Learning Mathematics Through the Machine

Can machines, built on math, be taught to discover math itself?

This question has ignited a flurry of exciting research in AI and mathematical reasoning. In this project, we take a classical statistical-learning approach to show that bare, simple, and accessible algorithms, running on ordinary computers, can rediscover classical formulas such as the area of a circle, the laws of logarithms, and even the prime number theorem.

These rediscoveries aren’t the result of rules being preprogrammed or learned from text (as in large language models). Rather, the methods uncover patterns directly from data. Now imagine that same algorithm stumbling upon a brand-new identity, a hidden symmetry, or a surprising relation that no one has ever written down. That is the thrill this project will chase.

We’re looking for students with a basic background in differential equations and applied linear algebra, as well as some programming experience in Python, MATLAB, or C. Familiarity with numerical methods or elementary statistics is a plus, but not required.