Prof. Yen-Wei Chen
Ritsumeikan University, Japan
 

Speech title: From Data-Centric to Knowledge-Guided Deep Learning in Advanced Biomedical Image Analysis

Abstract: Deep Learning (DL) has achieved remarkable success across various domains, particularly in computer vision and image recognition. In medical imaging, although DL models have demonstrated state-of-the-art performance, their translation into real-world clinical practice remains limited. This gap stems from two fundamental challenges: the scarcity of large-scale annotated datasets and the "black-box" nature of conventional DL models, which undermines clinical trust and interpretability. In this talk, I will present a paradigm shift from data-centric deep learning toward knowledge-guided multimodal learning—an approach that integrates clinical wisdom as an active component of the model itself, ad introduce two complementary strategies for infusing domain knowledge into deep learning:

(1) Explicit knowledge guidance, which integrates unstructured radiology reports with medical images using cross-modal attention mechanisms to enhance both accuracy and explainability.

(2) Implicit knowledge distillation, where genetic information serves as a teacher during training, transferring its predictive power to a student network that relies solely on CT images at inference time—enabling scalable, cost-effective precision diagnostics.

Biography: Yen-Wei Chen received the B.E. degree in 1985 from Kobe Univ., Kobe, Japan, the M.E. degree in 1987, and the D.E. degree in 1990, both from Osaka Univ., Osaka, Japan. He was a research fellow with the Institute for Laser Technology, Osaka, from 1991 to 1994. From Oct. 1994 to Mar. 2004, he was an associate Professor and a professor with the Department of Electrical and Electronic Engineering, Univ. of the Ryukyus, Okinawa, Japan. He is currently a professor with the college of Information Science and Engineering, Ritsumeikan University, Japan. He is the founder and the first director of Center of Advanced ICT for Medicine and Healthcare, Ritsumeikan University, Japan. Since April 2024, he has been a Foreign Fellow of the Engineering Academy of Japan.
His research interests include medical image analysis, computer vision and computational intelligence. He has published more than 300 research papers in a number of leading journals and leading conferences including IEEE Trans. Image Processing, IEEE Trans. Medical Imaging, CVPR, ICCV, MICCAI. He has received many distinguished awards including ICPR2012 Best Scientific Paper Award, 2014 JAMIT Best Paper Award. He is/was a leader of numerous national and industrial research projects.

 

Prof. Mu-Yen Chen
National Cheng Kung University, Taiwan
 

Speech title: From Model Capacity to Inductive Bias: Rethinking Deep Learning Models for Non-Stationary Financial Time Series

Abstract: Over the past decade, the evolution of deep learning for time-series forecasting has largely followed a single trajectory: increasing model size. From LSTM to Transformer and its many variants, and more recently to time-series foundation models pretrained across domains, mainstream research has implicitly assumed that greater model capacity leads to higher accuracy. Financial time series, however, do not conform to this assumption. They are highly non-stationary, have low signal-to-noise ratios and weak long-term memory, behave much like random walks, and undergo continual distribution shifts as market conditions change. What, then, does added complexity offer for such data?

This study argues that deep learning for financial time series is undergoing a paradigm shift from expanding model capacity to adopting inductive biases that match the underlying data structure. Two recent studies published in Applied Soft Computing provide the starting point. The first deliberately combines all mechanisms considered effective into a single hybrid architecture as a complexity probe, comparing it against a range of baselines under a fair benchmark with shared normalization and training protocols. The results show that instance-level distribution alignment, rather than architectural complexity, is the key determinant of accuracy, with lightweight recurrent and linear models performing better. The second uses a decomposition–residual architecture as a controlled testbed to systematically compare fusion strategies in hybrid systems. The results show that parameter-free direct addition outperforms learnable gating mechanisms. A model with just over 20,000 parameters can perform on par with the statistically leading group while substantially outperforming heavyweight attention models. Together, these studies indicate that shorter lookback windows and parsimonious parameterization yield better performance. They also show that improvements in statistical accuracy do not necessarily translate into economic value; under the efficient market hypothesis (EMH), these two dimensions must be evaluated separately.

Looking ahead, this study will discuss three research directions: (1) treating distribution-shift handling as a core model component, developing more adaptive normalization mechanisms and model designs that adapt to market conditions and transfer across settings; (2) combining the statistical advantages of large-scale pretrained foundation models with task-specific structural inductive biases within affordable computational budgets, while acknowledging the inherent predictability ceiling of financial time series; and (3) establishing a culture of fair, reproducible evaluation that incorporates economic value, allocating limited model capacity where it is most effective, and advancing trustworthy financial AI.

Biography: Dr. Chen is a Distinguished Professor of Engineering Science at National Cheng Kung University, Taiwan. He received his PhD from Information Management from National Chiao-Tung University in Taiwan. His current research interests include artificial intelligent, soft computing, data mining, deep learning, context-awareness, machine learning, and social network mining with more than 200 publications in these areas. He has co-edited several special issues in International Journals (e.g. IEEE Transactions on Engineering Management, IEEE Access, ACM Transactions on Management Information Systems, ACM Transactions on Sensor Networks, Computers in Human Behavior, Applied Soft Computing, Soft Computing, Information Fusion, Journal of Real-Time Image Processing, Sustainable Cities and Society, Neurocomputing, Supercomputing, Enterprise Information Systems, Journal of Medical and Biological Engineering, Computational Economics). He has served as Editor in Chief and Associate Editor of international journals [e.g. International Journal of Big Data and Analytics in Healthcare, IEEE Transactions on Engineering Management, IEEE Access, Applied Soft Computing, Granular Computing, Human-centric Computing and Information Sciences, Journal of Information Processing Systems, International Journal of Social and Humanistic Computing] while he is an editorial board member on several SCI journals.

 

Prof. Wanida Kanarkard
Khon Kaen University (KKU), Thailand
 

Speech title: Scaling AI-Powered Environment and Energy Innovations for a Sustainable Tomorrow

Abstract: As global climate action reaches a critical juncture, Artificial Intelligence (AI) emerges as a vital catalyst in accelerating the transition toward Net Zero and sustainable development. The talk explores the future trajectory of AI-driven green technologies, illustrating how higher education institutions act as living laboratories for scalable sustainability solutions. Through the Smart City Operation Center (SCOPC), demonstrates how university-led research leverages real-time IoT networks, spatial analytics, and emerging AI technologies ranging from predictive machine learning and digital twins to multi-agent optimization to optimize campus energy consumption, reduce carbon footprints, and automate waste management, establishing an operational blueprint for a Green University. Extending beyond campus boundaries, this session highlights applied AI research directly influencing Thai public policy and industrial practices. Key case studies showcase data-driven frameworks designed for tracking and managing industrial hazardous waste, alongside optimization models that accelerate the transformation of organic waste into bioenergy. By integrating policy intelligence with cutting-edge AI technologies, these innovations enhance circular economy strategies, improve regulatory enforcement, and accelerate to achieve decarbonization goals, build climate resilience, and advance a sustainable future.

Biography: Prof. Dr. Wanida Kanarkard graduated with second class honors in Computer Engineering from Khon Kaen University, Thailand. She further pursued her academic journey with a master’s degree in Advanced Computing from Imperial College London, UK, and Ph.D degree in Computer Engineering from the University of Hertfordshire, UK. Her academic journey was further enriched by postdoctoral research in Mobile Computing at the National Institute of Informatics (NII), Japan. Since joining Khon Kaen University in 1995, Prof. Dr. Wanida has been a driving force in integrating digital innovation with urban and environmental sustainability. As the Director of the Smart City Operation Center, she has spearheaded initiatives to transform the university into a Green Campus, focusing on energy efficiency and environmental stewardship. Her pioneering work extends to the development of smart cities across Thailand, where she leverages digital technologies to advance the circular economy, particularly in waste management—including plastic, industrial, electronic, and biomass waste—turning challenges into valuable opportunities. With numerous internationally published research papers and a few awards to her credit, reflecting her commitment to bridging technology, sustainability, and societal impact.