Keynote Speakers
Table of Contents
Chang D. Yoo
Thaicom Public Company Limited
Chang D. Yoo is a Professor in the School of Electrical Engineering and the Department of Computer Science at the Korea Advanced Institute of Science and Technology (KAIST). Since joining KAIST in 1999, he has been a leading contributor to the fields of Generative AI, Multimodal Foundation Models, Physical AI, Robotics, and Agentic Intelligence. His research spans diffusion-based generative models, continual learning, reinforcement learning, trustworthy AI, and embodied agents capable of interacting with complex real-world environments. Prof. Yoo has played a key role in advancing agentic and physical AI systems that bridge perception, reasoning, and autonomous action. He has served in major international leadership roles, including Local Chair of ICML 2026, and Area Chair for top-tier conferences such as ICCV, ECCV, and ICASSP. He has also been an Associate Editor for several IEEE Transactions journals and has held important administrative positions at KAIST, including Dean of International Relations and Dean of Special Projects. Prof. Yoo actively collaborates with industry and national research initiatives, including partnerships with Samsung Research and large-scale projects supported by NRF and MSIT. He is a Senior Member of IEEE and has received multiple honors for excellence in research, education, and professional service. He received his B.S. degree from the California Institute of Technology, his M.S. degree from Cornell University, and his Ph.D. degree in Electrical Engineering from the Massachusetts Institute of Technology (MIT).
The Agentic Shift: From Passive Models to Autonomous Actors
Abstract
The emergence of Agentic Artificial Intelligence marks a watershed moment in the evolution of AI—a fundamental transition from models that merely perceive and predict to autonomous agents that reason, plan, and act. Unlike traditional AI pipelines restricted by fixed supervision and static outputs, agentic AI integrates foundation models, multimodal understanding, and reinforcement learning with sophisticated tool-use capabilities. This allows systems to navigate complex tasks within dynamic, real-world environments. This shift is propelled by synergistic advances in Large Language Models (LLMs), diffusion-based generative models, and embodied intelligence. These technologies enable AI agents to interface seamlessly with human users, digital infrastructures, and physical systems. The potential applications are vast, spanning robotics, scientific discovery, healthcare, and cyber-physical automation. By offering unprecedented scalability and adaptability, Agentic AI serves as the bedrock for the next generation of Physical AI. However, this transition from "passive" to "actor" introduces significant technical and ethical risks. Autonomous decision-making complicates the landscape of reliability, alignment, and accountability. Because agentic systems operate through recursive actions, they may amplify errors or exhibit unpredictable behaviors in open-world settings, creating novel vulnerabilities in security-sensitive domains. Achieving trustworthy deployment necessitates breakthroughs in robust learning, causal reasoning, and ethical governance. This talk explores the transformative opportunities of this agentic era while critically examining the research directions required to build resilient, human-aligned, and responsible autonomous systems.