AI Enhances Gravitational Wave Detection at LIGO Observatory

Machine learning approach can enhance observatory's hunt for gravitational waves

Finding patterns and reducing noise in large, complex datasets generated by the gravitational wave-detecting LIGO facility just got easier, thanks to the work of scientists at the University of California, Riverside. The UCR researchers presented a paper at a recent IEEE big-data workshop, demonstrating a new, unsupervised machine learning approach to find new patterns in the auxiliary channel data of the Laser Interferometer Gravitational-Wave Observatory, or LIGO. The technology is also potentially applicable to large scale particle accelerator experiments and large complex industrial systems. LIGO is a facility that detects gravitational waves -- transient disturbances in the fabric of spacetime itself, generated by the acceleration of massive bodies. It was the first to detect such waves from merging black holes, confirming a key part of Einstein's Theory of Relativity. LIGO has two widely-separated 4-km-long interferometers -- in Hanford, Washington, and Livingston, Louisiana -- that work together to detect gravitational waves by employing high-power laser beams. The discoveries these detectors make offer a new way to observe the universe and address questions about the nature of black holes, cosmology, and the densest states of matter in the universe. Each of the two LIGO detectors records thousands of data streams, or channels, which make up the output of environmental sensors located at the detector sites. "The machine learning approach we developed in close collaboration with LIGO commissioners and stakeholders identifies patterns in data entirely on its own," said Jonathan Richardson, an assistant professor of physics and astronomy who leads the UCR LIGO group. "We find that it recovers the environmental 'states' known to the operators at the LIGO detector sites extremely well, with no human input at all. This opens the door to a powerful new experimental tool we can use to help localize noise couplings and directly guide future improvements to the detectors." Richardson explained that the LIGO detectors are extremely sensitive to any type of external disturbance. Ground motion and any type of vibrational motion -- from the wind to ocean waves striking the coast of Greenland or the Pacific -- can affect the sensitivity of the experiment and the data quality, resulting in "glitches" or periods of increased noise bursts, he said. "Monitoring the environmental conditions is continuously done at the sites," he said. "LIGO has more than 100,000 auxiliary channels with seismometers and accelerometers sensing the environment where the interferometers are located. The tool we developed can identify different environmental states of interest, such as earthquakes, microseisms, and anthropogenic noise, across a number of carefully selected and curated sensing channels." Vagelis Papalexakis, an associate professor of computer science and engineering who holds the Ross Family Chair in Computer Science, presented the team's paper, titled "Multivariate Time Series Clustering for Environmental State Characterization of Ground-Based Gravitational-Wave Detectors," at the IEEE's 5th International Workshop on Big Data & AI Tools, Models, and Use Cases for Innovative Scientific Discovery that took place last month in Washington, D.C. The work is published on the arXiv preprint server. "The way our machine learning approach works is that we take a model tasked with identifying patterns in a dataset and we let the model find patterns on its own," Papalexakis said. "The tool was able to identify the same patterns that very closely correspond to the physically meaningful environmental states that are already known to human operators and commissioners at the LIGO sites." Papalexakis added that the team had worked with the LIGO Scientific Collaboration to secure the release of a very large dataset that pertains to the analysis reported in the research paper. This data release allows the research community to not only validate the team's results but also develop new algorithms that seek to identify patterns in the data. "We have identified a fascinating link between external environmental noise and the presence of certain types of glitches that corrupt the quality of the data," Papalexakis said. "This discovery has the potential to help eliminate or prevent the occurrence of such noise." The team organized and worked through all the LIGO channels for about a year. Richardson noted that the data release was a major undertaking. "Our team spearheaded this release on behalf of the whole LIGO Scientific Collaboration, which has about 3,200 members," he said. "This is the first of these particular types of datasets and we think it's going to have a large impact in the machine learning and the computer science community." Richardson explained that the tool the team developed can take information from signals from numerous heterogeneous sensors that are measuring different disturbances around the LIGO sites. The tool can distill the information into a single state, he said, that can then be used to search for time series associations of when noise problems occurred in the LIGO detectors and correlate them with the sites' environmental states at those times. "If you can identify the patterns, you can make physical changes to the detector -- replace components, for example," he said. "The hope is that our tool can shed light on physical noise coupling pathways that allow for actionable experimental changes to be made to the LIGO detectors. Our long-term goal is for this tool to be used to detect new associations and new forms of environmental states associated with unknown noise problems in the interferometers." Pooyan Goodarzi, a doctoral student working with Richardson and a co-author on the paper, emphasized the importance of releasing the dataset publicly. "Typically, such data tend to be proprietary," he said. "We managed, nonetheless, to release a large-scale dataset that we hope results in more interdisciplinary research in data science and machine learning." Richardson, Papalexakis, and Goodarzi were joined in the research by Rutuja Gurav, a doctoral student working with Papalexakis; Isaac Kelly, a summer undergraduate REU student; Anamaria Effler of the LIGO Livingston Observatory; and Barry Barish, a UCR distinguished professor in physics and astronomy.

