The fast development of the Internet of Things (IoT) and deep learning enables learning useful patterns from the massive amount of collected data with sporadic nodes in IoT systems. Federated learning has received increasing attention in distributed machine learning where only intermediate parameters are exchanged with training samples that resided at local nodes. Nonetheless, most of the existing federated learning schemes assume a homogeneous distribution of data. The assumption, however, does not apply to IoT systems because of the heterogeneity of the IoT architecture. The nonindependent and identical distribution (non-IID) property in data volume and statistical distribution of IoT nodes can impact the performance of an aggregated global model that fits all nodes. Existing federated learning solutions for non-IID data sets either have to train additional models or require extra data exchange to check the node distribution. However, due to resource constraints in IoT systems, these approaches will increase the burden on limited computation capacity and cause network overhead. To address the issue, in this article, a novel weight-similarity-based client clustering (WSCC) approach is proposed, in which clients are split into different groups based on their data set distributions. An affinity-propagation-based method with the cosine distance of the client’s weight parameters is designed to iteratively and automatically determine dynamic clusters. The proposed approach is ideal for IoT systems since there are no auxiliary models and extra data transmissions are needed. Through the theoretical convergence analysis and empirical results, we show that our proposed WSCC scheme outperforms the representative federated learning schemes under different non-IID settings, achieving up to 20% improvements in accuracy.

The simultaneous development of deep learning techniques and Internet of Things (IoT)/Cyber-physical Systems (CPS) technologies has afforded untold possibilities for improving distributed computing, sensing, and data analysis. Among these technologies, federated learning has received increased attention as a privacy-preserving collaborative learning paradigm, and has shown significant potential in IoT/CPS-driven large-scale smart-world systems. At the same time, the vulnerabilities of deep neural networks, especially to adversarial attacks, cannot be overstated and should not be minimized. Moreover, distributed nature of federated learning makes defense against such adversarial attacks a more challenging problem due to unavailability of local data and resource heterogeneity. To tackle these challenges, in this paper, we propose ZeKoC, a Zero Knowledge Clustering approach to mitigate adversarial attacks. Particularly, we first formulate the problem of resource-constrained adversarial mitigation. Specifically, noting that global server has no access to training samples, we reformulate the unsupervised weight clustering problem. Our proposed ZeKoC approach allows the server to automatically split and merge weight clusters for weight selection and aggregation. Theoretical analysis of our approach demonstrates that convergence is guaranteed. Our experimental results illustrate that, in a non-i.i.d. (i.e., independent-and-identically distributed) data setting, our ZeKoC approach can successfully mitigate general attacks while outperforming state-of-art schemes.

Electric power systems are critical infrastructure and are vulnerable to contingencies including natural disasters, system errors, malicious attacks, etc. These contingencies can affect the world's economy and cause great inconvenience to our daily lives. Therefore, security of power systems has received enormous attention for decades. Recently, the development of the Internet of Things (IoT) enables power systems to support various network functions throughout the generation, transmission, distribution, and consumption of energy with IoT devices (such as sensors, smart meters, etc.). On the other hand, it also incurs many more security threats. Cascading failures, one of the most serious problems in power systems, can result in catastrophic impacts such as massive blackouts. More importantly, it can be taken advantage by malicious attackers to launch physical or cyber attacks on the power system. In this paper, we propose and investigate cascading failure attacks (CFAs) from a stochastic game perspective. In particular, we formulate a zerosum stochastic attack/defense game for CFAs while considering the attack/defense costs, budget constraints, diverse load shedding costs, and dynamic states in the system. Then, we develop a Q-CFA learning algorithm that works efficiently in power systems without any a priori information. We also formally prove that the convergence of the proposed algorithm achieves a Nash equilibrium. Simulation results validate the efficacy and efficiency of the proposed scheme by comparisons with other state-of-the-art approaches.

Cyber-physical systems (CPS) refer to systems that integrate communication, control, and computational elements into physical processes to facilitate the control of physical systems and effective monitoring. The systems are designed to interact with the physical world, monitor and control the physical processes while in operation, and generate data. Deep Neural Networks (DNN) comprise multiple layers of interconnected neurons that process input data to produce predictions. Spatial-temporal data represents the physical world and its evolution over time and space. The generated spatial-temporal data is used to make decisions and control the behavior of CPS. This paper systematically reviews the applications of DNNs, namely convolutional, recurrent, and graphs, in handling spatial-temporal data in CPS. An extensive literature survey is conducted to determine the areas in which DNNs have successfully captured spatial-temporal data in CPS and the emerging areas that require attention. The research proposes a three-dimensional framework that considers: CPS (transportation, manufacturing, and others), Target (spatial-temporal data processing, anomaly detection, predictive maintenance, resource allocation, real-time decisions, and multi-modal data fusion), and DNN schemes (CNNs, RNNs, and GNNs). Finally, research areas that need further investigation are identified, such as performance and security. Addressing data quality, strict performance assurance, reliability, safety, and security resilience challenges are the areas that are required for further research.

Cognitive Radio (CR) technology enables secondary users (SUs) to opportunistically access unused licensed spectrum owned by primary users (PUs). It has the potential to significantly enhance communication capacity, which is very critical to the next-generation wireless network design and has attracted intensive attention. One of the key issues in CR communications is to detect spectrum availability. Traditional approaches rely on spectrum sensing techniques to address this problem, which, however, consume considerable energy and time, and require complex prior information from PUs. In this paper, we develop a hierarchical spectrum learning system that takes advantage of the fine-tuned convolutional neural network (CNN) and the gated recurrent unit network (GRU), which is called the dual CNN and GRU (DCG), for spectrum availability prediction. Particularly, this model performs accurate predictions on local spectrum availability for each SU without any prior information of PUs. On the other hand, knowing their spectrum availability does not necessarily enable two SUs to successfully communicate on the same channel. This is a challenging problem and has been largely ignored by previous studies designing learning models for CR communications. Towards this goad, we design an enhanced DCG model called EDCG to enable two SUs to find the same channel to communicate with each other by performing channel selection prediction. The performance of the designed DCG and EDCG models is demonstrated through extensive and thorough simulations. The results show that our designed models achieve high prediction accuracy with limited training overhead.