Smart Cyber-Physical systems

Smart Cyber-Physical systems

The evolution of machine learning (ML) technology has expanded the capabilities of the Internet of Things (IoT). Nowadays, ML techniques are actively employed to address challenges within IoT environments and to develop ‘smart’ IoT services. However, due to the inherent constraints of the IoT environment, several open challenges exist when it comes to utilizing ML in this context. Distributed data, security vulnerabilities, and limited hardware resources can pose practical barriers to developing ideal ML systems in IoT. Our research team has conducted studies focused on ML systems that can overcome these limitations and effectively address problems within IoT environments.

Federated learning for distributed CPS

Federated learning (FL) is a distributed ML technique that emerged in response to growing privacy concerns and the need for decentralized data processing. Traditional machine learning models require centralized access to vast datasets, often raising serious privacy and security issues. FL addresses these concerns by enabling model training across distributed devices or servers while keeping the data localized. FL allows each distributed client to train its local model with its own data. Then, FL also appropriately aggregates the local models to obtain a global model. This FL technique does not require the exchange of local data, but it enhances the privacy of training data, making it a powerful tool for privacy-conscious applications such as healthcare, finance, and personalized recommendation systems while ensuring data remains in the hands of its owners.

Application: Federated learning for malware detection in IoT networks

With the rapid expansion of IoT devices, malware has become a critical risk for IoT networks. The large-scale Mirai DDoS attack that occurred in 2016 demonstrates the significant threat posed by IoT devices when used in security attacks. Many studies have been conducted to detect malware in IoT networks, and recently, the utilization of ML techniques has shown high performance in this field.

However, given that data collected from IoT devices tends to be widely distributed, the conventional method of training ML models with centralized data presents several challenges, including communication costs for gathering distributed data and security threats to the data. Considering these challenges, we have been working on applying federated learning (FL) techniques to detect malware in IoT networks. However, Despite the advantage of preserving privacy for clients’ data, federated learning also presents inherent challenges. One of the most typical problems is the risk of non-independent and identically distributed (non-IID) client data. Differences in client data characteristics result in variations in their local models, leading to instability and performance degradation for the global model combined with the local models. This problem is critical when detecting malware using FL in an IoT environment because the network traffic patterns for each IoT device vary significantly. Our research team has focused on addressing the challenge posed by non-IID data and finding a way to build a robust, reliable, and scalable FL system for malware detection.

Application: Smart agriculture

Smart agriculture is an approach to farming that leverages technologies to optimize agricultural practices and maximize crop yields while minimizing resource consumption. This innovative field harnesses the power of data analytics, Internet of Things (IoT) devices, remote sensing, and automation to monitor and manage various aspects of farming, including soil health, irrigation, pest control, and livestock management. By collecting real-time data from sensors and drones, farmers can make data-driven decisions to optimize planting, irrigation schedules, and fertilizer application. Smart agriculture not only increases productivity but also reduces environmental impact by minimizing water and chemical usage. It holds immense promise in addressing global food security challenges and creating sustainable agricultural systems capable of feeding a growing world population.