Keynote & Invited Talk

Abstract

This presentation introduces a methodology for developing probability-weighted commonsense knowledge networks that capture general world knowledge by assigning statistically derived weights to both entities and their relationships. We outline an automated construction approach that initially employs deep learning vision models to process visual data, creating a foundation layer of commonsense understanding. This visual knowledge base is then systematically expanded through carefully designed interactions with pre-trained language models, which are prompted to understand the initial network's context while minimizing inherent biases. By implementing chain-of-thought prompting techniques, the system enriches the knowledge structure with additional concepts and assigns probability-based linguistic qualifiers to newly generated information.

The dual-phase methodology – combining visual data mining with targeted language model enhancement – enables the automated creation of a comprehensive commonsense knowledge network, particularly rich in understanding of the physical world. We demonstrate how these probability-enhanced networks can effectively support possibilistic-based commonsense reasoning systems, bridging the gap between structured knowledge representation and human-like inferential capabilities. 

Biography

Marek Reformat received his M.Sc. degree (with honors) from the Technical University of Poznan, Poland, and his Ph.D. from the University of Manitoba, Canada. He is a Full Professor in the Department of Electrical and Computer Engineering at the University of Alberta. His research activities aim to develop methods and techniques for intelligent data modeling and analysis that lead to the translation of data into knowledge and to design systems that can imitate different aspects of human behavior. In this context, he recognizes the concepts of Computational Intelligence – with fuzzy computing and possibility theory in particular – as crucial elements for capturing relationships between data and knowledge and mimicking human reasoning about opinions and facts. He is an Associate Editor of several international journals. In addition, he has served as a general and program chair, and as a member of the program committees of numerous international conferences on Computational Intelligence and Software Engineering.

He is a past president of the North American Fuzzy Information Processing Society (NAFIPS) and the International Fuzzy Systems Association (IFSA).    

Abstract

    This talk presents a Large Language Model (LLM)-based fuzzy agent framework designed to enhance Human-Machine Interaction (HMI) in Quantum Computational Intelligence (QCI) and Artificial Intelligence (AI) learning, as well as Taiwanese-English co-learning. By integrating QCI&AI STEM experiential learning with language co-learning applications, the framework creates an adaptive and personalized learning environment using fuzzy logic-based Human Intelligence (HI) and evolutionary computation-based Machine Intelligence (MI). The framework comprises three main components: (1) the QCI&AI STEM experiential learning agent, which integrates QCI software and AI-driven models to facilitate interactive STEM learning; (2) the Trustworthy AI Dialogue Engine (TAIDE)-based Taiwanese-English co-learning agent, which utilizes the Meta AI UST model, TAIDE, and fine-tuned Open AI Whisper-Taiwanese model for real-time Taiwanese-English co-learning; and (3) the fuzzy Knowledge Graph (KG) agent, which constructs teacher’s and learners’ knowledge graphs to evaluate the students’ learning performance. Besides, an HMI-based Observation with Comfortable Intelligence (HMI-OwCI) model is proposed to create intelligent technologies that are seamlessly integrated and intuitively aligned with human needs. Additionally, two supporting agents enhance the proposed framework: the Genetic Algorithm-based Neural Network (GANN) Agent, which optimizes QCI learning models, and the LLM fine-tuned agent, which refines Taiwanese-English co-learning and generates various Whisper-Taiwanese models. Experimental results highlight that GANN adapts well in QCI&AI STEM learning, Whisper-Taiwanese models generalize effectively, and the fuzzy KG agent enables adaptive learning assessments.

Biography

    Chang-Shing Lee (IEEE Senior Member) obtained his Ph.D. degree from the Department of Computer Science and Information Engineering at National Cheng Kung University in 1998. He is currently a professor in the Department of Computer Science and Information Engineering at the National University of Tainan, Taiwan. Professor Lee served as the Director of the Computer Center from February 2006 to July 2011 and as the Dean of the Research and Development Office from January 2011 to July 2015. In 2009, he was honored as an "Outstanding Contributor to Information Education and Taiwan Academic Network for the 2009 academic year" by the Ministry of Education. During his tenure as the director of the Computer Center, he co-chaired the Taiwan Academic Network Conference (TANET 2010) and was the program chair of the International Conference on Fuzzy Systems (FUZZ-IEEE 2011). In 2017, he was awarded the MOST's FutureTech Breakthrough Award, and in 2023, he received the Excellent Award at the NSTC's Industry-Academia Project.

Abstract

We present a retrospective of our participation in the VDI Automotive Driving Challenge, focusing on how an academic autonomous driving stack becomes competition-ready under tight constraints. The talk covers our workflow from requirements and architecture to integration and track testing, highlighting robustness measures that mattered most in practice: diagnostics, safety fallbacks, parameter management, acceptance tests, and log-driven debugging. We share concrete failure modes (e.g., calibration drift, perception edge cases, integration regressions, timing issues) and the engineering changes that resolved them. In addition, we reflect on team organization and learning outcomes—interfaces, reproducibility, and the balance between research features and operational reliability. The result is a set of transferable lessons for research groups and educators building autonomous systems that must work outside the lab.

