Speaker
Description
Third-Generation (3G) observatories will observe Binary Neutron Star coalescences for hours before merger. Over these extended windows, Earth's motion induces signal modulations that render standard approaches computationally prohibitive for low-latency searches. To overcome this, we propose a 2D Convolutional Neural Network (CNN) processing multi-detector, complex Short Fourier Transform (SFT) tensors to implicitly marginalize over extrinsic parameters, effectively reducing the pipeline to a mass-only search. By optimizing a Binary Cross-Entropy loss, the network learns the signal's non-linear time-frequency morphology, approximating the optimal Bayesian detection statistic. We conduct a large-scale campaign of software-simulated signals to compare the CNN's performance with a GPU-optimized SFT-based Matched Filtering baseline for both $\triangle$ and 2L configurations for the Einstein Telescope, across three signal duration regimes: $T_\text{sig}\approx 25$ minutes ($f_{\text{min}} = 5$ Hz), $\approx 2$ hours ($3$ Hz), and $\approx 6$ hours ($2$ Hz). We found that the CNN achieves comparable detection sensitivity while reducing computational costs by over five orders of magnitude, slashing inference latency from minutes to milliseconds. This efficiency unlocks crucial low-latency early-warning triggers for 3G multi-messenger astronomy.