives and learning motivation, enhancing the overall efficiency of the network. Finally, the efficiency of the system is verified through related simulations.

As the IoT expands exponentially, the amount of data generated by individuals has increased. To process big data efficiently, machine learning (especially deep learning) has emerged. However, existing machine learning has the disadvantage of being vulnerable to data privacy because it sends raw data to the center. Therefore, federated learning (FL) was introduced to address this privacy problem, in which only learning parameters are sent to the center after training the user’s own local model with their own raw data. However, FL remains vulnerable to various attacks. In this paper, we propose an efficient and safe FL framework using the Interplanetary File System (IPFS) that minimizes the effect of data poisoning attacks on FL. In this system, the roles of nodes are divided into three: leader node, A-node (Aggregation-node), and T-node (Training-node). In this way, the A-node and T-node cannot manipulate the learning information, allowing the sharing of information and data safely through IPFS while protecting raw data with a similarity-based data shuffling scheme used by the A-node. Moreover, nodes with high accuracy receive more incentives and learning motivation, enhancing the overall efficiency of the network. Finally, the efficiency of the system is verified through related simulations.

Recently, big data related to human movement, air quality, and meteorology have been generated in urban computing through sensing technology and the computing infrastructure. However, security problems arise as data utilization increases. If the sensing data from internet of things devices are constantly exposed, the users’ private information can be determined, a critical security risk that could result in privacy breaches. This paper proposes a secure data processing system using the blockchain and differential privacy for data security and privacy protection in urban computing. When a service provider requests information, the system generates it from urban computing data using machine learning. We apply differential privacy to these data to protect privacy. However, if a query repeats, differential privacy may provide insufficient privacy protection. Therefore, we reduce the total privacy cost by reusing noise for the same data and privacy parameters using the blockchain. Machine learning accuracy may decrease when noisy data are used for training. Thus, we increase accuracy by storing and appropriately using the model parameters generated by the same data in the blockchain. We design, simulate, and analyze the results of an experimental environment for reusing noise for differential privacy and parameter utilization of machine learning using the blockchain. The proposed approach reduces privacy costs compared to the existing mechanism while protecting data privacy. We demonstrate that, through parameter utilization, the accuracy improves compared to conventional mechanisms.

Unmanned Aerial Vehicles (UAVs) are increasingly integral in various sectors, simultaneously encountering rising security threats as UAV and Urban Air Mobility (UAM) networks continue to expand. This paper addresses the challenge of securing UAM networks while also emphasizing generalizability of the security solution to protect heterogeneous UAVs against threats that compromise their stability, reliability and can cause catastrophic failures such as a crash landing. The deployment of traditional machine learning (ML) based intrusion detection systems (IDSs) is often hampered in real-world applications due to a lack of generalizability of the security solution. As a result, the system fails to provide adequate security across the varying models and platforms of UAVs, each with its unique statistical properties and data distributions. To address these challenges, we focus on employing a comprehensive set of UAV sensor parameters, tailored feature engineering and selection to develop multi-stage cross-validated ensemble learning systems to facilitate generalized detection of attack and non-attack cases. For additional analysis, we cross-validate the models using two different cross-validation techniques. The proposed stacking ensemble systems provide the overall best performance, with AUC within the range of 92% to 100% across different cross-validations.

Over the past few years, Generative Adversarial Network(GAN) has experienced significant growth in various applications as a generative model. However, the stability issues in training remain a challenge in GANs. To mitigate these problems, this paper proposes a novel GAN model that applies dual-parallelized generators. This study designs new methodologies by inputting three sets of data to the discriminator and updating the average of the loss values. Experimental results show that the proposed model shows an ideal convergence graph and reduces the loss by about 40%. The results also show an improvement in the quality of the generated data, with the model achieving stability during the training process.

