Ideas move
through people.

Yexin Liu is a postgraduate researcher at the Hong Kong University of Science and Technology (HKUST), working with Harry Yang. His research spans image and video generation, multimodal models, and computer vision.

His work includes OpenSubject, on subject-driven image generation and manipulation, and AlignVid, on improving semantic fidelity in text-guided image-to-video generation. He has also contributed to research on multimodal model understanding and efficient video diffusion models.

Taehwan Kim is an Associate Professor in the Graduate School of Artificial Intelligence and the Department of Computer Science and Engineering at Ulsan National Institute of Science and Technology (UNIST), Korea. He leads the Interactive Multimodal Machine Learning Lab, working on multimodal learning, generative AI, and interactive AI, including vision-language models, embodied AI, and video understanding.

He received his Ph.D. in Computer Science from the Toyota Technological Institute at Chicago in 2015. Before joining UNIST, he held research positions at Caltech, ObEN, and Amazon Alexa AI.

Yaofu Liu is a Ph.D. student at the Hong Kong University of Science and Technology (HKUST), working with Harry Yang. His research focuses on efficient generative AI, especially accelerating and compressing large video diffusion models for practical inference.

His recent work includes Model Reveals What to Cache (ICCV 2025), which studies adaptive feature reuse for video diffusion models, and RotateAttention (2026), which develops INT4 quantized attention for efficient video generation.

Presentation

Talk title to be announced.

Yu Chen is a designer, entrepreneur, and founder & CEO of Artisk.ai, an AI-native branding platform. He previously studied at the Maryland Institute College of Art and worked with leading design studios including Pentagram and Sagmeister & Walsh.

His design work has received more than 100 international awards, appeared in over 50 professional books, and been exhibited in more than 30 countries. At Artisk.ai, he is exploring intelligent, adaptive brand systems that respond to changing markets, trends, and user behavior.

Presentation

Talk title to be announced.

Jiajun Zha is a Ph.D. student at the Hong Kong University of Science and Technology (HKUST). His research includes generative modelling and efficient generation, with work on accelerating diffusion models and invertible attention mechanisms.

He is a co-author of Continuous Semi-Implicit Models (ICML 2025), which introduces a continuous framework for efficient training and multistep distillation of generative models, and Invertible Attention (2021).

Presentation

Fast Generative Modelling or World Modelling; final topic to be confirmed.

Guijia Zhang researches agent security, multimodal reasoning, and reliable AI systems. He holds a B.Eng. in Electronic Information Engineering from Shenzhen University. His research experience includes work with HKUST, KAUST, and Tsinghua University, as well as an algorithm evaluation internship at DeepWisdom (MetaGPT).

His recent work studies how AI agents use visual evidence, audit tool and skill invocations, and preserve safe behavior across state changes. He has collaborated with Harry Yang on multimodal agent reliability and security.

Selected recent papers

Aborted but Not Forgotten: KV-Cache Retention Breaks Rollback Consistency in Language Agents (2026 preprint). Studies how retained inference state can undermine an agent’s rollback.

Do GUI Agents Believe Their Eyes? Diagnosing State-Belief Reliance on Pixels versus Structure (2026 preprint). Examines how GUI agents resolve conflicts between screenshots and structured interface data.

Hallucination as Exploit: Evidence-Carrying Multimodal Agents (2026). Requires verified evidence before authorizing consequential agent actions.

STARS: Skill-Triggered Audit for Request-Conditioned Invocation Safety in Agent Systems (2026). Audits skill invocations using the user request and runtime context.

Presentation

Talk title and abstract to be announced.

Tae-Hyun Oh is an Associate Professor in the School of Computing at KAIST, where he leads the Advanced Machine Intelligence Lab. His research focuses on machine learning and its applications in computer vision and graphics, including audio-visual and vision-language understanding, 3D scene understanding, and embodied AI.

Before joining KAIST in 2025, he was a faculty member at POSTECH and held research positions at Facebook AI Research and MIT CSAIL. He received his Ph.D. in Electrical Engineering from KAIST in 2017.

Presentation

Talk title and abstract to be announced.

Sunil Manghani is Professor of Theory, Practice & Critique at Winchester School of Art, University of Southampton. His research spans AI design, image studies, and critical theory, connecting visual arts practice with interdisciplinary inquiry.

He is Editor of Journal of Visual Art Practice and Managing Editor of Theory, Culture & Society. His books include Image Studies (2013), Zero Degree Seeing (2019), and Rhythm and Critique (2020). His curatorial work includes exhibitions at the John Hansard Gallery and Tate Exchange, Tate Modern.

Xian Xu (徐嫻) is an Assistant Professor in the Department of Digital Arts and Creative Industries at Lingnan University. Her research and creative practice span AI for creativity, data-driven storytelling, human–computer interaction, and immersive arts education using virtual reality.

Before joining Lingnan, she was a Research Assistant Professor jointly appointed in HKUST’s Division of Arts and Machine Creativity and Division of Emerging Interdisciplinary Areas. She received her Ph.D. in Computational Media and Arts from HKUST under the supervision of Huamin Qu, and holds an MFA and a BA from the Central Academy of Drama.

Her work on learning film lighting in virtual reality received a Best Paper Honorable Mention at IEEE VR 2025. Her current projects explore human–AI co-creation in art, AI cinematic video, and virtual production for data storytelling.

Presentation

Talk title and abstract to be announced.