2026 Spring KAIST School of Computing Colloquium

On Mon March 09, 2026

Speaker

Leonidas J. Guibas

Title

The Space Between The Images -- Visual Learning From Relations

Abstract

In understanding or generating images or videos, visual relations play a fundamental role, reflecting basic principles that underlie the physical world. These can range from symmetries and repetitions, to groupings and compositional structures, to invariance and equivariance under various transformations, to cycle consistency. Today, almost all supervision we provide to our models is first-order and value-driven, such as specifying desired color or semantic class for pixels, expected depth for 3D points, etc. Yet the most important visual structure is encoded in binary or multiway relations between these values --- reflecting compositional scene hierarchies or respect for geometric or physical laws. In this talk we examine a number of ways that second-order, relational supervision can be provided, from being baked into the model to contrastive learning, to consistency losses for RL. We show that relation-awareness can vastly reduce the amount of training data needed and lead to superior performance across multiple applications, including classification, segmentation, reconstruction, and VQA.

Bio

Leonidas Guibas is the Paul Pigott Professor of Computer Science at Stanford University and a Principal Scientist at Google Deep Mind. He has worked in numerous areas of computer science, such as geometric algorithms, 3D computer vision and geometric deep learning, computer graphics, robotics, discrete mathematics, and biocomputation. Dr. Guibas has been elected to the US National Academy of Engineering, the US National Academy of Sciences, the American Academy of Arts and Sciences, and the Siggraph Academy, He is is an ACM Fellow, an IEEE Fellow, and has won the ACM-AAAI Allen Newell Award, the ICCV Helmholtz prize, and Siggraph's Steven Anson Coons award.

Language

English (Offline)

On Mon March 16, 2026

Speaker

Sooyon Cho (조수연)

Title

인간 시스템의 취약점, '법'으로 패치하라: 소중한 내 돈 지키는 법

Abstract

우리가 매일 사용하는 네트워크가 정교한 프로토콜 위에 돌아가듯, 현대 사회는 '법'이라는 보이지 않는 거대한 설계도 위에 구축되어 있습니다. 그러나 많은 이들이 법을 막연하고 어렵게 느끼며 무관심하게 지내다가 예상치 못한 분쟁에 휘말리거나 사기 피해를 입고 나서야 큰 충격을 받곤 합니다. 이번 강연에서는 여러분이 반드시 알아야 할 실무적인 기초 법률 지식을 쉽고 재미있게 소개하고, 각종 사례를 통하여 우리의 소중한 돈을 지키는 방법에 대하여 알아봅니다. 나아가 여러분의 전공 역량이 미래에 어떻게 범죄 수사와 자금 추적의 핵심 도구가 될 수 있을지 그 가능성까지 함께 조망해보는 시간이 될 것입니다.

Bio

조수연 교수는 KAIST 산업디자인학과와 고려대학교 법학과를 졸업하고 제45회 사법시험에 합격하였다. 17년간 판사로 근무하다가 청주지방법원 부장판사를 끝으로 법원생활을 마무리하고 2025년부터 한국외국어대학교 법학전문대학원에서 민사소송실무를 가르치고 있다.

Language

Korean (Offline)

On Mon March 23, 2026

Speaker

Dooyoung Jung (정두영)

Title

Emotional Care Using AI Conversational Agents (AI 대화형 에이전트를 활용한 정서적 돌봄)

Abstract

최근 대규모 언어모델(LLM)을 포함한 생성형 인공지능의 발전은 정신건강 서비스 전달 방식에 새로운 가능성을 제시하고 있다. 특히 심리상담 챗봇은 접근성 향상, 비용 절감, 그리고 지속적인 상호작용이 가능하다는 점에서 디지털 정신건강 서비스의 중요한 구성요소로 주목받고 있다. 그러나 실제 임상 및 연구 현장에서는 이용자 참여 저하, 치료적 관계 형성의 한계, 안전성과 윤리 문제 등 여러 도전과제가 동시에 제기되고 있다. 본 강의에서는 LLM 기반 심리상담 챗봇의 최근 발전 동향을 소개하고, 대학생 및 일반인을 대상으로 한 디지털 정신건강 개입 연구 사례를 통해 실제 활용 가능성과 한계를 논의한다. 또한 챗봇 기반 상담 시스템이 효과적인 심리개입 도구로 발전하기 위해 필요한 설계 원칙과 향후 연구 방향을 제안하고자 한다.

