Dependable System

We focus on designing algorithms to improve the dependability and performance of various distributed systems such as the Internet, ITS, IoT, blockchain (bitcoin), sensor networks, and nano-scale systems.

• Self-stabilizing algorithms that autonomously recover from failures
• Mobile agents that autonomously move around networks
• Application of random algorithms in sequential processing to distributed algorithms
• Probabilistic analysis in natural random procedures such as random walk
• Swarm robotics
• Distributed algorithm for dynamic graphs
• Distributed ledgers used in block chains
• Population protocols for nano-scale systems
• Dynamic distributed algorithms

Integrated circuits are core devices in various systems. The miniaturization of transistors enabling larger scale, high performance, and more power-efficient integrated circuits supports today's information society and AI advancement. However, testing integrated circuits has become increasingly complex, and reliability concerns due to transitor aging and security concerns such as hardware Trojan circuit insertion due to complicated supply chains have emerged. Furthermore, as CMOS transistor miniaturization reaches physical limits, new devices and architectures are expected, making reliability assurance a crucial challenge. Our laboratory addresses various integrated circuit dependability issues, including reliable design of memristor neural networks, improving integrated circuit security through hardware Trojan detection, and enhancing test quality using machine learning.

• Improving memristor neural network reliability through fault injection training
• Hardware Trojan detection methods using power side-channel learning consideringrandom variations
• Improving integrated circuit test quality using machine learning

Quantum information science has attracted lots of attention in recent years. In particular, facing the threat from quantum computers, researchers proposed the Quantum Key Distribution (QKD), a quantum communication protocol, as a potent countermeasure. QKD uses polarization states of photons to generate secret keys, and it is able to detect and quantify any possible eavesdropping attack and ensure secure communications. However, vulnerabilities called side-channels exist in the sources and detection devices in practical QKD systems. To address such side-channels, we develop measurement-device-independent (MDI) QKD protocols and fully passive QKD sources that enable more practically secure QKD systems. Meanwhile, noisy intermediate-scale quantum devices (NISQ) saw much development in the past decade, and we are interested in exploring the new field of quantum machine learning on NISQ devices and in particular its security. Our laboratory collaborates with researchers from institutes in e.g. Canada, Spain, and Hong Kong on the following past/ongoing topics:

• Measurement-device-independent quantum key distribution protocols
• Fully passive quantum key distribution sources
• Security of quantum machine learning

Cyber attacks through hackers exploiting security vulnerabilities and new types of malware are increasing daily, posing serious threats to information and communication technology security. To address these cyber attacks, network security technologies to protect against communication eavesdropping and impersonation attacks, and privacy enhancement technologies to protect information contained in data are essential.
Our laboratory tackles various security and privacy challenges, including zero-trust network access to prevent information leakage via malware, privacy protection during data collection and utilization for AI model construction, and prevention of poisoning attacks against AI models themselves. In collaboration with research institutions in the United States, France, and Bangladesh, we are currently working on the following research themes:

• Zero-trust architecture design for blocking lateral movement by malware
• Secure GPS data collection method using Differential Privacy
• Contaminated data separation method for preventing AI model poisoning attacks