Sangho Lee, a researcher at HippoTNC, has been selected to deliver an Oral Presentation at AAAI 2025, one of the most prestigious conferences in artificial intelligence.

※ Oral Presentations are reserved for approximately the top 6.1% of all accepted papers.

The paper, titled “MAMS: Model-Agnostic Module Selection Framework for Video Captioning,” was co-authored with Professors Ilyong Jeon and Hogun Park from Sungkyunkwan University. It introduces MAMS, a novel framework that addresses the limitations of conventional static frame processing in video captioning by automatically selecting the optimal number of frames and captioning modules for each video.

Sangho Lee played a pivotal role in the research by designing the frame and token selection algorithms, developing an adaptive attention masking mechanism, and leading extensive performance evaluation experiments.

The proposed framework demonstrated consistent performance improvements when applied to state-of-the-art multimodal video-language models such as SwinBERT, UniVL, and mPLUG-2, and has been recognized as a meaningful advancement in multimodal AI research.

This breakthrough, which integrates vision and natural language processing (NLP) technologies, is expected to make a practical contribution to HippoTNC’s core projects—including the development of predictive disease models through integrated analysis of diverse medical data, and the implementation of personalized digital therapeutics.

Article Link : https://ojs.aaai.org/index.php/AAAI/article/view/32478