Speaker
Description
Estimating the source parameters of gravitational wave signals is commonly performed with Bayesian inference or, more recently, simulation-based inference using deep learning techniques. In the context of next generation detectors these methods must address the significant challenge posed by long-duration signals, which may span several minutes to hours. Such extended signals result in extremely high-dimensional inputs, hindering neural networks from learning effectively and generalizing, leading to inaccurate inference.
We present an approach that leverages deep neural networks to construct compressed representations of long-duration gravitational wave data. We then perform parameter estimation using these compact representations, enabling efficient inference for extended signals.