Researchers have introduced a novel, unsupervised machine learning method designed to uncover fresh patterns in the auxiliary channel data from the Laser Interferometer Gravitational-Wave Observatory (LIGO).
Scientists at the University of California, Riverside, have made it easier to pinpoint patterns and minimize noise in the vast and intricate datasets produced by the gravitational wave-detecting LIGO facility.
At a recent IEEE big-data workshop, the UCR team revealed their findings about a machine learning method that autonomously identifies patterns in LIGO’s auxiliary channel data. This innovative technology may also be relevant for large-scale particle accelerator studies and complex industrial systems.
LIGO serves as a facility for detecting gravitational waves—brief ripples in the fabric of spacetime caused by moving massive objects. It was the pioneer in detecting waves from colliding black holes, thereby confirming a vital element of Einstein’s Theory of Relativity. Comprising two distant 4-km-long interferometers located in Hanford, Washington, and Livingston, Louisiana, LIGO uses high-powered lasers to identify gravitational waves. The information gathered from these detectors provides new insights into the universe and informs inquiries about the nature of black holes, cosmology, and the densest forms of matter in existence.
Each of the LIGO detectors generates massive amounts of data streams, or channels, from various environmental sensors situated at the observation sites.
“Our machine learning approach, developed in close collaboration with LIGO officials and stakeholders, autonomously identifies patterns in the data,” stated Jonathan Richardson, assistant professor of physics and astronomy and lead of the UCR LIGO group. “We have found it effectively captures the environmental ‘states’ known to LIGO operators without any human intervention. This capability paves the way for a potent new experimental resource to assist in isolating noise interferences and guide enhancements to the detectors.”
Richardson noted that LIGO detectors are highly sensitive to external disturbances. Vibrations caused by factors such as wind or ocean waves can impede the experiment’s precision and data integrity, leading to “glitches” or periods marked by increased noise, he said.
“Monitoring the environmental circumstances at the sites is an ongoing process,” he added. “LIGO is equipped with over 100,000 auxiliary channels with seismometers and accelerometers that track the environmental conditions around the interferometers. The tool we developed can discern varied environmental states of interest, like earthquakes, microseisms, and human-caused noise, across multiple carefully selected monitoring channels.”
Vagelis Papalexakis, associate professor of computer science and engineering and 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 held last month in Washington, D.C.
“Our machine learning method is designed to allow a model to autonomously discover patterns within a dataset,” said Papalexakis. “The tool successfully identified patterns aligning closely with the significant environmental states already recognized by LIGO operators.”
Papalexakis also indicated that the team collaborated with the LIGO Scientific Collaboration to facilitate the release of a substantial dataset related to their research. This data release enables the scientific community to validate their findings and create new algorithms for identifying data patterns.
“We found a compelling connection between environmental noise and specific glitches that degrade data quality,” Papalexakis remarked. “This finding could assist in mitigating or preventing such noise.”
The team engaged in organizing and analyzing LIGO’s channels for approximately a year. Richardson pointed out that releasing the data was a major effort.
“Our team spearheaded this effort on behalf of the entire LIGO Scientific Collaboration, which comprises about 3,200 members,” he noted. “This marks the first release of these particular datasets, and we believe it will significantly influence the machine learning and computer science communities.”
Richardson clarified that the developed tool can assimilate data from a variety of sensors that monitor different disturbances surrounding the LIGO facilities. The tool synthesizes this information into a singular state, which can be utilized to track when noise issues occurred in the LIGO detectors and correlate them with environmental conditions at those times.
“Identifying these patterns could lead to physical modifications to the detectors, like replacing components,” he explained. “We hope our tool will illuminate physical noise coupling pathways to enable practical experimental adjustments to the LIGO detectors. Ultimately, we aim for this tool to identify new associations and unexplored environmental states linked to unknown noise challenges in the interferometers.”
Coauthor Pooyan Goodarzi, a doctoral student working with Richardson, underscored the significance of making the dataset publicly available.
“Usually, such data is proprietary,” he said. “However, we succeeded in releasing a large-scale dataset, which we hope will foster more interdisciplinary research in data science and machine learning.”
The National Science Foundation supported the team’s research through a unique program, Advancing Discovery with AI-Powered Tools, which focuses on applying artificial intelligence and machine learning to resolve challenges in the physical sciences.
Alongside Richardson, Papalexakis, and Goodarzi, the research involved Rutuja Gurav, a doctoral student working with Papalexakis; Isaac Kelly, a summer undergraduate REU student; Anamaria Effler from the LIGO Livingston Observatory; and Barry Barish, a distinguished professor in physics and astronomy at UCR.