Bio

정두영 교수는 정신건강의학과 전문의이자 디지털 정신건강 연구자로, 현재 KAIST 디지털인문사회과학부 부교수로 재직 중이다. 서울대학교 의과대학에서 의학박사 학위를 취득하고 서울대학교병원에서 정신건강의학과 전문의 과정을 수료하였다. 이후 울산과학기술원(UNIST)에서 교수로 재직하며 대학 구성원의 정신건강 서비스 제공과 디지털 정신건강 연구를 수행하였다. 정두영 교수는 Digital Mental Health 분야에서 모바일 기반 정신건강 평가, 디지털 개입, 그리고 인공지능 기반 상담 시스템 연구를 수행하고 있으며, 모바일 앱, 가상현실, 챗봇 등을 활용한 다양한 디지털 정신건강 중재 프로그램을 개발하고 효과성을 검증해 왔다. 최근에는 생성형 AI와 대규모 언어모델을 활용한 심리상담 챗봇 및 디지털 행동개입 연구를 진행하고 있으며, 정신건강 서비스의 접근성과 확장성을 높이기 위한 기술 기반 개입 모델 개발에 관심을 두고 있다. 또한 국내외 학술지에 다수의 연구를 발표하며 디지털 정신건강 연구 분야에서 활발한 학술 활동을 이어가고 있다.

Language

Korean (Offline)

On Mon March 30, 2026

Speaker

Kazuhiro Nakadai

Title

Robot Audition in the Wild: Toward an Inclusive Society

Abstract

Robot Audition is a concept originally proposed by Nakadai and colleagues to enable robots to perceive and understand complex acoustic scenes in real-world environments where noise, reverberation, and multiple sound sources coexist. In this invited talk, I revisit the development of robot audition from the perspective of “in the wild” sensing, highlighting how auditory and multimodal perception must evolve when robots operate beyond controlled laboratory settings. The talk begins by introducing the core technologies of robot audition, including sound source localization and separation, and by discussing the fundamental challenges that arise when these techniques are deployed in real environments. Building on this foundation, I present a series of research efforts that extend robot audition toward real-world applications, such as locating humans using sound in search-and-rescue scenarios, inferring environmental and surface properties from acoustic signals, and analyzing bird songs for ecological monitoring in outdoor environments. I also discuss how the same technical principles can be extended toward human-centered interaction, particularly sign-language-based human–robot interaction, where communication relies on non-verbal and multimodal signals rather than speech alone. Although these research topics address different application domains, they are unified by a common technical direction: expanding the signals, agents, and environments that intelligent systems are designed to perceive and reason about. By framing robot audition as a foundation for multimodal perception and interaction in the wild, this talk presents a technical pathway through which such expansions can lead toward the realization of an inclusive society, enabling intelligent systems to engage not only with diverse humans, but also with challenging environments and even non-human entities.

Bio

Kazuhiro Nakadai received a B.E. in electrical engineering in 1993, an M.E. in information engineering in 1995, and a Ph.D. in electrical engineering in 2003 from the University of Tokyo. He worked at Nippon Telegraph and Telephone as a system engineer from 1995 to 1999, at the Kitano Symbiotic Systems Project, ERATO, JST as a researcher from 1999 to 2003, and at Honda Research Institute Japan, Co., Ltd. as a principal scientist from 2003 to 2022. Currently, he is a professor at the Department of Systems and Control Engineering, School of Engineering, Institute of Science Tokyo (formerly Tokyo Institute of Technology). He concurrently served as a visiting associate professor at Tokyo Institute of Technology from 2006 to 2010, a visiting professor from 2011 to 2017, and a specially appointed professor from 2017 to 2022. He also held a concurrent position as a guest professor at Waseda University from 2011 to 2018. His research interests include artificial intelligence, robotics, signal processing, computational auditory scene analysis, multimodal integration, and robot audition. He has served as an executive board member for the Japanese Society for Artificial Intelligence (JSAI) from 2015 to 2016 and from 2024 to 2025, and for the Robotics Society of Japan (RSJ) from 2017 to 2018. He is recognized as a Fellow of both the IEEE and RSJ.

