This repository serves as a comprehensive reading list and resource hub for Deep Learning for Mathematical Reasoning (DL4MATH). It is targeted at researchers and practitioners in NLP, AI, and machine learning who are interested in advancing AI's capabilities in mathematical problem-solving, theorem proving, and quantitative reasoning. The primary benefit is a curated collection of papers, blogs, workshops, and benchmarks, providing a structured overview of the field.

How It Works

The repository organizes resources into categories such as Related Surveys, Blogs, Workshops, Talks, Benchmarks (Math Word Problems, Theorem Proving, Geometry Problem Solving, Math Question Answering, Other Quantitative Problems), Neural Network Architectures (General, Seq2Seq, Graph-based), Pre-trained Language Models (General, Self-supervised, Task-specific fine-tuning), and In-context Learning. This categorization allows users to navigate the landscape of DL for Math effectively.

Quick Start & Requirements

This repository is a curated list of resources and does not have a direct installation or execution command. Users are expected to access the linked papers, blogs, and websites for their specific needs.

Highlighted Details

• Extensive lists of benchmarks for various mathematical reasoning tasks, including Math Word Problems (e.g., GSM8K, MATH), Theorem Proving (e.g., MiniF2F, NaturalProofs), and Geometry Problem Solving (e.g., GEOS, UniGeo).
• Categorization of neural network architectures relevant to mathematical reasoning, from foundational models like LSTMs and Transformers to specialized Graph-to-Tree networks.
• Detailed sections on pre-trained language models and their application to mathematical reasoning, including self-supervised learning techniques and task-specific fine-tuning.
• Coverage of in-context learning strategies for LLMs in mathematical tasks, highlighting methods like Chain-of-Thought (CoT) and Program-aided Language Models (PAL).

Maintenance & Community

The project is maintained by contributors Pan Lu, Liang Qiu, Wenhao Yu, Sean Welleck, and Kai-Wei Chang. Suggestions and contributions are welcomed via email, Twitter, or GitHub issues.

Licensing & Compatibility

The repository itself does not contain code and is not subject to software licensing. The linked resources are governed by their respective licenses.

Limitations & Caveats

As a curated reading list, the repository does not provide executable code or benchmarks. The rapidly evolving nature of the DL4MATH field means that the list may not be exhaustive or perfectly up-to-date, though the inclusion of "Latest Work" sections aims to mitigate this.