Metraipolis › archive › dc_7ec2b33dd480

Structural-Test Geometry as an Inductive Bias for Partially Observable World Models — arXiv:2610.38903

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.38903
License (as archived)
CC0-1.0
Archived
2026-11-01T02:24:55.876Z
SHA-256 of the archived text
f6a3ed523af4009bff5869644d82745ee2dea14fda1f03e46185e0aa9c538665
Stable URL of this page
https://metraipolis.com/documento/dc_7ec2b33dd480

How to cite

Structural-Test Geometry as an Inductive Bias for Partially Observable World Models — arXiv:2610.38903 — arXiv (en), https://arxiv.org/abs/2610.38903 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7ec2b33dd480 (Metraipolis archive dc_7ec2b33dd480, sha256 f6a3ed523af4009bff5869644d82745ee2dea14fda1f03e46185e0aa9c538665)
BibTeX
@misc{dc7ec2b33dd480,
  title = {Structural-Test Geometry as an Inductive Bias for Partially Observable World Models — arXiv:2610.38903},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.38903},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7ec2b33dd480; sha256 f6a3ed523af4009bff5869644d82745ee2dea14fda1f03e46185e0aa9c538665},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.38903. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Predictive world-model objectives need not organize latent histories according to distinctions required for later control. With privileged simulator access during training, should response-derived structural information be predicted, decoded, or imposed directly on latent geometry--and how do these choices compare with structured belief distillation? We define posterior-marginalized structural tests whose response distributions induce a pseudometric over histories, then train quotient-preserving (QP) world models by aligning shared latent Euclidean distances to that metric. Across a fixed six-task diagnostic suite, the largest differences appear under held-out structural variation: QP improves planning success by 26.1 points $[21.5,30.7]$ over metric decoding and 6.4 points $[3.8,9.0]$ over structured belief distillation. On GridWorld, metric decoding accurately recovers held-out structural distances ($\rho=.91$) while its native latent geometry remains more distorted than QP's (paired IQD difference $.22$ $[.20,.24]$), showing that relational recoverability and direct geometric alignment can diverge.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.38903 — 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 2026-11-01T02:12:44.140Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).