Earticle

초록

영어

Recent advances in generative modeling have sparked growing interest in image generation from electroencephalography (EEG) signals. A critical yet technically challenging component of this task lies in effectively encoding EEG signals to capture semantic information corresponding to visual stimuli. In this preliminary study, we investigate the feasibility of employing CLIP (Contrastive Language–Image Pre-training), a state-of-the-art pretrained multimodal contrastive learning model, to semantically align EEG representations with image and caption feature vectors. Our analysis explores the potential of CLIP-based EEG encoding as a foundation for brain-to-image generation systems.

목차

Abstract
1. 서론
2. 방법
2.1 EEG로부터 특징을 추출하기 위한 모델
2.2 CLIP을 활용한 EEG-Image 의미적 정렬
3. 실험방법
3.1 데이터셋
3.2 실험 환경
4. 실험 결과
5. 결론
Acknowledgement
참고문헌

저자

• Gyu Seok Lee [ Dept. of AI Biomedical Engineering, Sun Moon University ]
• Jörg Stadler [ Combinatorial NeuroImaging, Leibniz Institute for Neurobiology, Germany ]
• André Brechmann [ Combinatorial NeuroImaging, Leibniz Institute for Neurobiology, Germany ]
• Wonsang You [ Dept. of AI Biomedical Engineering, Sun Moon University/AIIP Lab, Dept. of Information and Communication Engineering, Sun Moon University ] 교신저자

참고문헌