Language

English (Offline)

On Mon April 06, 2026

Speaker

Seong Joon Oh (오성준)

Title

Deploying General AI in the Private World

Abstract

General-purpose AI has achieved remarkable capabilities but struggles in real-world private settings. This talk examines three barriers to deployment. 1. Human-to-machine communication. Encoding human intent into AI remains hard. Modularity and agentic architectures offer promising solutions. 2. Machine-to-human communication. Users must understand AI decisions. This drives research in explainable AI and training data attribution. 3. Privacy and security. Deployment in sensitive domains is blocked by PII leakage, adversarial attacks, and membership inference risks. Regulatory compliance adds further constraints. I present our recent work on these challenges. Future directions. AI research is shifting from model scaling to adaptation, personalisation, and agent-based interfaces. I conclude with a new direction - separating knowledge from intelligence in AI systems. Software engineering decoupled code from data 50 years ago. A similar decoupling in AI could unlock transparent, editable, and trustworthy systems.

Bio

Seong Joon Oh is an associate professor at KAIST AI. He believes trustworthiness is the last barrier to the AI productivity revolution. He researches Scalable Trustworthy AI: making AI systems reliable, explainable, and aligned with human intent at scale. He has led the Scalable Trustworthy AI (STAI) group since 2022. He moved the group from the University of Tübingen to KAIST in February 2026. He was a research scientist at NAVER AI Lab for 3.5 years. He received his PhD in computer vision and machine learning at the Max-Planck Institute for Informatics in 2018, under the supervision of Bernt Schiele and Mario Fritz, with a focus on the privacy and security implications of CV and ML (Thesis). He received the Master of Mathematics with Distinction in 2014 and Bachelor of Arts in Mathematics as a Wrangler in 2013, both at the University of Cambridge.

Language

English (Offline)

On Mon April 13, 2026

Speaker

Tae-Ho Kim (김태호)

Title

Trends in Inference Optimization and Lightweighting for Sustainable AI (지속 가능한 AI를 위한 추론 최적화·경량화 기술 트렌드)

Abstract

AI 모델의 크기는 폭발적으로 커지고 있지만, 이를 실제 디바이스에 배포하는 일은 여전히 복잡하고 어렵습니다. 학습용 칩은 단일하지만 추론 환경은 수십 가지 하드웨어로 파편화되어 있고, 그 사이에서 연산 미지원·메모리 부족·레이턴시 초과 같은 현실적인 장벽이 존재합니다. 이 강연에서는 이 격차를 메우기 위해 Nota AI가 개발한 AI 모델 최적화/경량화 플랫폼 및 솔루션에 대해 소개하고, 실증 사례를 소개합니다. 나아가 기존 Computer Vision의 한계를 VLM(Vision-Language Model)으로 극복하는 영상 분석 솔루션과, ICLR 2026에 채택된 고해상도 VLM 효율화 연구(ERGO)도 소개하며, 엣지 AI 시대의 기술적 도전과 Nota AI의 접근 방식을 공유합니다.

