In 2020, the author attempted to apply neural networks to math by searching for Lyapunov functions, inspired by their undergraduate research. They tried replicating Lample and Charton’s work but faced challenges. Years later, with improved programming skills and a change in job focus, they successfully fine-tuned MathBERT to determine integrability of functions. Generating random functions and integrating them symbolically was difficult. Implementing parallel Python code to speed up integrations using Sympy and overcoming processes that clogged resources were key challenges. The author eventually ran the program on Temporal to distribute workload across multiple computers for faster results.
https://carlkolon.com/2024/08/12/integration-temporal/