How it works

NiCLIP adapts CLIP to pair fMRI activations with text embeddings. A contrastive loss pulls matching brain-text pairs together in a shared space.