Metonymic Circuits for Abstract Concept Grounding in Vision Transformers
Uncovering how Vision Transformers map abstract concepts through concrete visual proxies
We study how Vision Transformers ground abstract concepts (e.g., angry) when training data provide limited direct referential evidence. We hypothesize a metonymic grounding mechanism in which abstract predictions are driven by concrete, interpretable anchor concepts (e.g., fire) ...
ArXiv CS.AI
Source
Generated insight by OnAirToday AI
This research reveals how Vision Transformers use concrete visual proxies to infer complex abstract emotions, effectively creating a bridge between raw pixels and human-like conceptual understanding. By identifying these metonymic circuits, researchers can significantly improve the interpretability and reliability of black-box AI vision models.

Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.