The potential growth in data mining has an important aspect on security due to the consideration of the data as an asset. The provisioning of protection in a public infrastructure fails to ensure privacy disclosure of an individual’s information. Differential Privacy (DP) is a promising solution for assuring privacy protection by injecting noise using the Laplace mechanism or Exponential mechanism. The access of data by analysts is performed via edge devices. A common problem identified from previous research work is the leakage of privacy at the edge layer and data accessed by unauthorized people. To address the problem, this paper proposes DP-FCNN, that implements Differential Privacy using a Fuzzy Convolution Neural Network (FCNN) with Laplace Mechanism for injecting noise. The processes handled here are data processing and query processing. The dataset is uploaded by the data owner to the data provider, who is responsible for injecting noise and then encrypting with Piccolo encryption before uploading it into the cloud. Based on the uploaded dataset, the data owner constructs a hash index from the extracted key attributes by using the BLAKE2s algorithm for performing hashing. The hash index is fed into the edge server to form a Merkle hash tree due to the data leakage at the edge is eliminated. On the other hand, requests/queries by the data analyst are authenticated by the data provider. The hash tree in the edge server then searches for the corresponding data, extracting it from the cloud and delivers it to the data analyst in an encrypted format. Every authenticated data analyst is provided with a decryption key for retrieving the query result. This is implemented using Java and the results show better efficiency in terms of scalability, processing time and accuracy.

As contemporary systems are being operated in dynamic situations alternating into decentralized and distributed environments from conventional centralized frameworks, Federated Learning (FL) has been gaining attention for an effective architecture when aggregating the information and correspondingly trained AI model of geographically distributed datasets. However, the convergence of FL suffers from the skewed and biased non-IID local dataset acquired from the heterogeneous environment, which is common in real-world practice. To address this issue, we propose a novel Label-wise clustering algorithm in FL (FedLC) that guarantees the convergence among local clients that hold unique distribution. We theoretically analyze the non-IID attributes that potentially affect the performance, defining the six non-IID scenarios in hierarchical order. Through conducting experiments on the suggested non-IID scenarios, we empirically show that the vanilla FL aggregation model is incapable of gaining stable convergence, generating a biased model. By pre-evaluating the local dataset prior to training models, FedLC determines the potential contributions with respect to trainability in a global view, and adaptively selects the locals to collaboratively take part in while aggregation. Our experimental results show that the FedLC outperforms the state-of-the-art non-IID FL optimization studies, offering a robust convergence in a highly skewed and biased non-IID dataset.

Federated learning (FL) stands as a robust framework facilitating collaborative learning in distributed Internet of Things (IoT) settings. Nevertheless, the proliferation of nonidentically and independently distributed (non-IID) data sets presents formidable challenges, revealing distinct attributes among local participants within the IoT network. In response, clustering has been incorporated into FL methodologies, bolstering adaptability to heterogeneous data set characteristics and ameliorating the effects of non-IID conditions. Despite its effectiveness, the interpretive aspects and ramifications of clustering within FL have received insufficient scrutiny, constraining its broader utility in artificial intelligence-infused IoT ecosystems. This study exposes a notable drawback in traditional clustering methods within FL, labeled as “Cluster Collapse Disorder,” wherein cluster boundaries blur, leaving models susceptible to non-IID data and data drift. Inspired by this finding, we propose clustering-based semantic FL (CLSM-FL). Our framework employs a soft clustering technique based on the inherent attributes of local models, assigning them to semantic clusters for interpretable evaluation of their capabilities. Beyond its interpretive benefits, CLSM-FL ensures resilient performance in non-IID scenarios and against data-perturbed adversarial attacks. We substantiate its efficacy through extensive experiments covering various scenarios and data sets. This contribution not only underscores the crucial role of clustering in FL but also advances comprehension and practical application of FL methodologies in non-IID IoT environments.

Multimodal sentiment analysis plays a critical role in numerous IoT-driven applications, such as personalized smart assistants, healthcare monitoring systems, and intelligent transportation networks, where accurate interpretation of user emotions is vital for enhancing service quality. However, a severe threat of privacy leakage in the multimodal sentiment analysis has been overlooked by previous works. To fill this gap, we propose a differentially private correlated representation learning (DPCRL) model to achieve privacy-preserving multimodal sentiment analysis by combining a correlated representation learning scheme with a differential privacy protection scheme. Our correlated representation learning scheme aims to achieve heterogeneous multimodal data transformation to meet the requirements of privacy-preserving multimodal sentiment analysis by learning the correlated and uncorrelated representations, where especially, a predetermined correlation factor is employed to flexibly adjust the expected correlation among the correlated representations. The differential privacy protection scheme is used to obtain the disturbed correlated and uncorrelated representations by adding Laplace noise for $\epsilon $ -differential privacy. In particular, the correlation factor can help alleviate the side-effect of the added Laplace noise on the sentiment prediction performance. Finally, via conducting a series of real-data experiments, we validate that our proposed DPCRL model is superior to the state-of-the-art for privacy-preserving multimodal sentiment analysis.

There is growing attention in the metaverse from a variety of fields, and many Internet-of-Things (IoT) companies are exploring the possibility of integrating their existing business into the metaverse space. However, it still remains challenging to put the metaverse into practice. Among those challenges, trust management is critical to guarantee secure interactions and effective sharing among metaverse assets. In this article, we construct a TMETA system as a pioneer work for further establishing the metaverse and utilizing digital assets, which helps IoT startup companies expand their business during the cold-start phase with limited business knowledge and seed budget. In the TMETA system, IoT companies treat their interaction experience with resource providers as digital assets. When an IoT startup initiates IoT services with the assistance of expertise-diverse resource providers, task assignments are determined by leveraging the knowledge and trustworthy advice of expert IoT companies. With the aid of digital twin (DT) and blockchain technology, IoT tasks will be assigned to their suitable resource providers as the secured instruction from trust DTs of expert companies, which are evaluated and updated by trust evolution and advice aggregation. We demonstrate our proposed TMETA system’s practicability with synthetic and real-world data. The experimental results indicate that our proposed TMETA system can help IoT startups attain more favorable outcomes and reduce the task failure rate in various trust environments, thereby accelerating the cold-start process.

This paper proposes a novel data-owner-driven privacy-aware cloud data acquisition framework for intelligent big data analytics for service providers and users. To realize this idea, we propose three main components. The first one is a new global identity provider concept to support fine-grained access control for a federated outsourcing cloud, namely called P-FIPS (Privacy-enhanced Federated Identity Provider System), in which data owners perform identity access control with the operator of the federated outsourcing cloud so that the service providers can selectively use their encrypted data on the cloud for various purpose such as intelligent big data analytics. In P-FIPS, data owners manage the access privilege of service providers over their encrypted data on the cloud by (a) labeling the scope of use (e.g., user connection, user disconnection, user tracking) on each encrypted data on the cloud, and (b) by selectively providing the information regarding the data owners to the service provider. The label also includes the attributes related to the data owner’s identity, and this allows service providers to locate the target data with the assist of cryptographic computation according to the scope of the use at the cloud outsourcing server. The second one is a new ambiguous data acquisition mechanism integrated with P-FIPS from a cloud to a service provider. The last one is the Decentralized Audit and Ordering (DAO) Chain mechanism which provides the correctness of obtained data to the service provider as well as ensures the owners that their data is being used for the approved purpose only. Most importantly, we show that our framework is much more efficient than the existing alternative in the scheme.