As the Internet of Things (IoT) grows exponentially, it is becoming deeply embedded in our daily lives. As the quantity and quality of data produced by devices have also gradually increased, there have been increasing attempts to use these useful IoT big data for various applications and to combine IoT with machine learning and deep learning to process a large amount of useful data. However, in the centralized deep learning method, privacy issues have been raised because the server can use personal data collected from the user’s IoT. Due to this reason, Federated Learning (FL) method that can protect users' personal data while doing machine learning has been studied. However, current FL also has the possibility of data poisoning attacks and other problems. Therefore, this work, by suggesting distributed FL framework combined with Interplanetary File System and Differential Privacy, proposes a method that allows users to participate in FL safely and efficiently. Through this method, participants share some parts of data, and these data are collected by specific nodes. These data are combined to make a new dataset of FL network for defending against data poisoning attack and vouch for training’s accuracy. Also, an aggregation mechanism is proposed to suppress the effect of a malicious node poisoning attack. Finally, this framework is tested in python environment. With this method, one can freely open a project and anyone can join in with distributed condition, even when he or she has no enough dataset for learning but computing capability, vice versa. If a malicious node tries to interrupt the learning with poisoned dataset, aggregation mechanism and combined validation set from the network’s nodes will suppress the bad effect. We have tested through python and open-source code to verify the efficiency and privacy.

Recently, given the current situation the world is undergoing with the COVID-19 pandemic, a lot of companies encourage remote working from home. As a result, the use of the cloud has increased, but as a consequence, cyberattacks on the cloud systems have escalated. However, organizations and companies still have concerns about the use of the cloud as most of them remain unaware of the security threats against cloud system. This work researches through major concerns of the organizations and companies for cloud security and tries to explain why it is important for the organization to be aware of the threats that the cloud is facing. It also proposes a three-step strategy to help companies and organizations to secure their cloud system.

Due to the explosive increase in IoT devices and traffic, big data is developing into smart data that helps the data science experts understand human activities, through the relationship between mobility and resource application of the users in public spaces. For example, smart data markets help to predict crimes or understand the infection the cause of COVID-19 infections. For these smart services, the users agree to the privacy policy so that the personal and sensitive information can be collected by a third party. But the conditions of the privacy policy do not specify whether the information of the users can be tracked. To ensure data transparency, many systems are applying consortium/private blockchains with raft algorithm. The raft algorithm requires nodes to check countless messages for a single transaction. Eventually, as the number of nodes increases, the overall system degradation is derived from the burden of the leader node. This paper proposes a method to process the collected transactions by dividing a certain amount of transactions into cells, without any extra protocol. The proposed scheme also uses the federated learning model with high accuracy and data privacy, in order to determine the optimized cell size in a blockchain system that should lead to consensus on multiple servers. Therefore, the proposed CBR(Cell-based Raft) consensus algorithm proposes a protocol that reduces the number of messages, without interfering with the concept of the existing raft algorithm, in order to maintain stable throughput in the smart data market where massive transactions occur.

In urban computing composed of various Internet of Things (IoT) devices, data is collected from IoT as learning data for other IoT devices in a circular structure. Therefore, data are shared resources and crucial in urban computing. In particular, group communication is essential for effective data sharing and for updating software in an IoT system. However, changing the group key is imperative for data security in group communication whenever the group member changes. When the movement of IoTs is high, group key redistribution must be repeated because the group membership keeps changing. In changing and redistributing the group keys, overhead occurs, inevitably degrading system performance. In addition, since much data traffic generates in IoT devices’ crowded areas, transmission delay and data loss could degrade system performance. Due to IoT’s low-power and low-capacity characteristics, one-to-one communication is inefficient, requiring efficient group management and group key management. This work proposes a hierarchical blockchain-based group and group key management scheme to establish an efficient communication environment in urban computing. We adopted blockchains to track the movement and density of IoT and secure node authorization. Using the upper layer blockchain, the unmanned aerial vehicle (UAV) determines the movement and density of IoTs. Using the lower-layer blockchain, base stations (BSs) identify the IoT’s movement of information in each group. We included only nodes determined to be safe in the group. Through the hierarchical blockchain, while protecting the IoT’s privacy, we can record the information in the blockchain to determine the mobility of nodes and the node density of a group. We set several experimental environments, and analyzed the efficiency by simulating the addition of secondary gateways and group integration. As a result, we showed that our proposal establishes an efficient communication environment and reduces computational and communication overhead.