Bio

- 몬트리올 대학 연구원(Y.Bengio)
- KAIST 전기 및 전자공학 석사
-KAIST AI연구소 선임연구원
현) 노타 CTO

Language

Korean (Offline)

On Mon April 27, 2026

Speaker

Hwalsuk Lee (이활석)

Title

The Current and Future of the AI B2B Market

Abstract

This lecture examines how artificial intelligence is reshaping business operations, particularly in document-driven workflows. Focusing on Upstage’s industry insights, it traces AI’s evolution through four transformative stages: from augmenting traditional software to pioneering AI-native systems that redefine work processes. The session begins with an overview of the current AI landscape and Upstage’s contributions, followed by a breakdown of each evolutionary phase. It highlights AI’s progression from enhancing existing tools to becoming the core driver of autonomous workflows. Key insights include the shift in document processing from rigid template-based systems to adaptive multimodal models, and the strategic distinctions between roles, tasks, missions, and workflows in AI integration. Through case studies in industries like insurance and banking, the lecture equips students with practical understanding of AI’s impact on enterprise operations. It underscores how AI is not just optimizing processes but creating entirely new paradigms for business innovation.

Bio

- Upstage CTO (2020 ~ Present)
- Naver Clova Visual AI Lead (2017 ~ 2020)
- NCSOFT AI Center AI Lab Vision TF Researcher (2016 ~ 2017)
- Hanwha Techwin Advanced Technology Researcher, Vision Technology Group (2011 ~ 2016)
- Ph.D. in Electrical and Electronic Engineering, KAIST (2006 ~ 2011)

Language

Korean (Offline)

On Mon May 04, 2026

Speaker

Minki Hhan (한민기)

Title

Cryptography in Quantum World

Abstract

Ever since Shor’s quantum algorithm for integer factorization, researchers have warned that quantum computers could pose a serious threat to modern cryptography. Yet much of today’s digital infrastructure still relies on cryptographic systems that would be vulnerable to large-scale quantum attacks. As quantum computing continues to advance, this raises an urgent question: what should we do to prepare?

In this colloquium talk, I will introduce the basic ideas of quantum computing, discuss the current state and future prospects of quantum computers, and explain their implications for cryptography. I will also describe what we can do now, including post-quantum cryptography, and what new possibilities may emerge in the longer-term quantum future in the context of cryptography.

Bio

Minki Hhan is an Assistant Professor in the School of Computing at KAIST. His research focuses on quantum algorithms and (post-)quantum cryptography, with broader interests in theoretical computer science and computational complexity. Before joining KAIST, Dr. Hhan was a Postdoctoral Fellow at UT Austin and a QUC Fellow/Research Fellow at KIAS. He received his PhD and BSc in Mathematics from Seoul National University, and has served on program committees for conferences including Eurocrpy, Asiacrypt, and QCrypt.

Language

English (Offline)

On Mon May 11, 2026

Speaker

Insu Yun (윤인수)

Title

The Dark and Bright Side of AI in Cybersecurity

Abstract

LLM agents extend beyond simple natural language interaction; they are intelligent systems capable of autonomously executing tasks through tool usage, code execution, and interaction with external systems. Recent advancements have enabled these agents to evolve rapidly, positioning them as viable alternatives or complements to traditional software analysis techniques, including static analysis, vulnerability detection, and code security auditing. However, this evolution presents a dual-use challenge. On the one hand, LLM agents offer significant potential for automating repetitive tasks traditionally performed by security professionals, thereby improving the efficiency and speed of vulnerability identification and mitigation. On the other hand, from an adversarial perspective, these agents may also be leveraged as powerful tools, and inadequately designed systems can introduce new attack surfaces. In this presentation, we examine both the opportunities and risks associated with LLM-based security automation. Furthermore, we aim to discuss how AI-driven approaches are expected to reshape the cybersecurity landscape.

Bio

Insu Yun is an associate professor at KAIST, currently leading Hacking Lab. He is interested in system security in general, especially, binary analysis, automatic vulnerability detection, and automatic exploit generation. His work has been published to the major computer conferences such as IEEE Security & Privacy, USENIX Security, and USENIX OSDI. Particularly, his research won the best paper award from USENIX Security and OSDI in 2018, and he also won DARPA AIxCC with Team Atlanta. In addition to research, he has been participating in several hacking competitions as a hacking expert. In particular, he won Pwn2Own 2020 by compromising Apple Safari and won DEFCON CTF in 2015 and 2018, which is the world hacking competition. Prior to joining KAIST, he received his Ph.D. degree in Computer Science from Georgia Tech in 2020.

