Keynote Presentations
Keynote Presentations
The keynote speakers offer wide-ranging perspectives on what it takes to build AI fit for high-stakes settings, drawing on sustainable computing, reliable natural language reasoning, interpretable models, and genomic science.
Chair, Department of CSET
KC Santosh, Ph.D.
Green Computing is All What You Need!
Professor & Person-in-Charge, Division of AI
Haoran Xie, Ph.D.
Reliable Reasoning of Large Language Models via Implicit Sentiment Analysis
Senior Professor
Qiangfu Zhao, Ph.D.
Concept Generation: A Way Towards Interpretable AI
Associate Professor, Department of CS
Ke Li, Ph.D.
From Genomic Language Models to Virtual Scientists
Green Computing is All What You Need!
Abstract
How large should datasets be for machine learning scientists to start meaningful work? Can we afford to wait years to collect data, risking limited testing opportunities? This challenge mirrors what we faced during the COVID-19 pandemic, where rapid response was essential. With future epidemics likely, isn't human-in-the-loop machine learning essential for effective public healthcare? And as we aim to harness AI for good, what about its carbon footprint? Tech giants like Amazon, Meta, Google, and Tesla are amassing massive data—but should we not address the carbon emissions tied to these AI models? Now is the time to seriously explore sustainable computing with human AI. Green computing is key to achieving sustainable AI solutions that prioritize social well-being.
Biography
KC Santosh is the Chair of the Computer Systems Engineering Technology Department and Inaugural Director of the Center of Excellence at Oregon Institute of Technology. He previously served as Chair of the Department of Computer Science and Founding Director of the Applied AI Research Lab at the University of South Dakota, with additional prior experience as a research fellow at NIH and a postdoctoral researcher at LORIA Research Center, France. With over $8 million in research funding, 10 published books, and more than 250 peer-reviewed articles, he serves as associate editor for IEEE Transactions on AI and program chair for the IEEE Conference on AI. He joined NIST's AI Safety Institute Consortium in 2024 and has received the Visionary Leadership Award (2023), the Cutler Award (2021), and the President's Research Excellence Award (2019).
Concept Generation: A Way Towards Interpretable AI
Abstract
Deep learning with attention mechanisms is making AI systems powerful enough to solve many problems more efficiently than humans. However, the process for reaching results is often "invisible". Even if an LLM can provide an explanation, it can be a "hallucination". Although many hidden concepts can be generated as latent factors for inference, most do not have counterparts in human culture. It is important to capture these concepts and redefine them properly to understand AI-based decision making, enrich human knowledge, and build a harmonic human-AI co-existence society. I propose using a tree structured deep neural network (NNTree) to generate various super/sub-concepts useful for categorizing the "world". The obtained concepts can be used directly as a "knowledge base", where human experts can confirm and refine concepts to make systems more interpretable.
Biography
Qiangfu Zhao received his D. Eng. in Electronic Engineering from Tohoku University in 1988. Prof. Zhao is currently a senior professor at Zhejiang Normal University and was a tenured full professor at the University of Aizu from 1999 to 2026. He is a foreign fellow of the Engineering Academy of Japan and chairs the Technical Committee on Awareness Computing in the IEEE Systems, Man, and Cybernetics Society. He served as Vice President of the University of Aizu from 2020 to 2024 and as associate editor for IEEE Transactions on Cybernetics and IEEE SMC Magazine. His research spans system design, signal/image processing, machine learning, and evolutionary computation.
Reliable Reasoning of Large Language Models via Implicit Sentiment Analysis
Abstract
Implicit sentiment analysis (ISA) poses a significant challenge with the absence of salient cue words. It necessitates reliable reasoning to understand how sentiment is aroused. Encoder-Decoder LLMs serve as backbone models for SA applications with impressive text comprehension, while Decoder-only LLMs exhibit superior generation and in-context learning but may contain misleading information. This study proposes RVISA, a two-stage reasoning framework harnessing DO LLMs' generation ability and ED LLMs' reasoning ability to train an enhanced reasoner. We adopt three-hop reasoning prompting to furnish sentiment elements as cues and develop a verification mechanism to ensure reasoning reliability. The method achieved state-of-the-art results on benchmark datasets.
Biography
Haoran Xie received his Ph.D. degree in Computer Science from City University of Hong Kong and his Ed.D. degree in Digital Learning from the University of Bristol. He is currently a Professor and the Person-in-Charge at the Division of Artificial Intelligence, Director of LEO Dr David P. Chan Institute of Data Science, and Associate Dean of the School of Data Science, Lingnan University, Hong Kong. He served as the Editor-in-Chief of Natural Language Processing Journal, Computers & Education: X Reality, Computers & Education: Artificial Intelligence, and Artificial Intelligence in Language Education. His research interests include natural language processing, large language models, language learning, and AI in education. He has published around 500 research publications, including more than 300 journal articles. His Google Scholar citation is more than 32,000 with the h-index of 70 and i10-index of 258. He has obtained more than 30 research awards, including the World's Top 2% Scientists by Stanford University, Top 0.5% Highly Ranked Scholar, and Best Computer Science Scientists.
From Genomic Language Models to Virtual Scientists: Decoding and Designing Biological Complexity
Abstract
Genomic language models (gLMs) offer a powerful paradigm for capturing biological complexity and designing new functions. This talk presents our recent advances in gLMs, starting with PlantRNA-FM, an interpretable foundation model that decodes RNA regulatory codes across plant species. I will then introduce OmniGenome, a structure-contextualized model achieving state-of-the-art bidirectional mapping between RNA sequences and structures, facilitating generative design. To support the community, we present OmniGenBench, a comprehensive benchmarking platform for gLM research. The talk concludes with our latest progress in building "virtual plant scientists" using LLM-empowered agents and automated knowledge graphs.
Biography
Ke Li is an Associate Professor at the University of Exeter, specializing in computational and artificial intelligence for black-box multi-objective optimization and decision-making. His research develops intelligent agents to solve complex problems in natural sciences, healthcare, and engineering using evolutionary methods and large language models. He has published over 150 papers in venues including IEEE Transactions on Evolutionary Computation, NeurIPS, AAAI, ACL, and ICSE, and has been ranked among Stanford's World's Top 2% Scientists since 2020. His research is supported by funding exceeding £11.5 million, including the UKRI Future Leaders Fellowship and the Royal Society Faraday Discovery Fellowship.