EmoMind: Decoding Affective Captions from Human Brain fMRI
Researchers have developed EmoMind, a novel system capable of generating affective captions directly from fMRI brain activity. Unlike previous methods that focus on semantic content or use discrete emotion labels, EmoMind decodes a continuous 34-dimensional emotion vector from fMRI signals. This vector is then used to rewrite semantically grounded scene descriptions, allowing for a controlled balance between factual accuracy and emotional expression. EmoMind significantly outperforms existing methods, including GPT-4 prompted with emotion labels, in generating person-specific affective captions. AI
IMPACT This research opens new avenues for personalized AI-generated content and understanding individual affective states through brain activity.