Nankai University International E-Forum on Artificial Intelligence and Robotics

（第65期）

2025年南开大学人工智能与机器人国际学术讲坛

College of Artificial Intelligence, Nankai University

报告时间：2025年11月25日（周二）13:00-14:00

报告嘉宾：Chanho Eom 韩国中央大学

飞书会议：987406154

报告题目：Toward Robust and Structured Representation Learning for Real-world Vision Tasks

报告摘要：

This talk explores recent advances in vision-language learning, particularly focusing on how to improve alignment between complex visual inputs and lengthy, detailed text descriptions. While large-scale models like CLIP have demonstrated impressive capabilities in connecting images and text, they often fall short when handling paragraph-level language due to their reliance on short, global associations. To address this, recent research has explored hybrid strategies that combine global and local alignment mechanisms, identifying fine-grained correspondences between visual regions and semantic units within the text. This allows for more structured and representative embeddings, which are especially effective in retrieval and grounding tasks that demand nuanced understanding. The talk will highlight key ideas behind local alignment, token-level learning objectives, and their integration into scalable fine-tuning pipelines for multimodal models.

报告人简介：

Chanho Eom is an Assistant Professor at the Graduate School of Advanced Imaging Science, Multimedia & Film at Chung-Ang University, where he leads the Perceptual AI Lab. Prior to his faculty appointment, he worked at Samsung Advanced Institute of Technology (SAIT) and NAVER CLOVA. He earned his Ph.D. from Yonsei University under the supervision of Prof. Bumsub Ham. His research lies at the intersection of computer vision and representation learning, with a particular focus on person re-identification, vision-language alignment, and generative modeling for video and image understanding. His works have been published in top-tier venues including CVPR, ICCV, ECCV, NeurIPS, and IEEE TPAMI. In addition to academic research, he actively participates in multiple government and industry-funded AI projects.