Speaker
Description
Matched-filtering techniques are the standard approach for detecting compact binary coalescences (CBCs) and have been employed in all confirmed gravitational-wave (GW) detections to date. However, these methods rely on large banks of accurately modeled waveform templates, making them unsuitable for poorly modeled or unmodeled sources such as core-collapse supernovae (CCSN) and other unknown transient events. To address this limitation, several pipelines have explored waveform-agnostic approaches for gravitational-wave anomaly detection. In this work, we present GWAK2, an updated and improved version of our previous anomaly detection pipeline, Gravitational Wave Anomalous Knowledge (GWAK). GWAK2 introduces new training strategies and model orchestration techniques to enhance detection performance. The pipeline constructs a low-dimensional embedding space using a ResNet-based feature extractor and a normalizing flow model, enabling it to capture distinctive signal features and identify potential gravitational-wave events beyond the scope of modeled searches.
