Metraipolis › archive › dc_559022ddec59

Mechanistic Interpretability of Atmospheric Rivers in GraphCast — arXiv:2610.07583

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.07583
License (as archived)
CC0-1.0
Archived
2026-10-07T11:20:52.714Z
SHA-256 of the archived text
ffa4fe0d83c01a5d6e8f7d22827c549432dee6260273b1b870e0674e4a2f330d
Stable URL of this page
https://metraipolis.com/documento/dc_559022ddec59

How to cite

Mechanistic Interpretability of Atmospheric Rivers in GraphCast — arXiv:2610.07583 — arXiv (en), https://arxiv.org/abs/2610.07583 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_559022ddec59 (Metraipolis archive dc_559022ddec59, sha256 ffa4fe0d83c01a5d6e8f7d22827c549432dee6260273b1b870e0674e4a2f330d)
BibTeX
@misc{dc559022ddec59,
  title = {Mechanistic Interpretability of Atmospheric Rivers in GraphCast — arXiv:2610.07583},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.07583},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_559022ddec59; sha256 ffa4fe0d83c01a5d6e8f7d22827c549432dee6260273b1b870e0674e4a2f330d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.07583. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

While AI weather models now rival operational forecasts, how they represent the atmosphere internally remains an open question: feature attribution reveals which input patterns matter, not what the model computes or how it combines information internally. We train sparse autoencoders (SAEs) on GraphCast to uncover its learned concepts, using atmospheric rivers as our phenomenon of focus. Both standard and Matryoshka SAEs show GraphCast computes atmospheric river intensity, measured by integrated vapor transport (IVT), as a stable internal variable, despite IVT being neither an input nor a target. In contrast to the unstructured concept retrieval of the standard SAE, the Matryoshka SAE orders concepts by importance and exposes their relations. Atmospheric river concepts persist across depth and direct interventions confirm causality. This method offers a way to find internal variables and determine which of them the model actually relies on, which is a prerequisite for asking whether those variables remain meaningful as the phenomenon changes under a warming climate.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.07583 — 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 2026-10-07T11:15:24.735Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).