#Autoformalization
mathematics). To mitigate the inefficiency of manual formalization, we introduce a novel human-in-the-loop autoformalization pipeline that integrates: (1) specialized large language models (LLMs) for statement autoformalization, (2) multi-LLM semantic [3/7 of https://arxiv.org/abs/2505.02735v1]
May 6, 2025 at 6:02 AM Everybody can reply
ATLAS generated ~117k theorem statements and fine-tuned Llama 3.1-8B-Instruct with LoRA adapters, yielding statistically significant gains (p < 0.05). https://getnews.me/atlas-framework-advances-ai-theorem-autoformalization-with-large-dataset/ #atlas #llama31 #neurips2025
October 3, 2025 at 7:12 AM Everybody can reply
unique challenges. In this survey, we provide a comprehensive overview of recent advances in autoformalization from both mathematical and LLM-centric perspectives. We examine how autoformalization is applied across various mathematical domains and [3/5 of https://arxiv.org/abs/2505.23486v1]
May 30, 2025 at 5:58 AM Everybody can reply
WIth Lean-FIRE, we achieved the first end-to-end autoformalization for 13 Putnam problems, but also showed that conjecturing is still a massive reasoning challenge.
October 24, 2025 at 11:05 AM Everybody can reply
Next up: Real-world autoformalization by Siddhartha Gadgil. This is gonna be lit, very much looking forward to this.
researchseminars.org - View talk
Welcome to researchseminars.org, a list of research seminars and conferences!
buff.ly
January 15, 2025 at 1:01 PM Everybody can reply
1 likes
Why is this a worthwhile project?

1) It will create a hard dataset for autoformalization AI's;

2) It will force us to formalize the definitions of mathematical objects which are being used today in the top journals, thus making Lean's mathematics library more relevant to modern math researchers.
I am advertising for 4 post-docs to come to Imperial and formalize, in Lean, *statements* of theorems from recent issues of the top generalist pure mathematics journals.

www.imperial.ac.uk/jobs/search-...

Positions are for 2 years, start date 1st Oct this year. Deadline 15th August.
Description
Please note that job descriptions are not exhaustive, and you may be asked to take on additional duties that align with the key responsibilities ment...
www.imperial.ac.uk
July 28, 2025 at 12:12 PM Everybody can reply
10 likes
ProofNet: A benchmark for autoformalizing and formally proving undergraduate-level mathematics problems. ~ Zhangir A Azerbayev, Bartosz Piotrowski, Jeremy Avigad. mathai2022.github.io/papers/20.pdf #Autoformalization #ITP #LeanProver #Math
November 1, 2023 at 9:53 AM Everybody can reply
[2025-08-27] 📚 Updates in #AIMat

(1) <a href="https://researchtrend.ai/papers/2508.18914" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">FormaRL: Enhancing Autoformalization with no Labeled Data
(2) FormaRL: Enhancing Autoformalization with no Labeled Data

🔍 More at researchtrend.ai/communities/AIMat
August 27, 2025 at 3:07 AM Everybody can reply
Autoformalization performance of LLMs as measured by standard benchmarks such as ProofNet. Crucially, our approach outperforms pretrained models using a minimal number of tokens. We also show, through strategic prompting and [6/8 of https://arxiv.org/abs/2502.15795v1]
February 25, 2025 at 5:53 AM Everybody can reply
Merlin Carl
Improving the Diproche CNL through Autoformalization via Large Language Models
https://arxiv.org/abs/2303.17513
April 3, 2024 at 3:13 PM Everybody can reply
Research shows data alignment, not size, significantly influences LLM performance, especially in Autoformalization. There is a strong negative correlation between alignment and perplexity, indicating a need to adjust LLM training methodologies. https://arxiv.org/abs/2501.08496
Quantifying the Importance of Data Alignment in Downstream Model Performance
ArXiv link for Quantifying the Importance of Data Alignment in Downstream Model Performance
arxiv.org
July 4, 2025 at 7:30 AM Everybody can reply
2 likes
arXiv:2505.23486v1 Announce Type: new
Abstract: Autoformalization, the process of transforming informal mathematical propositions into verifiable formal representations, is a foundational task in automated theorem proving, offering a new perspective [1/5 of https://arxiv.org/abs/2505.23486v1]
May 30, 2025 at 5:58 AM Everybody can reply
Nilay Patel, Jeffrey Flanigan, Rahul Saha
A New Approach Towards Autoformalization. (arXiv:2310.07957v1 [cs.CL])
http://arxiv.org/abs/2310.07957
October 13, 2023 at 2:03 AM Everybody can reply
@rohanpaul_ai https://x.com/rohanpaul_ai/status/1963195517980291255 #x-rohanpaul_ai

a simple reinforcement approach that learns formal math from unlabeled text and boosts accuracy.

4x to 6x pass@1 gains using only 859 unlabeled problems.

Autoformalization means converting textbook...
September 3, 2025 at 11:15 AM Everybody can reply
We need so much more work on autoformalization (well that and practical code writing, like where the AI has to write a whole project, since I think they share many of the same difficulties like writing APIs, using libraries, and breaking down problems into sub-problems).
December 23, 2024 at 1:52 AM Everybody can reply
3 likes
Chan, Souliman, Nordhagen, Miranda, Obbad, Koyejo: Lean-ing on Quality: How High-Quality Data Beats Diverse Multilingual Data in AutoFormalization https://arxiv.org/abs/2502.15795 https://arxiv.org/pdf/2502.15795 https://arxiv.org/html/2502.15795
February 25, 2025 at 5:53 AM Everybody can reply