Metraipolis › archive › dc_5b412e2d3750
Metonymic Circuits for Abstract Concept Grounding in Vision Transformers — arXiv:2610.06928
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.06928
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T11:18:11.672Z
- SHA-256 of the archived text
30ec24dba2083afcd664515a88d5bc01415784261acacd7b06c1d31f0faa4c7e- Stable URL of this page
https://metraipolis.com/documento/dc_5b412e2d3750
How to cite
Metonymic Circuits for Abstract Concept Grounding in Vision Transformers — arXiv:2610.06928 — arXiv (en), https://arxiv.org/abs/2610.06928 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5b412e2d3750 (Metraipolis archive dc_5b412e2d3750, sha256 30ec24dba2083afcd664515a88d5bc01415784261acacd7b06c1d31f0faa4c7e)
BibTeX
@misc{dc5b412e2d3750,
title = {Metonymic Circuits for Abstract Concept Grounding in Vision Transformers — arXiv:2610.06928},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.06928},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5b412e2d3750; sha256 30ec24dba2083afcd664515a88d5bc01415784261acacd7b06c1d31f0faa4c7e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.06928. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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) that bridge visual signals to abstract semantics. By applying Transcoders on CLIP and DINO vision encoders, we recover intermediate features that can be associated with semantic labels for more concrete concepts, and trace their contributions in circuits underlying abstract concept recognition. Experiments on a carefully curated icon dataset reveal structured metonymic circuits, in which perceptual primitives dominate early layers and object-like anchors precede abstract targets. Images containing rendered text instead recruit a distinct perceptual-to-textual route. Causal interventions further validate that metonymic intermediates are functionally involved in grounding abstract concepts. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.06928 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T11:15:24.459Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).