Self-supervised learning for gravitational wave signal identification

TL;DR

This paper introduces a self-supervised learning approach for gravitational wave detection, significantly reducing computational costs by leveraging contrastive learning on simulated signals in noise for space-based detectors.

Contribution

It presents a novel self-supervised model tailored for gravitational wave signal identification, demonstrating high efficiency in low-latency detection scenarios.

Findings

Effective detection of simulated GW signals in noise

Reduced computational cost compared to traditional methods

Potential applicability to space-based GW detectors

Abstract

The computational cost of searching for gravitational wave (GW) signals in low latency has always been a matter of concern. We present a self-supervised learning model applicable to the GW detection. Based on simulated massive black hole binary signals in synthetic Gaussian noise representative of space-based GW detectors Taiji and LISA sensitivity, and regarding their corresponding datasets as a GW twins in the contrastive learning method, we show that the self-supervised learning may be a highly computationally efficient method for GW signal identification.