> 학술행사 > 국내학술대회
Recent progress in Physical AI has been driven largely by models: vision-language-action architectures, reinforcement learning, and large-scale demonstration data collected through teleoperation. Robots, however, inherently involve mechanical contact, and current vision-centric approaches face fundamental limitations in physical interaction.
This talk addresses contact perception as the missing element. It first examines why teleoperation-based, vision-only demonstrations fail to capture physical interaction, and then reviews tactile and force sensing that fills this gap, from fingertip sensors to whole-body robot skin. It further argues that mechanical compliance is what makes such interaction safe, and that haptic feedback should close the demonstration loop if robots are expected to learn from contact. The talk concludes with how foundation models are beginning to incorporate touch.
Bio:
Kyungseo Park is an Assistant Professor in the Department of Robotics and Mechatronics Engineering at DGIST, where he leads the Interactive Robot Lab. His research addresses tactile sensing, safe and dependable robots, whole-body multimodal perception, and contact-rich control for physical human-robot interaction. He received his B.S., M.S., and Ph.D. from KAIST, and was a postdoctoral researcher at the KIMLAB, University of Illinois Urbana-Champaign.
Image stylization aims to transform the visual appearance of an image while preserving its content and semantic information.
This tutorial introduces the evolution of image stylization from early neural style transfer to recent diffusion-based and personalized generation methods.
We will cover key concepts and methods, including content-style representation, reference-guided stylization, and personalization using diffusion models.
Finally, the tutorial will discuss practical applications and open challenges, including style evaluation, content leakage, and dataset construction.
Jae Woong Soh is an Assistant Professor in the Department of Electrical Engineering and Computer Science (EECS) at Gwangju Institute of Science and Technology (GIST).
He received the B.S. and Ph.D. degrees in Electrical and Computer Engineering from Seoul National University in 2016 and 2021, respectively.
Before joining GIST, he worked as a Staff Engineer at Samsung Research. His research interests include image processing, computer vision, and generative models.
Spatial reasoning is a critical capability for embodied agents, enabling them to understand, navigate, and interact with complex environments. In this talk, I will discuss key challenges, including frame-of-reference reasoning and long-term spatial reasoning, as well as the abilities agents need to operate robustly in partially observable environments.
Hyounghun Kim is an assistant professor in the Graduate School of Artificial Intelligence and the Department of Computer Science and Engineering at POSTECH. He earned his Ph.D. from the Department of Computer Science at UNC-Chapel Hill, advised by Prof. Mohit Bansal. His research spans natural language processing and multimodal learning with focus on large language models, conversational agents, commonsense reasoning, image/video QA, as well as embodied AI. He has served as a senior action editor, action editor/area chair, and reviewer for ACL Rolling Review and AI top conferences. His professional experience includes internships at Adobe Research and Amazon Alexa AI, and a position as a software engineer at Samsung Electronics.
유창동 교수 (KAIST)
Title: 생성형 AI 시대, 다시 묻는 통계적 학습 이론
Abs:
인공지능은 입력을 출력으로 매핑하는 함수를 연구하는 학문이다. 초기에는 사람이 규칙(Rule)을 설계하여 함수를 정의하였으나, 오늘날에는 데이터를 통해 함수를 학습하는 패러다임으로 발전하였다. 이러한 변화는 학습된 함수의 일반화 능력(Generalization Ability)을 이해하기 위해 모델 복잡도(Model Capacity)와 샘플 복잡도(Sample Complexity)를 정의하는 문제로 이어졌으며, VC Dimension, Rademacher Complexity, PAC Learning, Agnostic PAC Learning과 같은 통계적 학습이론이 그 수학적 기반을 제공하였다. 이후 함수의 표현력은 Transformer 기반의 생성형 AI와 Agentic AI를 거쳐 Physical AI로 확장되고 있으며, 여전히 모델의 복잡도와 필요한 데이터의 양을 어떻게 이해할 것인가는 핵심적인 연구 주제로 남아 있다. 본 강연에서는 통계적 학습이론의 핵심 개념을 다시 돌아보고, 생성형 AI 시대에도 여전히 유효한 질문과 새롭게 제기되는 도전 과제를 함께 살펴본다. 역사를 되돌아보는 것은 과거를 배우기 위함이 아니라, 변하지 않는 질문을 통해 현재를 이해하고 미래를 준비하기 위함이다.
Professor, School of Electrical Engineering, KAIST Adjunct Professor, Department of Computer Science, KAIST Professor, Robotics Program, KAIST Member of Tau Beta Pi, Sigma Xi Member of ISCA Senior Member of IEEE Consultant for the Korean Foundation for Advance Studies Vice President of the Institute of Electronics and Information Engineers (IEIE) Dean of the Office of International Relations, KAIST (2013-2015) Dean of the Office of Special Projects and Institutional Relations, KAIST (2011-2013) Ph.D. in Electrical Engineering from the Massachusetts Institute of Technology M.S. in Electrical Engineering from Cornell University B.S. in Engineering and Applied Science from the California Institute of Technology
