Space-based gravitational wave signal detection and extraction with deep neural network
Problem
Matched filtering for space-based gravitational wave detectors would require a computationally unaffordable template bank, so a feasible detection and extraction method was needed.
Why it matters
Space-based detectors could observe sources nearly impossible to detect from the ground, but only if signal detection is computationally practical.
What the AI did
The authors developed a science-driven, uniform multi-stage self-attention-based deep neural network that detects and extracts gravitational wave signals, demonstrated as a proof of concept on synthetic signals in Gaussian noise.
Results
Detection rate exceeding 99% at signal-to-noise ratio 50 with 1% false alarm rate; at least 95% similarity to target signals; interpretability and generalization demonstrated
Impact
A proof of concept on synthetic data; not yet demonstrated on real detector data.
Record
| Venue | Communications Physics |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Tianyu Zhao, Ruoxi Lyu, He Wang, Zhoujian Cao, Zhixiang Ren |
| Citations | 35 (OpenAlex, at import time) |
| References | 73 |
| Model or method | Multi-stage self-attention-based deep neural network |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s42005-023-01334-6 |
| Open access | gold |
| Keywords | gravitational waves, deep neural network, signal detection, space-based detectors, self-attention |
Importance rating
significant — thousands to millions affected
Abstract states space-based detectors could observe otherwise nearly impossible sources and matched filtering is computationally unaffordable; detection feasibility is foundational for this scientific field, but only synthetic-data proof of concept.
More in Physics & Astronomy
| # | Paper | Year | Citations | Importance |
|---|---|---|---|---|
| 1 | Magnetic control of tokamak plasmas through deep reinforcement learning | 2022 | 739 |
7
|
| 2 | Solving the quantum many-body problem with artificial neural networks | 2017 | 2,205 |
7
|
| 3 | Learning high-accuracy error decoding for quantum processors | 2024 | 90 |
7
|
| 4 | Optimizing Quantum Error Correction Codes with Reinforcement Learning | 2019 | 158 |
7
|
| 5 | Machine-learning-assisted correction of correlated qubit errors in a topological code | 2018 | 141 |
7
|