Machine Learning & Surrogates

We develop machine learning techniques and surrogate models to accelerate gravitational wave analysis and improve detection efficiency. Our research combines advanced signal processing methods with deep learning to enhance the capabilities of current and future GW detectors.

GW Denoising and Waveform Reconstruction

Given the low signal-to-noise ratios (SNRs) in GW interferometer data, effective noise reduction is critical for reliable detection and parameter estimation. This project enhances GW data analysis by implementing advanced denoising methods based on total-variation regularization and machine learning. These techniques will be integrated into the coherent Wave Burst (cWB) pipeline to improve the detectability of unmodeled burst sources such as CCSN and magnetar flares.

Machine Learning for GW Data Analysis

Deep learning and convolutional neural networks (CNNs) are becoming powerful tools in GW astronomy, particularly for event classification and parameter inference. This project aims to develop machine learning models for detecting CCSN signals and compact binary coalescences (CBC). Previous work by the group has demonstrated that CNN-based approaches can achieve high accuracy in identifying BBH mergers using spectrograms. This research will extend these methods to full CBC searches in existing and future LVK transient catalogs.

Surrogate Models and Rapid Parameter Estimation

Numerical relativity simulations are computationally expensive, making it challenging to perform comprehensive parameter estimation studies. We develop surrogate models that provide fast and accurate waveform predictions across large parameter spaces. These models enable real-time parameter estimation and rapid follow-up of gravitational wave events, which is crucial for multi-messenger astronomy campaigns.