Identifying and mitigating machine-learning biases for the gravitational-wave detection problem

Aug 3, 2026

Deep learning for gravitational-wave search has shown promise in learning from detector noise, yet generalization can fail due to learning biases that reduce detection sensitivity. This work identifies 11 interconnected biases in supervised training for the gravitational-wave detection problem, explains why common sensitivity metrics become unreliable, and evaluates trustworthiness of prior machine-learning results. Using domain knowledge, the authors build the Sage binary black hole search pipeline and validate it with challenge injection studies, reporting improved signal recovery and robust behavior across noise and non-Gaussian artifacts.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages40

Reading Time1 h 41 min

AccessFree to download

Introduction

Deep learning and generalization limits

Matched filtering vs ML-based detection

Scope: biases and mitigation tactics

Biases and mitigation

Identifying 11 interconnected biases

Reliability of detection sensitivity metrics

Validation and experiments

Injection study and challenge data

Comparisons with PyCBC and prior ML pipelines

Ablation of Sage components

Identifying and mitigating machine-learning biases for the gravitational-wave detection problem Narenraju Nagarajan* and Christopher Messenger† SUPA, School of Physics and Astronomy, University of Glasgow, Glasgow G12 8QQ, United Kingdom (Received 27 January 2025; accepted 3 October 2025; published 4 November 2025) Matched filtering is a long-standing technique for the optimal detection of known signals in stationary Gaussian noise. However, it has known departures from optimality when operating on unknown signals in real noise and suffers from computational inefficiencies in its pursuit of near-optimality. A compelling alternative that has emerged in recent years to address this problem is deep learning. Although it has shown significant promise when applied to the search for gravitational-waves in detector noise, we demonstrate the existence of a multitude of learning biases that hinder generalization and lead to significant loss in detection sensitivity. Our work identifies the sources of a set of 11 interconnected biases present in the supervised learning of the gravitational-wave detection problem and contributes mitigation tactics and training strategies to concurrently address them. In light of the identified biases, we demonstrate that existing detection sensitivity metrics are not reliable for machine-learning pipelines and discuss the trustworthiness of previous results. We use gravitational-wave domain knowledge to build a bespoke machine-learning-based binary black hole search pipeline called Sage that addresses these biases. Via the injection study presented in the Machine-Learning Gravitational-Wave Search Challenge, we show that Sage detects ≈11 .2% more signals than the benchmark PyCBC analysis at a false-alarm rate of one per month in O3a noise. Moreover, we also show that it can detect ≈48 .29% more signals than the previous best-performing machine-learning pipeline on the same dataset. We empirically prove that Sage can (i) effectively handle out-of-distribution noise power spectral densities, (ii) strongly reject non-Gaussian transient noise artifacts, and (iii) achieve higher detection sensitivities using less data than network architectures of a similar size. By studying machine-learning biases and conducting empirical investigations to understand the reasons for performance improvement or degradation, we aim to address the need for interpretability of machine-learning methods for gravitational-wave science. All code and implementations are available. DOI: 10.1103/zwj9-ycyz I. INTRODUCTION The promise of deep learning in the field of applied data science has been to develop fast approximation machines capable of emulating physical principles. Although it has been successful in seemingly approaching this ideal, it falls short due to its inability to truly generalize or adapt outside its training distribution (otherwise called brittleness) [1,2], being biased [3] and data intensive [4] . We hope that, by acknowledging the issues with machine-learning (ML), we can be better informed of their capabilities and, more importantly, their deficits. While these deficits exist, it is *Contact author: [n.nagarajan.1@research.gla.ac.uk](n.nagarajan.1@research.gla.ac.uk) †Contact author: [christopher.messenger@glasgow.ac.uk](christopher.messenger@glasgow.ac.uk) Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. possible to mitigate them and produce reliable pipelines via careful consideration of domain-specific knowledge. In this paper, we focus on identifying and addressing some of the biases that prevail in the application of ML to the gravitational-wave (GW) detection of binary black-hole (BBH) mergers. Although we focus on the BBH detection problem and provide explanations from this persp

FAQs

What problem do the authors focus on in machine-learning gravitational-wave detection?

They focus on identifying learning biases that prevent models from generalizing beyond the training distribution in the gravitational-wave detection of binary black hole mergers.

How many interconnected biases does the paper identify?

The paper identifies a set of 11 interconnected biases present in supervised learning for the gravitational-wave detection problem.

What improvements does the Sage pipeline achieve in the injection study?

The injection study shows Sage detects about 11.2% more signals than the PyCBC benchmark at a false-alarm rate of one per month in O3a noise, and about 48.29% more signals than the previous best-performing machine-learning pipeline on the same dataset.

Aug 3, 2026

Upload by Genevieve