Enhancing Gravitational-Wave Detection: A Machine Learning Pipeline Combination Approach with Robust Uncertainty...

Area of Science:

• Astrophysics
• Data Science
• Signal Processing

Background:

• Gravitational-wave data contains noise, artifacts, and rare astrophysical signals.
• Existing search algorithms for compact binary coalescences have variable performance, complicating data interpretation.

Purpose of the Study:

• To present a machine-learning-driven approach for combining gravitational-wave search results.
• To provide robust, calibrated uncertainty quantification using conformal prediction.
• To improve the detection efficiency and confidence in multipipeline analyses of gravitational-wave events.

Main Methods:

• Utilized a machine-learning model to integrate outputs from multiple gravitational-wave search pipelines.
• Employed conformal prediction techniques for uncertainty quantification.
• Validated the approach using simulated gravitational-wave data.
• Applied the model to the Gravitational Wave Transient Catalog 3 (GWTC-3).

Main Results:

• Demonstrated improved detection efficiency through simulations.
• Enhanced confidence in multipipeline detections of gravitational-wave signals.
• Successfully applied the model to GWTC-3 data.
• Increased confidence in identifying subthreshold events, such as the binary neutron star candidate GW200311_103121.

Conclusions:

• The machine-learning approach offers a robust method for analyzing gravitational-wave data.
• Conformal prediction provides reliable uncertainty quantification for gravitational-wave event detection.
• This method improves the interpretation of complex gravitational-wave signals and enhances astrophysical discovery potential.