주최: 서울대학교 뇌영상센터(SNUBIC)

과제명: 인프라 고도화: 고경사 확산 자기공명 영상장치 도입을 통한 뇌 구조와 기능에 대한 난제 해결(RS-2024-00435727)

후원: 국가연구시설장비진흥센터(NFEC), 교육부

This presentation explores “Large Brain Models,” applying the paradigm of foundation models—characterized by scale and emergent abilities—to neuroscience. To overcome the scalability limitations of current neuroimaging research, we developed SwiFT, an end-to-end transformer model for spatiotemporal fMRI data.

Trained on a large-scale dataset, SwiFT significantly outperforms previous models on prediction tasks like age and sex, and demonstrates high adaptability through pre-training and fine-tuning. These findings suggest that Large Brain Models, coupled with high-performance computing, can become a transformative tool for accurately analyzing the brain’s spatiotemporal representations and interpreting the non-verbal world, much like LLMs have for language.