Abstract:We investigate the use of Convolutional Neural Networks (including the modern ConvNeXt network family) to classify transient noise signals (i.e.~glitches) and gravitational waves in data from the Advanced LIGO detectors. First, we use models with a supervised learning approach, both trained from scratch using the Gravity Spy dataset and employing transfer learning by fine-tuning pre-trained models in this dataset. Second, we also explore a self-supervised approach, pre-training models with automatically generated pseudo-labels. Our findings are very close to existing results for the same dataset, reaching values for the F1 score of 97.18% (94.15%) for the best supervised (self-supervised) model. We further test the models using actual gravitational-wave signals from LIGO-Virgo's O3 run. Although trained using data from previous runs (O1 and O2), the models show good performance, in particular when using transfer learning. We find that transfer learning improves the scores without the need for any training on real signals apart from the less than 50 chirp examples from hardware injections present in the Gravity Spy dataset. This motivates the use of transfer learning not only for glitch classification but also for signal classification.

What problem does this paper attempt to address?

This paper attempts to address the problem of classifying transient noise signals (i.e., "glitches") and gravitational wave signals in gravitational wave data streams. Specifically, the researchers explore the application of convolutional neural networks (including the modern ConvNeXt network family) to this task. They employ both supervised and self-supervised learning methods and test the performance of these models with actual gravitational wave signals. ### Main Issues 1. **Transient Noise Signal Classification**: How to effectively identify and classify transient noise signals (glitches) in gravitational wave data, which may interfere with the detection of real gravitational wave signals. 2. **Model Performance Optimization**: How to utilize advanced deep learning techniques (such as convolutional neural networks and self-supervised learning) to improve the performance of classification models, especially when dealing with data from different observation runs. ### Research Background - **Development of Gravitational Wave Astronomy**: Since the first detection of gravitational waves from the merger of two black holes in 2015, the LIGO-Virgo-KAGRA (LVK) collaboration has reported results from multiple observation runs, confirming a total of 90 gravitational wave events. - **Noise Issues**: Gravitational wave detectors operate under extremely low noise conditions, but environmental and instrumental noise (such as ground motion, earthquakes, storms, etc.) can interfere with the normal operation of the detectors, generating transient noise signals (glitches) that may mimic real gravitational wave signals, increasing the false alarm rate. - **Existing Methods**: Various methods are currently used to classify and remove transient noise signals, such as BayesWave and gwsubtract, but these methods have limitations when processing large amounts of data. ### Research Objectives - **Improve Classification Accuracy**: Improve the accuracy of transient noise signal classification by using convolutional neural networks (especially the ConvNeXt network family) and self-supervised learning methods. - **Generalization Ability**: Evaluate the generalization ability of the models when processing data from different observation runs, especially without retraining on new data. - **Optimize Training Process**: Optimize the training process of the models using techniques such as the 1cycle training strategy and mixed precision training to reduce training time and resource consumption. ### Research Methods 1. **Supervised Learning**: - Train models from scratch using ResNet and ConvNeXt network architectures. - Use pre-trained models for transfer learning to reduce training time and improve performance. 2. **Self-Supervised Learning**: - Generate pseudo-labels, train the models on pretext tasks, and then fine-tune on real labeled data. 3. **Dataset**: - Use the Gravity Spy dataset, which contains transient noise signals from LIGO observation runs O1 and O2. - Split the dataset into training, validation, and test sets to evaluate model performance. ### Experimental Results - **Supervised Learning Models**: - The best model achieved an F1 score of 97.18%, comparable to existing methods. - Transfer learning significantly improved model performance, especially when processing O3 data. - **Self-Supervised Learning Models**: - The best model achieved an F1 score of 94.15%, slightly lower than supervised learning models but still performing well. - Self-supervised learning methods performed well with a small amount of labeled data. ### Conclusion - **Model Performance**: Convolutional neural networks (especially the ConvNeXt network family) performed excellently in classifying transient noise signals, achieving performance close to existing methods. - **Transfer Learning**: Transfer learning significantly improved the generalization ability of the models, allowing them to perform well on data from different observation runs. - **Future Work**: Further optimize model architectures and training strategies, explore more self-supervised learning methods to improve classification performance and reduce training time.

Convolutional Neural Networks for the classification of glitches in gravitational-wave data streams

Enhancing the Rationale of Convolutional Neural Networks for Glitch Classification in Gravitational Wave Detectors: A Visual Explanation

Glitch Classification and Clustering for LIGO with Deep Transfer Learning

Advancing Glitch Classification in Gravity Spy: Multi-view Fusion with Attention-based Machine Learning for Advanced LIGO's Fourth Observing Run

Convolutional neural network search for long-duration transient gravitational waves from glitching pulsars

GSpyNetTree: A signal-vs-glitch classifier for gravitational-wave event candidates

On Improving the Performance of Glitch Classification for Gravitational Wave Detection by using Generative Adversarial Networks

Application of Machine Learning Algorithms to the Study of Noise Artifacts in Gravitational-Wave Data

Machine Learning based Glitch Veto for inspiral binary merger signals using Linear Chirp Transform

Mitigating the impact of noise transients in gravitational-wave searches using reduced basis timeseries and convolutional neural networks

Detection of anomalies amongst LIGO's glitch populations with autoencoders

Gravitational-wave searches for cosmic string cusps in Einstein Telescope data using deep learning

Transient Classification in low SNR Gravitational Wave data using Deep Learning

Gravity Spy: Integrating Advanced LIGO Detector Characterization, Machine Learning, and Citizen Science

A new method to distinguish gravitational-wave signals from detector noise transients with Gravity Spy

Cross-Temporal Spectrogram Autoencoder (CTSAE): Unsupervised Dimensionality Reduction for Clustering Gravitational Wave Glitches

Real-Time Detection of Unmodelled Gravitational-Wave Transients Using Convolutional Neural Networks

Detection and mitigation of glitches in LISA data: A machine learning approach