Transformers generate 3-manifold triangulations using 3 exaFLOPs

Researchers have utilized GPT-2 style transformers to generate triangulations of 3-dimensional manifolds by encoding dehydration signatures. Experiments involved models with up to 200 million parameters and 3 exaFLOPs of compute, trained on millions of sphere triangulations. The optimal model achieved a generation rate of 6.9% for closed manifolds, with 82% of these being spheres. Further analysis using IsoFLOP curves on 3-sphere triangulations revealed a parameter scaling exponent of approximately 0.49 and an optimal token-to-parameter ratio of 234. AI

IMPACT Demonstrates potential for large language models in complex mathematical and geometric problem-solving.

RANK_REASON Academic paper detailing a novel application of transformers to geometric topology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• 3-manifold
• 3-sphere
• arXiv
• GPT-2
• Hugging Face
• IsoFLOP

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →