OzGrav

  • Home
  • About
    • Vision & Mission
    • Join OzGrav
    • Mental Health and Wellbeing
    • Getting started in OzGrav
    • Funding Opportunities
    • Diversity and Inclusion
    • Code of Conduct
    • OzGrav Mentoring Program
    • Nodes & Partners
    • Facilities & Capabilities
    • Reports >
      • Annual Reports
      • Industry Success Stories
      • Strategic Plan
    • Member resources
  • Our People
    • Chief Investigators
    • Partner Investigators
    • Associate Investigators
    • Postdocs and Students >
      • Faces of OzGrav
    • Professional & Outreach staff
    • Governance Advisory Committee
    • Scientific Advisory Committee
    • Executive Committee
    • Equity & Diversity Committee
    • Early Career Researcher Committee
    • Professional Development Committee
    • Research Translation Committee
    • OzGrav Alumni
  • Research Themes
    • Instrumentation
    • Data/Astro
  • Education and Outreach
  • Events
    • OzFink workshop 2023
    • 2022 OzGrav ECR Workshop and Annual Retreat
    • Upcoming and Past Events
  • News/Media
    • News
    • Newsletter
    • How to write a research brief
  • Contact Us
  • Home
  • About
    • Vision & Mission
    • Join OzGrav
    • Mental Health and Wellbeing
    • Getting started in OzGrav
    • Funding Opportunities
    • Diversity and Inclusion
    • Code of Conduct
    • OzGrav Mentoring Program
    • Nodes & Partners
    • Facilities & Capabilities
    • Reports >
      • Annual Reports
      • Industry Success Stories
      • Strategic Plan
    • Member resources
  • Our People
    • Chief Investigators
    • Partner Investigators
    • Associate Investigators
    • Postdocs and Students >
      • Faces of OzGrav
    • Professional & Outreach staff
    • Governance Advisory Committee
    • Scientific Advisory Committee
    • Executive Committee
    • Equity & Diversity Committee
    • Early Career Researcher Committee
    • Professional Development Committee
    • Research Translation Committee
    • OzGrav Alumni
  • Research Themes
    • Instrumentation
    • Data/Astro
  • Education and Outreach
  • Events
    • OzFink workshop 2023
    • 2022 OzGrav ECR Workshop and Annual Retreat
    • Upcoming and Past Events
  • News/Media
    • News
    • Newsletter
    • How to write a research brief
  • Contact Us

RESEARCH HIGHLIGHT: Extraction of binary black hole gravitational wave signals from detector data using deep learning

12/7/2021

0 Comments

 
​One of the major challenges involved in gravitational wave data analysis is accurately predicting      properties of the progenitor black hole and neutron star systems from data recorded by LIGO and Virgo. The faint gravitational wave signals are obscured against the instrumental and terrestrial noise.    
 
LIGO and Virgo use data analysis techniques that aim to minimise this noise with software that can ‘gate’ the data – removing parts of the data which are corrupted by sharp noise features, called ‘glitches’. They also use methods that extract the pure gravitational-wave signal from noise altogether. However, these techniques are usually slow and computationally intensive; they’re also potentially detrimental to multi-messenger astronomy efforts, since observation of electromagnetic counterparts of binary neutron star mergers—like short-gamma ray bursts—relies heavily on fast and accurate predictions of the sky direction and masses of the sources.
 
In our recent study, we’ve developed a deep learning model that can extract pure gravitational wave signals from detector data at faster speeds, with similar accuracy to the best conventional techniques. As opposed to traditional programming, which uses a set of instructions (or code) to perform, deep learning algorithms generate predictions by identifying patterns in data. These algorithms are realised by ‘neural networks’ – models inspired by the neurons in our brain and are ‘trained’ to generate almost accurate predictions on data almost instantly.      

Picture
Figure 1

​The deep learning architecture we designed, called a ‘denoising autoencoder’, consists of two separate neural networks: the Encoder and the Decoder. The Encoder reduces the size of the noisy input signals and generates a compressed representation, encapsulating essential features of the pure signal. The Decoder ‘learns’ to reconstruct the pure signal from the compressed feature representation. A schematic diagram of a denoising autoencoder model is shown in Figure 1.
 
For the Encoder network, we’ve included a Convolutional Neural Network (CNN) which is widely used for image classification and computer vision tasks, so it’s efficient at extracting distinctive features from data. For the Decoder network, we used a Long Short-Term Memory (LSTM) network—it learns to make future predictions from past time-series data.
 
Our CNN-LSTM model architecture successfully extracts pure gravitational wave signals from detector data for all ten binary-black hole gravitational wave signals detected by LIGO-Virgo during the first and second observation runs. It’s the first deep learning-based model to obtain > 97% match between extracted signals and ‘ground truth’ signal ‘templates’ for all these detected events.  Proven to be much faster than current techniques, our model can accurately extract a single gravitational wave signal from noise in less than a milli-second (compared to a few seconds by other methods). 
 
The data analysis group of OzGrav-UWA is now using our CNN-LSTM model with other deep learning models to predict important gravitational wave source parameters, like the sky direction and ‘chirp mass’. We’re also working on generalising the model to accurately extract single signals from low-mass black hole binaries and neutron star binaries.    
 
https://arxiv.org/abs/2105.03073
 
Paper status: Accepted.
 
Written by OzGrav researcher Chayan Chatterjee, UWA

0 Comments



Leave a Reply.

         


    OzGrav News


    Archives

    March 2023
    December 2022
    November 2022
    September 2022
    July 2022
    June 2022
    May 2022
    April 2022
    March 2022
    February 2022
    January 2022
    December 2021
    November 2021
    October 2021
    September 2021
    August 2021
    July 2021
    June 2021
    May 2021
    April 2021
    March 2021
    February 2021
    January 2021
    December 2020
    November 2020
    October 2020
    September 2020
    August 2020
    July 2020
    June 2020
    May 2020
    April 2020
    March 2020
    February 2020
    January 2020
    December 2019
    August 2019
    July 2019
    June 2019
    April 2019
    March 2019
    December 2018
    November 2018
    October 2018
    September 2018
    August 2018
    July 2018
    April 2018
    March 2018
    February 2018
    December 2017
    November 2017
    October 2017
    September 2017
    August 2017
    June 2017
    May 2017
    April 2017
    March 2017
    February 2017
    December 2016
    September 2016
    June 2016
    February 2016

    Categories

    All
    Event
    Media

      Keep up to date with ozgrav news and events

    Subscribe to Newsletter

    RSS Feed

Picture
Picture
Picture
Picture
Picture
Picture
Picture
Picture
We acknowledge and pay respects to the Elders and Traditional Owners of the land on which our six Australian nodes stand

​© 2022   The ARC Centre of Excellence for Gravitational  Wave Discovery (OzGrav)
Banner images: An artist's impression of gravitational waves generated by binary neutron stars.  Credits: R. Hurt/Caltech-JPL
Picture