The smart city vision has driven the rapid development and advancement of interconnected technologies using the Internet of Things (IoT) and cyber-physical systems (CPS). In this paper, various aspects of IoT and CPS in recent years (from 2013 to May 2023) are surveyed. It first begins with industry standards which ensure cost-effective solutions and interoperability. With ever-growing big data, tremendous undiscovered knowledge can be mined to be transformed into useful applications. Machine learning algorithms are taking the lead to achieve various target applications with formulations such as classification, clustering, regression, prediction, and anomaly detection. Notably, attention has shifted from traditional machine learning algorithms to advanced algorithms, including deep learning, transfer learning, and data generation algorithms, to provide more accurate models. In recent years, there has been an increasing need for advanced security techniques and defense strategies to detect and prevent the IoT and CPS from being attacked. Research challenges and future directions are summarized. We hope that more researchers can conduct more studies on the IoT and on CPS.

Applying deep learning (DL) algorithms for image classification tasks becomes more challenging with insufficient training data. Transfer learning (TL) has been proposed to address these problems. In theory, TL requires only a small amount of knowledge to be transferred to the target task, but traditional transfer learning often requires the presence of the same or similar features in the source and target domains. Cross-modality transfer learning (CMTL) solves this problem by learning knowledge in a source domain completely different from the target domain, often using a source domain with a large amount of data, which helps the model learn more features. Most existing research on CMTL has focused on image-to-image transfer. In this paper, the CMTL problem is formulated from the text domain to the image domain. Our study started by training two separately pre-trained models in the text and image domains to obtain the network structure. The knowledge of the two pre-trained models was transferred via CMTL to obtain a new hybrid model (combining the BERT and BEiT models). Next, GridSearchCV and 5-fold cross-validation were used to identify the most suitable combination of hyperparameters (batch size and learning rate) and optimizers (SGDM and ADAM) for our model. To evaluate their impact, 48 two-tuple hyperparameters and two well-known optimizers were used. The performance evaluation metrics were validation accuracy, F1-score, precision, and recall. The ablation study confirms that the hybrid model enhanced accuracy by 12.8% compared with the original BEiT model. In addition, the results show that these two hyperparameters can significantly impact model performance.

Predicting the occurrence of stressed and drowsy driving provides sufficient time to alert drivers and prevent undesired driving behaviours that result in road traffic accidents. Typically, the predictive machine learning models for stress and drowsy driving are formulated using supervised learning with labelled data. Labelled data are available only in training datasets, where labelling new data becomes infeasible and costly. With the ever-growing quantity of new and unlabelled data, once the model is deployed, unlabelled data quickly dominate the whole data population, and bias toward unlabelled data may increase the error rate of the prediction model. A preliminary study was thus conducted to train the predictive model using semi-supervised learning. We analyze and compare the performance of four algorithms, namely extreme gradient boosting (XGBoost), autoregression (AR), autoregressive integrated moving average (ARIMA), and exponential smoothing (ES). Various portions of unlabeled data were introduced, and more window sizes (in advance intervals) were chosen for the analysis.