Biography

Prof. Dr. Michael Danner is a Professor for Autonomous Systems at Bochum University of Applied Sciences, focusing on artificial intelligence, robotics, machine vision, and mobile systems. His work bridges applied research and practice-oriented teaching, with an emphasis on robust perception and decision-making for real-world robotic platforms. He completed his PhD at the University of Surrey and has extensive experience in higher-education teaching and curriculum development in robotics, software engineering, and AI. In addition to academic research, he collaborates closely with industry partners—particularly in the automotive and mobility sector—on transfer projects and technology development, ranging from autonomous driving and advanced sensing to dependable software architectures and validation workflows.

Abstract

Quantum Fuzzy Federated Learning (QFFL) integrates fuzzy reasoning, quantum-inspired computation, and federated learning to support semantic-level inference with limited data exposure. However, in continual multi-task environments, QFFL remains susceptible to catastrophic forgetting and lacks a systematic mechanism for inference-time privacy protection. This paper enhances QFFL by integrating a suite of continual learning (CL) strategies, including replay-based, regularization-based, distillation-based, and lightweight architectural approaches, to improve stability under sequential and non-IID task conditions. In addition, inference-time local differential privacy (LDP) is implemented by perturbing fuzzy inference outputs prior to Quantum Federated Inference (QFI), enabling privacy-preserving semantic aggregation. Experiments conducted on the Permuted MNIST dataset with non-IID task partitions show that continual learning consistently reduces forgetting and improves robustness and generalization. Results further demonstrate that moderate privacy budgets preserve most inference utility, revealing a clear privacy–utility trade-off. These findings provide practical guidance for equipping QFFL with adaptive learning capability and inference-time privacy protection.

Biography

Chu Kiong Loo (IEEE Senior Member) is currently a professor in the Department of Artificial Intelligence, Faculty of Computer Science & Information Technology, Universiti Malaya. As an expert in Machine Intelligence and Neurorobotics, he specializes in Lifelong Machine Learning, Explainable AI, and Sustainable Machine Intelligence. He has authored over 200 high-impact papers and holds 10 patents, bridging the gap between theoretical AI and real-world applications in digital healthcare and cognitive robotics. Prof. Loo’s international accolades include the Fulbright, JSPS, Belt and Road, and Humboldt Fellowships. He was honored with the Malaysia Research Star Award (2019) and the Top Research Scientist Malaysia (2020) award for his profound impact on the scientific community.

Abstract

 Community robots serving diverse populations require safety, adaptability, and sustainability without continuous power consumption. This talk demonstrates how energy-free adaptive mechanisms enable gravity balancing through passive mechanical design, providing safer and more energy-efficient community robots. We present a progression from fixed-payload assistive devices to reconfigurable systems that adapt to variable users and multiple task configurations. Examples include lower-limb rehabilitation machines, upper-limb support devices, and workspace assistance systems. Advanced solutions such as multi-degree-of-freedom balancing using single energy elements and emerging magnetic technologies are presented. We discuss fundamental design principles, implementation approaches, and open challenges, including achieving fully automatic, position-independent adjustment.

Biography

    Dr Chin-Hsing Kuo is a Senior Lecturer in the School of Engineering at the University of Wollongong, Australia. His research focuses on the design of mechanisms and robotic systems that optimise mechanical energy interactions—including those stored in springs, weights, and magnets—to enable next-generation technologies. He currently serves as Chair of the ASME Mechanisms and Robotics Committee and Chair of the IFToMM Technical Committee for Linkages and Mechanical Controls. He has served as Associate Editor of the ASME Journal of Mechanical Design, Journal of Mechanisms and Robotics, Mechanism and Machine Theory, and IEEE Robotics and Automation Letters, and was Conference Chair of the 2021 ASME Mechanisms and Robotics Conference at IDETC/CIE and General Chair of the 6th IFToMM International Symposium on Robotics and Mechatronics (ISRM 2019). Dr Kuo is a Fellow of ASME and a former JSPS International Research Fellow.

Abstract

In this talk I will discuss how the move toward quantum and quantum-inspired AI may change the relationship between people and intelligent systems. Although quantum AI promise major gains in optimisation and problem-solving, it may also amplify existing AI challenges such as bias, opacity, privacy, inequality and digital divide. Moreover, they may also introduce new quantum-specific challenges linked to probabilistic behaviour, hardware dependence, and deeper forms of uncertainty. The key questions are: What is the new meaning of Human-Centric Systems (HcS) in quantum-AI era, and how human can understand, trust, and ethically work with such accelerated AI systems. Explainability must evolve from technical transparency to practical support for human decision-making, and accountability and fairness must be reconsidered in systems that are inherently uncertain. Drawing on fuzzy logic and human-like reasoning, I will also discuss how a so-called Quantum-Fuzzy approach can help bridging the gap between quantum AI and humanised computing.    