Language

English (Offline)

On Mon May 18, 2026

Speaker

Jean Oh

Title

Creative Physical AI

Abstract

Do robots need creativity? I will share my stance that they do need creativity to solve general problems and support human values. Physical AI is a type of AI that enables robots to perceive and interact with a physical world. Trendy approaches in physical AI such as Vision-Language-Action (VLA) models directly map the observations to actions where robots make decisions dominantly based on sensed information. While sensing is crucial for understanding the current physical environments, this paradigm of physical AI is fundamentally limited to support general tasks where humans see around corners and solve problems creatively based on not only what they can observe now but also various predictions of the latent spatiotemporal and social contexts. I will illustrate the examples where robots without creativity can fail to fulfill even simple goals and how we can develop physical AI for creative problem solving. If equipped with creative physical AI, can such robots promote human creativity as in creating arts? Generative AI has brought us numerous types of convenience in the digital art world. To create artifacts in the real world, creative physical AI is needed, for instance, to preserve traditional craftsmanship such as wood carving or claymation, which faces declining participation due to its labor-intensive nature. More broadly, our innovations in creative physical AI aim to encourage people to participate in more creative activities such as educational and therapeutic art sessions. I would like to invite the audience to think about how we can use technologies to promote human creativity for the next generation.

Bio

Jean H. Oh is an Associate Research Professor and Director of roBot Intelligence Group (BIG) at the Robotics Institute at Carnegie Mellon University (CMU). Jean’s research goal is to create robots that can co-exist and collaborate with humans in shared environments, continuously learning to improve themselves over time through training, exploration, and interaction. The philosophical goal of Jean’s research is to develop technologies to remind us of “what makes us human,” promoting humanity such as safety, creativity, and compassion. Jean is best known for her contributions to socially compliant navigation and interdisciplinary physical AI research between robotics and arts. Jean’s work has won numerous best paper awards at various robotics conferences such as IEEE International Conference on Robotics and Automation (ICRA) and ACM/IEEE International Conference on Human-Robot Interaction (HRI), and featured in media worldwide including New York Times, CNET, Arirang TV, ABC, CBS, and KBS. Jean received a PhD from CMU, MS from Columbia University, and BS from Yonsei University. Jean is the founder of Lavoro AI where she brings her creative, safe, and compassionate Physical AI innovations to market.

Language

English (Offline)

On Mon June 01, 2026

Speaker

Ziwei Liu

Title

From Multimodal Generative Models to Dynamic World Modeling

Abstract

Beyond the confines of flat screens, multimodal generative models are crucial to create immersive experiences in virtual reality, not only for human users but also for robotics. Virtual environments or real-world simulators, often composed of complex 3D/4D assets, significantly benefit from the accelerated creation enabled by Gen AI. In this talk, we will introduce our latest research progress on multimodal generative models for objects, avatars, scenes, motions, and ultimately dynamic world models.

Bio

Ziwei Liu is currently an Associate Professor and a Provost’s Chair in AI at Nanyang Technological University, Singapore. His research revolves around computer vision, machine learning and computer graphics. He has published extensively on top-tier conferences and journals in relevant fields, including CVPR, ICCV, ECCV, NeurIPS, ICLR, SIGGRAPH, TPAMI, TOG and Nature Machine Intelligence. He is the recipient of PAMI Mark Everingham Prize, CVPR Best Paper Award Candidate, International Congress of Basic Science Frontiers of Science Award, MIT Technology Review Innovators under 35 Asia Pacific, Asian Young Scientist Fellowship and Singapore President’s Young Scientist Award. He serves as an Area Chair of CVPR, ICCV, ECCV, NeurIPS and ICLR, as well as an Associate Editor of TPAMI and IJCV.

Language

English (Offline)

On Mon June 08, 2026

Speaker

Jung-hee Ryu (류중희)

Title

The Era of Physical AI: Challenges in East Asia (Physical AI 시대, 동아시아의 도전)

Abstract

TBD

Bio

TBD

Language

Korean (Offline)