Application of the Random Feature Method for Operator Learning to Tumor Growth Models

This repository contains an implementation for the numerical solution of the Cahn-Hilliard initial-condition-to-solution operator for early timesteps. The Cahn-Hilliard equation can be seen as the prototypical model for tumor growth [Fritz2023].

Project Structure

• generate_data.py for genereting a high resolution dataset of input-output pairs that can be used for supervised learning
• generate_grf_g.py for generating a large amount of parameters $\theta \in \Theta$
• RFM_CahnHilliardy.py containing the implementation of the random feature method for the Cahn-Hilliard equation.
• run_*.py-scripts for the individual experiments conducted

Dependencies

The dependencies are not explicitly listed in a requirements.txt file, but the following Python packages are required for numerical simulations:

• numpy

• scipy

• jax

• multiprocessing

• joblib

• matplotlib

• Optional: jupyter (for interactive notebooks)