Real Analysis

Xinyu Zhang / December 2023

Click here to view the PDF version

This project aims to enhance the problem-solving capabilities of LLMs in Real Analysis, with a specific focus on preparing them to tackle questions from the Harvard Math Entrance Exam. By fine-tuning GPT-3.5-turbo on a curated dataset of theorems, lemmas, and definitions from Real Analysis textbooks, we developed a specialized model called the “Lemma Finder.” This model serves as a mathematical assistant, offering relevant theoretical insights to support the logical steps in solving complex mathematical problems. The refined pipeline effectively complements LLMs like GPT-4, enhancing their mathematical reasoning and accuracy in Real Analysis.

My contributions to the project included dataset curation and data preprocessing. I extracted and organized Real Analysis theorems, structured them into a custom dataset, and fine-tuned GPT-3.5-turbo with OpenAI platform. This preparation was essential in equipping the model with the ability to provide contextually relevant mathematical statements, significantly improving its utility in addressing Real Analysis problems.