Seminars by Prof. Kouichi Sakurai and Prof. Chunhua Su

Date: Wednesday, 18 March 2026

Time: 2:00pm to 4:00pm

Venue: Room 6591 (via lift 29/30), HKUST

Speaker (I): Prof. Kouichi Sakurai
Kyushu University

Title: The Power and Limitations of Adversarial Machine Learning: Empirical Insights from One‑Pixel Attacks and Emerging Challenges for Large Language Models

Abstract:

Facial Expression Recognition (FER) is increasingly used in applications such as surveillance, healthcare, and video analytics. Although adversarial attacks have been widely studied, attacks specifically targeting FER systems remain limited. Our research group originally proposed the One‑Pixel Attack (OPA), which uses Differential Evolution to identify critical pixels and generate minimal perturbations capable of fooling deep learning models. However, exhaustive pixel search is costly and can reduce black‑box attack success when irrelevant regions are selected. We introduce an improved Differential Evolution–based method that shortens search time and enhances targeted and untargeted attack success rates. Experimental results on a FER system are presented. The talk also highlights recent research directions, including extending OPA to intrusion detection systems (IDS) and discussing its potential applicability to large language models (LLMs).

Biography:

Dr. Kouichi Sakurai is a Full Professor in the Department of Informatics at Kyushu University. He directs the Laboratory for Information Technology and Multimedia Security and holds concurrent appointments with the university’s Cybersecurity Center and the Center for Quantum Computing Systems. His recent research interests include AI security in addition to cryptography and cybersecurity. Dr. Sakurai has advanced international and industry–academia–government collaboration in information security, including NICT‑supported research among Japan, China, and Korea, and the first MoU between Japan and the Cryptology Research Society of India (CRSI). He was also affiliated with the Advanced Telecommunications Research Institute International (ATR), where he contributed to NEDO–SIP projects on supply‑chain security. He has published over 500 academic papers [ https://dblp.org/pid/16/3865.html ].

Speaker (II): Prof. Chunhua Su
University of Aizu

Title: Trustworthy AI for Collaborative Cyber-Physical Ecosystems: From Federated Sensing to Secure Governance

Abstract:

The rapid convergence of artificial intelligence (AI), Internet of Things (IoT), and cyber-physical systems (CPS) is transforming modern infrastructures such as smart cities, intelligent transportation, healthcare, and industrial control systems. However, this convergence also introduces new security, privacy, and governance challenges. Ensuring trustworthiness in collaborative cyber-physical ecosystems has become a fundamental requirement. In this talk, I will present our recent research on trustworthy AI frameworks for CPS environments, covering three key layers: sensing, learning, and governance. At the sensing layer, we explore AI-enabled transparent authentication and malicious behavior detection, including biometric-based IoT authentication and GPS spoofing detection in unmanned systems. At the learning layer, we design decentralized and privacy-enhanced federated learning architectures for IoT applications, featuring ring-topology-based communication-efficient aggregation and differential privacy mechanisms for secure medical and smart-city scenarios. At the governance layer, we introduce blockchain-assisted access control mechanisms, including NFT-based decentralized authentication models for IoT ecosystems. Finally, I will outline our future directions on LLM-assisted CPS threat intelligence, adaptive federated learning in dynamic networks, and biologically inspired intrusion detection systems. The goal of this talk is to demonstrate how trustworthy AI can serve as a foundational infrastructure for secure, privacy-preserving, and resilient cyber-physical ecosystems.

Biography:

Chunhua Su is a Senior Associate Professor at the University of Aizu, Japan. He received his B.S. degree from Beijing Electronic Science and Technology Institute in 2003, and his M.S. and Ph.D. degrees in Computer Science from Kyushu University, Japan, in 2006 and 2009, respectively. From 2009 to 2011, he was a Postdoctoral Fellow at Singapore Management University. Between 2011 and 2013, he served as a Research Scientist in the Cryptography & Security Department at the Institute for Infocomm Research (I²R), Singapore. He was an Assistant Professor at JAIST from 2013 to 2016, and at Osaka University from 2016 to 2017. Since 2017, he has been with the University of Aizu, where he is currently a Senior Associate Professor. His research interests focus on trustworthy AI and cyber-physical security, including privacy-enhanced federated learning, AIoT/WiFi sensing, blockchain-assisted authentication and reputation systems, physical-layer key generation, intrusion detection for critical infrastructures, and decentralized security governance. He has published over 200 papers in international journals and conferences.