Table of Contents

Author(s)

How to Obtain Support

General Information

Software and Tools

Needed datasets

Short Description of the Use Case

Gravitational waves (gw) are 'ripples' in space-time caused by some of the most violent and energetic processes in the Universe. Albert Einstein predicted the existence of gravitational waves in 1916 in his general theory of relativity. Einstein's mathematics showed that massive accelerating objects (such as neutron stars or black holes orbiting each other) would disrupt space-time in such a way that 'waves' of undulating space-time would propagate in all directions away from the source. These cosmic ripples would travel at the speed of light, carrying with them information about their origins, as well as clues to the nature of gravity itself.

...

As described above, as the lengths of the arms change, so too does the distance traveled by each laser beam. A beam in a shorter arm will return to the beam splitter before the beam in a longer arm, then the situation switches as the arms oscillate between being longer and shorter. Arriving at different times, the waves of light no longer meet up nicely when recombined at the beam splitter. Instead, they shift in and out of alignment or "phase" as they merge while the wave is causing the arm lengths to oscillate. In simple terms, this results in a flicker of light emerging from the interferometer.

Interferometers act as a pass band on the signal, having a different rensponse to different frequencies, for this reason signal need to be corrected (with frequencies dependent weight), we call it "whitening" step.

Once whitened it is time to search for signal, we generally apply a matched filter between the signal and a "template" (a simulated event).  

In this tutorial, we will follow another approach, instead of simulated an event and searching for it inside a signal we will answer a different question: does this acquisition contains different information compared to a signal with a gw inside it? We will usie a machine learning algorithm. 

This is in fact  a simple example on how use an autoenconder to efficient data codings and foldings

...

In the following we will use the autoencoder to analyse a gravitation wave format file and learn how to ignore some sources of noise, obtain a signal cleaned and a size reductioned

All data files used for this exercise are public and can be obtained from the Ligo website at https://www.gw-openscience.org/archive/O2_16KHZ_R1/ in the gwf format or hdf5 format

At https://www.gw-openscience.org/ you could find many interesting tutorial on how read plot and analyze with standard technique the gravitational files 

How to execute it

Download data files (any files at https://www.gw-openscience.org/data/ will be good) and execute the Jupyter notebook (https://github.com/luca-rei/ml-genoa). For convenience in the Jupyter we assume to work with hdf5 files,the interesting part is how the output of an encoded signal (gw) differ from an encoded noise (compare their size and their entropy). For example try to encode different data...

Annotated Description

References

https://blog.keras.io/building-autoencoders-in-keras.html

https://keras.io/api/models/

Attachments