Metraipolis › archive › dc_f59ef370dec4

Using Transformers to Generate 3-Manifold Triangulations — arXiv:2611.07514

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.07514
License (as archived)
CC0-1.0
Archived
2026-11-06T14:51:12.666Z
SHA-256 of the archived text
2148509e1699b65dab78de8d4110bc64927f39288853a4a894db41704e40d55e
Stable URL of this page
https://metraipolis.com/documento/dc_f59ef370dec4

How to cite

Using Transformers to Generate 3-Manifold Triangulations — arXiv:2611.07514 — arXiv (en), https://arxiv.org/abs/2611.07514 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f59ef370dec4 (Metraipolis archive dc_f59ef370dec4, sha256 2148509e1699b65dab78de8d4110bc64927f39288853a4a894db41704e40d55e)
BibTeX
@misc{dcf59ef370dec4,
  title = {Using Transformers to Generate 3-Manifold Triangulations — arXiv:2611.07514},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.07514},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f59ef370dec4; sha256 2148509e1699b65dab78de8d4110bc64927f39288853a4a894db41704e40d55e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.07514. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Using dehydration signatures as an encoding, we train GPT-2-style transformers to generate triangulations of 3-dimensional manifolds. Our main runs each use approximately 3 exaFLOPs, spanning a 200 million parameter model trained on 60 million sphere triangulations to an 11 million parameter model trained on 1.1 billion sphere triangulations. The optimal model has 10 blocks, width 640 and generates dehydration signatures of closed manifolds of various topological types at a rate of 6.9%, 82% of which are spheres. To investigate how our method scales, we use the technique of IsoFLOP curves on 15-tetrahedron triangulations of the 3-sphere, over a compute range of 6 to 3000 petaFLOPs, and observe neural scaling behaviour. We extract a parameter scaling exponent of $\alpha = 0.49$ ($[0.48, 0.50]$ stat, $\pm 0.02$ syst), and an optimal ratio of 234 tokens per parameter for training. In addition, we present a number of smaller experiments, including evidence that transformers are able to learn to count degree-one edges near-perfectly, but get progressively worse when counting edges with higher degrees.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Geometric Topology (math.GT) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.07514 — 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 2026-11-06T14:21:42.661Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).