Reduced-Order Modeling with Artificial Neurons for Gravitational-Wave Inference

Area of Science:

• Astrophysics
• Data Analysis
• Machine Learning

Background:

• Deep learning, particularly convolutional networks, is increasingly used for gravitational-wave data analysis.
• Current methods often treat time-series data as images, which may not be the most efficient approach.

Purpose of the Study:

• To develop an alternative, computationally efficient framework for gravitational-wave data analysis.
• To leverage reduced-order modeling and artificial neural networks for parameter estimation.

Main Methods:

• Representing gravitational waveforms as weighted sums over reduced bases (reduced-order modeling).
• Training artificial neural networks to map source parameters to basis coefficients.
• Performing statistical inference directly in the coefficient space.

Main Results:

• Demonstrated fast and accurate coefficient interpolation for a four-dimensional binary-inspiral waveform family.
• Neural networks provide analytic waveform derivatives, beneficial for sampling schemes.

Conclusions:

• The proposed framework offers a theoretically straightforward and computationally efficient alternative for gravitational-wave data analysis.
• This approach shows promising applications in parameter estimation for gravitational-wave events.