Internet of Things (IoT) is a large network of interconnected sensors, instruments, and devices, allowing a user to attain a higher degree of intelligent services. When embedded with sophisticated computational intelligence (CI) tools, the performance of IoT is further enhanced to attain higher operational efficiency, productivity, usability, safety, etc. For example, applications such as robotics, autopilot systems, and medical diagnosis were greatly enhanced by adapting the advanced techniques in neural network-based object detection and semi-supervised action learning. CI techniques used in IoT must be efficient and robust to perform well economically. The availability of accurately labelled datasets is critical to achieving this.

However, there are several challenges in regard to data collection in IoT. Firstly, given the size of IoT, and the different types, and the sheer volume of data produced, it’s a daunting task for companies to collect and store the data efficiently. Secondly, the execution of privacy laws (such as the Europe Union General Data Protection Regulation and California Consumer Privacy Act) further poses restrictions on data collection, making the task even more difficult. Thus, the data collection and storage issues and privacy laws result in many sparse datasets in IoT. It is crucial to find solutions to solve efficiently and securely these problems in IoT. There are several techniques such as differential privacy, secure multi-party computation (MPC), homomorphic encryption (HE), federated Learning (FL), transfer learning, meta-learning and generative adversarial network (GAN) that can be adapted to solve the sparse data challenge in IoT securely. Differential privacy allows one to publicly share information about a dataset by describing the patterns of groups within the dataset while withholding information about individuals. MPC is a novel method enabling parties to jointly compute a function over their inputs while keeping the inputs themselves private by applying cryptographic techniques like Homomorphic Encryption. FL allows multiple decentralized edge devices/clients to collaboratively train a machine learning model without exchanging their data. Transfer learning can retain the knowledge gained while solving one problem and apply it to a different but related problem. In meta-learning, it is possible to learn, select, alter, or combine different learning algorithms to solve a given learning problem effectively. Generative adversarial networks (GAN) can be used to generate (artificial) data to provide more samples for training purposes. There is a need to design secure data publication/exchange algorithms, build highly efficient collaborative training platforms, and generate artificial data. Moreover, there is a need to invent novel learning methods that can better utilize sparse data to enhance the IoT applications.

The aim of this Special Issue is to focus on the frameworks, platforms, software systems, applications, architectures, algorithms, and emerging technologies that solve sparse data challenges in IoT. We welcome original research and review articles in the field.

Potential topics include but are not limited to the following:

• Differential privacy-based data publication
• Federated learning for solving sparse data challenges in IoT
• Transfer learning for solving sparse data challenges in IoT
• Meta-learning for solving sparse data challenges in IoT
• Secure collaborative learning for solving sparse data challenges in IoT
• GAN-based solutions for solving sparse data challenges in IoT
• Hybrid computational solutions for solving sparse data challenges in IoT
• Sparse mobile crowdsensing methodologies for IoT applications
• Evaluating spatial and temporal correlation in sparse IoT data
• Inference and assessment of sparse IoT data quality/usability