GW Parameter Inference with Machine Learning

This tutorial applies simulation-based inference to gravitational wave parameter estimation. It uses simulated data (signal waveforms + noise) to train a neural network to represent the Bayesian posterior.

The tutorial works in a simplified setup, inferring just the two mass parameters for a binary black hole, so it runs in a few minutes on a laptop. The notebook should hopefully serve as a starting point for more advanced applications.

Getting started

To get started quickly, run the tutorial in Google Colab by clicking the button above. To run it locally (which may be faster), ensure that you create and activate a Python environment with the required packages. If using conda, this can be done with

conda create -c conda-forge -n gwml python=3.10 pytorch lalsuite glasflow corner numpy matplotlib jupyterlab
conda activate gwml

References

• My papers on the topic
• Dingo, a package for GW inference with neural posterior estimation
• SBI, a higher-level package for simulation-based inference