Deep Transfer Learning and Unsupervised Clustering for Classifying Transient Noise in Gravitational Wave Detectors

Gravitational wave detection requires a detailed understanding of the response of the LIGO and Virgo detectors to true signals in the presence of environmental and instrumental noise. Of particular interest is the study of anomalous non-Gaussian transients, known as glitches, since their occurrence rate in LIGO and Virgo data can obscure or even mimic true gravitational wave signals. Therefore, successfully identifying and excising these anomalies from gravitational wave data is of utmost importance for the detection and characterization of true signals, and to accurately compute their significance.

This research shows that we can automatically detect and group together anomalies in data from the LIGO detectors by using artificial intelligence algorithms based on neural networks that were already pre-trained to classify photographs of real-world objects. We achieved state-of-the-art accuracy at classifying new anomalies by transferring knowledge from these neural networks, and then training them on a dataset hand-labeled by the Gravity Spy citizen science project. This method could help in finding and eliminating transient noise artifacts, thus improving the sensitivity of the detectors in the future, and it could also enable the detection of unmodeled gravitational wave signals.

Furthermore, we showed that these neural networks can be used as feature extractors for unsupervised clustering algorithms to facilitate finding entirely new and unknown classes of glitches/anomalies without human supervision.

Different classes of glitches are clustered after applying the feature extractor.
Confusion matrix showing high accuracy for this method, particularly for rare classes of glitches.

Project Members: Daniel George, Hongyu Shen, Eliu Huerta


Training Machine:  NVIDIA DGX-1 with Tesla V100 and NVIDIA Tesla P100 GPUs

Publications: Classification and clustering of LIGO data with deep transfer learning