Machine Learning in the LIGO-Virgo Era

Document #:

LIGO-G1800074-v1

Document type:

G - Presentations (eg Graphics)

Abstract:

As LIGO-Virgo moved from the first to the second observation run, 2015-17, the rapid maturation of Machine Learning (ML) algorithms industry-wide has enabled an increasing number of researchers to engage in a diversity of applied ML projects at the LIGO Scientific and Virgo Collaborations. Furthermore, multiple detection events have enabled a transition from simulated signals to a more robust landscape of real data analysis and note-worthy results.

Currently several areas of ML research are being pursued by LV researchers, including: a means to both classify and locate the source of transient artifacts known as glitches; tested localization of desired signal as produced by coalescing binary black holes and neutron stars; as a single-detector case for supernovae; and as a potential, future means to lock an interferometer. The algorithms employed include Random Forest, Genetic Programming, Convolutional Neural Networks, RNN Auto-encoders, Deep Filtering and Deep Regression.

This talk will provide a comprehensive overview of the diverse applications of ML in the LIGO Scientific and Virgo Collaborations, the opportunity for engaging citizen scientists, and a deeper discussion of the application of Genetic Programming to understand the origin of mechanical couplings in the LIGO detector.

Files in Document:

• final version of the Abstract - 2018 01/15 (APS_April_abstract-STAATS-20180115.pdf, 41.6 kB)

Topics:

• Public Talk / Colloquium
• Data analysis software

Authors:

• Kai Staats

Keywords:

APS machine learning

Notes and Changes:

This is the final version, approved by Marco Cavaglia, 2018 01/15