Speaker
Description
The use of Artificial Intelligence (AI), Machine Learning (ML), and High-Performance Computing (HPC) is playing an increasingly important role in advancing how we detect gravitational waves (GW).GW signals from core-collapse supernovae (CCSN) are still undetected. These signals are inherently weak, unmodeled and often masked by environmental and instrumental noise, presenting significant challenges for detection. CCSN waveforms are derived from complex simulations in the HPC. The simulations yield signals of stochastic nature that mimic noise. This study explores how convolutional neural networks (CNN) can effectively distinguish diverse CCSN GW signals from background noise.
This study addresses a fundamental query: how can a CNN be comprehensively trained to capture all potential CCSN signatures, optimizing accuracy? The investigation presents a multivariate classification of the entire supernova waveform landscape to strategically select training waveforms that maximize the feature space. Rigorously tested on both known and unknown waveforms, the method achieves a classification accuracy of ≥90%. This approach has been seamlessly incorporated into the multilayer signal enhancement with coherent wave burst and CNN (MuLaSEcC) analysis pipeline, showcasing promising outcomes using LIGO O3b data. Noteworthy improvements include a reduction in background by ≥99%, along with the calculation of detection efficiencies for ten contemporary explosion models. The study evaluates the search pipeline’s performance by illustrating detection probability as a function of false alarm rate and false alarm probability. The results highlight a ≥50% detection efficiency within an SNR range of 20–35 for the ten analyzed models, whether trained or untrained. Time-frequency images of CCSN signals detected by the pipeline show broadband features of the CCSN waveforms that are predicted in the simulations.
This study underscores the potential of artificial intelligence in gravitational wave data analysis, paving the way for more accurate and scalable detection frameworks.
