Deep Residual Networks for Gravitational Wave Detection

Paraskevi Nousi,Alexandra E. Koloniari,Nikolaos Passalis,Panagiotis Iosif,Nikolaos Stergioulas,Anastasios Tefas

DOI: https://doi.org/10.1103/PhysRevD.108.024022

2023-06-29

Abstract:Traditionally, gravitational waves are detected with techniques such as matched filtering or unmodeled searches based on wavelets. However, in the case of generic black hole binaries with non-aligned spins, if one wants to explore the whole parameter space, matched filtering can become impractical, which sets severe restrictions on the sensitivity and computational efficiency of gravitational-wave searches. Here, we use a novel combination of machine-learning algorithms and arrive at sensitive distances that surpass traditional techniques in a specific setting. Moreover, the computational cost is only a small fraction of the computational cost of matched filtering. The main ingredients are a 54-layer deep residual network (ResNet), a Deep Adaptive Input Normalization (DAIN), a dynamic dataset augmentation, and curriculum learning, based on an empirical relation for the signal-to-noise ratio. We compare the algorithm's sensitivity with two traditional algorithms on a dataset consisting of a large number of injected waveforms of non-aligned binary black hole mergers in real LIGO O3a noise samples. Our machine-learning algorithm can be used in upcoming rapid online searches of gravitational-wave events in a sizeable portion of the astrophysically interesting parameter space. We make our code, AResGW, and detailed results publicly available at <a class="link-external link-https" href="https://github.com/vivinousi/gw-detection-deep-learning" rel="external noopener nofollow">this https URL</a> .

General Relativity and Quantum Cosmology,High Energy Astrophysical Phenomena

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are as follows: When dealing with binary black hole (BBH) systems with non - aligned spins, traditional gravitational - wave detection methods (such as matched filtering or wavelet - based model - free search) become impractical when exploring the entire parameter space, which severely limits the sensitivity and computational efficiency of gravitational - wave searches. Specifically:

1. **Limitations of matched filtering**: For BBH systems with non - aligned spins, if the entire parameter space is to be explored, the computational cost of matched filtering will be extremely high, resulting in limitations in sensitivity and computational efficiency.

2. **Limitations of model - free search**: Although the model - free search algorithm does not require specific waveform templates, its sensitivity is limited and depends on the specific gravitational - wave source.

To solve these problems, the author proposes a new method based on deep learning. In particular, on the basis of using techniques such as a 54 - layer deep residual network (ResNet), deep adaptive input normalization (DAIN), dynamic data augmentation, and curriculum learning, it achieves higher sensitivity and lower computational cost than traditional methods. Specific improvements include:

- **Greater sensitive distance**: The new method exceeds the sensitive distance of traditional techniques under specific settings.
- **Lower computational cost**: Compared with matched filtering, the computational cost of the new method is only a fraction.
- **Better handling of non - stationary noise**: By using DAIN to handle the non - stationary characteristics of LIGO O3a noise, the robustness of the model is improved.

These improvements enable this method to be applied in the upcoming rapid online gravitational - wave event search and cover a large amount of astrophysically interesting parameter space.