Thematic Modules: Physics


Generative Adversarial Networks


A python package for gravitational-wave astrophysics

A novel Machine Learning pipeline for data preprocessing, NN training and generation of glitches. Such generated glitches will ultimately be deployed in a low latency pipeline for vetoing and denoising of the strain signal at GW interferometers like Virgo. 

Release Notes

GlitchFlow is an under development collection of jupyter notebooks for the generation of glitches in the strain channel of GW interferometers starting from auxiliary channels’ data. The different notebooks contain a novel algorithm for channel correlation characterization, channel selection and data preprocessing. The data is then used to train a few different Generative Neural Network models in order to ultimately reproduce the signal in the strain channel of the Interferometer.

Future Plans

There is no released WP module and the functionalities developed so far. A proof of concept of the Preprocessing API and of the Training and Inference subsystems, are implemented in Jupyter notebooks. The next step is to release a set of Python packages in a separate repo implementing the required functionalities and deploy the workflow via Docker containers and Airflow DAGs. The ultimate goal is to build a veto/denoise glitch subtraction algorithm to be integrated in the low latency pipeline at Virgo and, Einstein Telescope.



Target Audience
  • Expert users
  • Developers



Source Code

Created by

Ligo and Virgo Collaborations