A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Yuqing Yang, Yan Ma, Pengfei Liu
Proposes a progressive weak-to-strong reasoning framework where a strong model refines its own training data without human or advanced model input, significantly improving reasoning on GSM8K and MATH.
Tony Feng, Trieu H. Trinh, G. Bingham, et al.
A case study using Gemini to semi-autonomously evaluate 700 open conjectures from Bloom's Erdős Problems database, finding obscurity rather than difficulty explains their open status.
Bernardino Romera‐Paredes, Mohammadamin Barekatain, Alexander Novikov, et al.
FunSearch pairs a pretrained LLM with an evolutionary evaluator to discover new mathematical constructions and heuristics, surpassing best-known results in extremal combinatorics and online bin packing.
Yibo Yan, Jiamin Su, Jianxiang He, et al.
First comprehensive survey of mathematical reasoning in multimodal LLMs, reviewing over 200 studies since 2021 across benchmarks, methodologies, and challenges.
Yiping Wang, Qing Yang, Zhiyuan Zeng, et al.
Reinforcement learning with verifiable reward using a single training example dramatically improves LLM math reasoning, matching or exceeding results from thousands of examples.
Adrian de Wynter
This paper argues that in-context learning fits the mathematical definition of learning but empirically shows limited generalization to unseen tasks, with accuracy insensitive to exemplar distribution and prompt style.
Thang Luong, Dawsen Hwang, Hoang Nguyen, et al.
Introduces IMO-Bench, a suite of Olympiad-level benchmarks for robust mathematical reasoning, enabling gold-level IMO performance via Gemini Deep Think.
Unknown
Integrates LLMs with Monte Carlo Tree Search to iteratively refine mathematical reasoning via self-evaluation and backpropagation.
Unknown
Trains an LLM with RL using a verifiable reward on just one example, matching performance of larger datasets in math reasoning.
Unknown
SPHERE is a self-evolving data generation pipeline that enhances mathematical reasoning in small language models by iteratively generating, correcting, and diversifying reasoning chains.
Unknown
Logic-RL uses rule-based RL on synthetic logic puzzles to teach LLMs advanced reasoning that generalizes to math benchmarks.
Unknown
Small language models rival large models in math reasoning via self-evolved deep thinking with Monte Carlo Tree Search and process preference models.