Biography

Dr Amir Pourabdollah (PhD, FHEA, MSc, MEng, BEng) joined NTU in 2017, and currently leads the modules of Artificial Intelligence (final year BSc) and Data Analytics (2nd year BSc). Dr. Pourabdollah is a member of the Computational Intelligence and Applications Research Group (CIA). His active areas of research include Computational Intelligence, Quantum Intelligence, Fuzzy Logic Systems, Deep Learning, Cloud-based AI, and Ambient Intelligence. Beyond his teaching and research, he serves as the Chair of the IEEE Task Force on Quantum Computational Intelligence, an Editor for Springer's Journal of Quantum Intelligence, and a member of the IEEE CIS Standards Committee and Education Subcommittee. As a Microsoft Certified Educator and Microsoft Program Advisor at NTU, he facilitates student certifications in AI, Cloud Computing, and Cybersecurity. Furthermore, he serves as the Lead External Subject Examiner for AI courses at Northumbria University and De Montfort University, and supervises PhD research in Quantum Intelligence and Distributed Fuzzy Systems.  

Abstract

This talk explores the development of “Reflection Science,” a research program that conceptualizes reflection as an internal action that transforms one’s own cognition and aims to computationally model the cognitive dynamics induced by reflection. Grounded primarily in the Free Energy Principle, Reflection Science focuses not only on an agent’s material interactions with the external world, but also on the dynamics by which an agent—driven into an unstable free-energy state by the gap (conflict) between ideals and reality—temporarily suspends overt action toward the external world and seeks reduction expected free energy by interpreting, reappraising, and reframing one’s construal of the situation. Within this framework, I aim to provide a unified computational perspective on contrasting outcomes of reflection dynamics, such as discovering creative solutions on the one hand, and, on the other, becoming entrenched in maladaptive beliefs through repetitive, rumination-like reflection. 

Biography

Yosuke Fukuchi received the Ph.D. degree in engineering from Keio University, Japan, in 2022. He was a Project Researcher with the National Institute of Informatics, Tokyo, Japan, from April 2022 to March 2024. He is currently an Assistant Professor with the Faculty of Systems Design, Tokyo Metropolitan University. His research interests include human-agent interaction, artificial intelligence, and computational cognitive science. He is a member of Japanese Society for Artificial Intelligence.

Abstract

　Cognitive function encompasses several components, including memory, attention, working memory, executive function, visuospatial attention, and language. These components are processed hierarchically and in parallel across various situations, enabling us to derive optimal solutions. Cognitive dysfunction after stroke influences motor recovery and motor learning. It is important to identify which component of cognitive function is impaired and to provide specific rehabilitation. In this seminar, I will talk about the relationship between cognitive and motor function based on the cognitive assessments I am developing and our findings to date.

Biography

　Katsuya Sakai received the Ph.D. degree in Physical Therapy from Tokyo Metropolitan University in 2021. He worked as a research fellow at Keio University in 2018, and then became an assistant professor at Chiba Prefectural University of Health Sciences. He subsequently joined the Department of Physical Therapy at Tokyo Metropolitan University as an assistant professor in 2023. His research focuses on cognitive and motor function after stroke, and neurophysiological effects during cognitive task performance.

Abstract

　This presentation introduces case studies of shared haptic experiences for pre-school children and people with disabilities. Touch is one of the most primitive forms of human perception. Through these cases, we explore how haptic interactions can be shared, negotiated, and experienced together.

Biography

　Mina Shibasaki is an Assistant Professor at Tokyo Metropolitan University. Her research focuses on interaction design using haptic technology, particularly for children and people with disabilities. She received her B.A. from Joshibi University of Art and Design, her M.A. from Keio University Graduate School of Media Design, and her Ph.D. from Keio University in 2022.

Abstract

　This talk introduces exploratory HCI studies that aim to support physical activity for manual wheelchair users, who often face limited exercise opportunities due to physical and environmental constraints. I will present two prototype “wheelchair instruments”: a color-sensing system that converts floor colors into sound, and a motion-sensing trial that investigates outdoor mobility and engagement through barrier-information capture and location-linked interactive content. I will also discuss a visualization approach for wheelchair sprint training that combines video-based observation with sensor data graphs to examine propulsion technique and facilitate coach–user discussion.

Biography

　Xu Han received the Ph.D. degree (Art Engineering) from Tokyo Metropolitan University, Japan, in 2021. He was a Project Assistant Professor at Tokyo Metropolitan University. He is currently a Lecturer at the Professional University of Information and Management for Innovation (iU), Japan. His research interests include assistive technologies for wheelchair users, sensor-based motion analysis, and interactive feedback using animation and sound.