| title | LANTERN: Illuminating Hidden Mathematical Knowledge in Language Models |
|---|---|
| authors | Pavel Tikhonov, Elena Tutubalina, Ivan Oseledets, Dmitry I. Ignatov, Mikhail Seleznyov |
| date | 2026-09-26 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2609.32264 |
| upvotes | 33 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | Language models can now prove theorems, but people still decide which problems to pursue. We ask whether a model's internal representations can help identify promising mathematical connections. We develop LANTERN, a fast, cost-efficient pipeline that uses a classifier over pretrained-model activations to rank candidate relations, followed by staged filtering, hypothesis generation, executable verification, and analytical checking. Applied to the On-Line Encyclopedia of Integer Sequences (OEIS), LANTERN ranked 50 million pairs among 10,000 frequently referenced sequences and produced 62 verified relations between pairs without an existing OEIS cross-reference. A content screen retained 13 relations worth presenting; nine of these are informative or insightful, including four which are entirely novel to the best of our knowledge: none appears in the OEIS or in our targeted literature search. The entire end-to-end process including classifier training, candidate ranking, filtering and verification took under 8 hours. |
cite
@misc{tikhonov2026lantern,
title={LANTERN: Illuminating Hidden Mathematical Knowledge in Language Models},
author={Pavel Tikhonov and Elena Tutubalina and Ivan Oseledets and Dmitry I. Ignatov and Mikhail Seleznyov},
year={2026},
eprint={2609.32264},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2609.32264}
}