Metraipolis › archive › dc_ac5e3e36628b
Small transformers track Bayesian evidence for latent common causes via a context-invariant mechanism — arXiv:2609.35161
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.35161
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T22:38:54.260Z
- SHA-256 of the archived text
4ebe1a0ced8a0a8e010f6b3d5b2b31f05e5957785342f2e516f12a421c9d499d- Stable URL of this page
https://metraipolis.com/documento/dc_ac5e3e36628b
How to cite
Small transformers track Bayesian evidence for latent common causes via a context-invariant mechanism — arXiv:2609.35161 — arXiv (en), https://arxiv.org/abs/2609.35161 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ac5e3e36628b (Metraipolis archive dc_ac5e3e36628b, sha256 4ebe1a0ced8a0a8e010f6b3d5b2b31f05e5957785342f2e516f12a421c9d499d)
BibTeX
@misc{dcac5e3e36628b,
title = {Small transformers track Bayesian evidence for latent common causes via a context-invariant mechanism — arXiv:2609.35161},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.35161},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ac5e3e36628b; sha256 4ebe1a0ced8a0a8e010f6b3d5b2b31f05e5957785342f2e516f12a421c9d499d},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.35161. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We present an in-depth investigation of how a form of Bayesian reasoning about common causes can emerge as a cross-contextual generalization in small, tractable transformers. Incrementing on recent work, our set-up (i) disentangles causal mechanisms in the model from the causal structure of the true data-generating process, (ii) orients more towards natural language prediction by considering inference of latent common causes, and (iii) considers whether and how Bayesian evidence accumulation for latent common causes can be implemented in representations and mechanisms that allow for cross-context generalization to novel test cases. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.LG). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.35161 — 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 2026-10-07T22:14:28.373Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).