New diagnostic tool will help LIGO hunt gravitational waves

The UCR researchers presented a paper at a recent IEEE big-data workshop, demonstrating a new, unsupervised machine learning approach to find new patterns in the auxiliary channel data of the Laser Interferometer Gravitational-Wave Observatory, or LIGO. The technology is also potentially applicable to large scale particle accelerator experiments and large complex industrial systems. LIGO is a facility that detects gravitational waves -- transient disturbances in the fabric of spacetime itself, generated by the acceleration of massive bodies. It was the first to detect such waves from merging black holes, confirming a key part of Einstein's Theory of Relativity. LIGO has two widely-separated 4-km-long interferometers -- in Hanford, Washington, and Livingston, Louisiana -- that work together to detect gravitational waves by employing high-power laser beams. The discoveries these detectors make offer a new way to observe the universe and address questions about the nature of black holes, cosmology, and the densest states of matter in the universe. Each of the two LIGO detectors records thousands of data streams, or channels, which make up the output of environmental sensors located at the detector sites. "The machine learning approach we developed in close collaboration with LIGO commissioners and stakeholders identifies patterns in data entirely on its own," said Jonathan Richardson, an assistant professor of physics and astronomy who leads the UCR LIGO group. "We find that it recovers the environmental 'states' known to the operators at the LIGO detector sites extremely well, with no human input at all. This opens the door to a powerful new experimental tool we can use to help localize noise couplings and directly guide future improvements to the detectors." Richardson explained that the LIGO detectors are extremely sensitive to any type of external disturbance. Ground motion and any type of vibrational motion -- from the wind to ocean waves striking the coast of Greenland or the Pacific -- can affect the sensitivity of the experiment and the data quality, resulting in "glitches" or periods of increased noise bursts, he said. "Monitoring the environmental conditions is continuously done at the sites," he said. "LIGO has more than 100,000 auxiliary channels with seismometers and accelerometers sensing the environment where the interferometers are located. The tool we developed can identify different environmental states of interest, such as earthquakes, microseisms, and anthropogenic noise, across a number of carefully selected and curated sensing channels." Vagelis Papalexakis, an associate professor of computer science and engineering who holds the Ross Family Chair in Computer Science, presented the team's paper, titled "Multivariate Time Series Clustering for Environmental State Characterization of Ground-Based Gravitational-Wave Detectors," at the IEEE's 5th International Workshop on Big Data & AI Tools, Models, and Use Cases for Innovative Scientific Discovery that took place last month in Washington, D.C. "The way our machine learning approach works is that we take a model tasked with identifying patterns in a dataset and we let the model find patterns on its own," Papalexakis said. "The tool was able to identify the same patterns that very closely correspond to the physically meaningful environmental states that are already known to human operators and commissioners at the LIGO sites." Papalexakis added that the team had worked with the LIGO Scientific Collaboration to secure the release of a very large dataset that pertains to the analysis reported in the research paper. This data release allows the research community to not only validate the team's results but also develop new algorithms that seek to identify patterns in the data. "We have identified a fascinating link between external environmental noise and the presence of certain types of glitches that corrupt the quality of the data," Papalexakis said. "This discovery has the potential to help eliminate or prevent the occurrence of such noise." The team organized and worked through all the LIGO channels for about a year. Richardson noted that the data release was a major undertaking. "Our team spearheaded this release on behalf of the whole LIGO Scientific Collaboration, which has about 3,200 members," he said. "This is the first of these particular types of datasets and we think it's going to have a large impact in the machine learning and the computer science community." Richardson explained that the tool the team developed can take information from signals from numerous heterogeneous sensors that are measuring different disturbances around the LIGO sites. The tool can distill the information into a single state, he said, that can then be used to search for time series associations of when noise problems occurred in the LIGO detectors and correlate them with the sites' environmental states at those times. "If you can identify the patterns, you can make physical changes to the detector -- replace components, for example," he said. "The hope is that our tool can shed light on physical noise coupling pathways that allow for actionable experimental changes to be made to the LIGO detectors. Our long-term goal is for this tool to be used to detect new associations and new forms of environmental states associated with unknown noise problems in the interferometers." Pooyan Goodarzi, a doctoral student working with Richardson and a coauthor on the paper, emphasized the importance of releasing the dataset publicly. "Typically, such data tend to be proprietary," he said. "We managed, nonetheless, to release a large-scale dataset that we hope results in more interdisciplinary research in data science and machine learning." The team's research was supported by a grant from the National Science Foundation awarded through a special program, Advancing Discovery with AI-Powered Tools, focused on applying artificial intelligence/machine learning to address problems in the physical sciences. Richardson, Papalexakis, and Goodarzi were joined in the research by Rutuja Gurav, a doctoral student working with Papalexakis; Isaac Kelly, a summer undergraduate REU student; Anamaria Effler of the LIGO Livingston Observatory; and Barry Barish, a UCR distinguished professor in physics and astronomy.