Kernel Equation Learning: KEqL

JAX implementation of

'Data-Efficient Kernel Methods for Learning Differential Equations and Their Solution Operators: Algorithms and Error Analysis' by Jalalian, Osorio, Hsu, Hosseini and Owhadi. See preprint on arXiv here.

Conda Environments

We use two separate conda environments due to compatibility issues between the kernel-based implementation and the neural-based implementation.

• keql_env: Main environment to run the kernel methods for equation learning. Additionally, you can install the tools inside keql_tools by running the command in the terminal pip install -e /path/to/keql_tools.

• pinnsr_env: Environment used to run code from the PINN-based method that we refer to as PINN-SR.

Usage

• Reproduce examples: See implementation under examples/.

• Learn a new PDE: Use source code implemented in keql_tools/.