Multi-Wavelength Analysis of Kilonova Associated with GRB 230307A: Accelerated Parameter Estimation and Model Selection Through Likelihood-Free Inference - Accepted at the Machine Learning and the Physical Sciences Workshop, NeurIPS 2024

The mergers of binary compact objects are of central interest to several areas of astrophysics, including as the progenitors of short gamma-ray bursts (GRBs). Dozens of GRBs have been confidently associated with rapidly-decaying optical transients (``afterglows'') and more recently with late-time emission, such as from a kilonovae. Traditional methods for modeling these phenomena are computationally expensive for comparison and parameter inference because of the high-dimension parameter space. We propose using Simulation-Based Inference (SBI) as a fast and scalable alternative to conventional likelihood-based approaches. Preliminary results using SBI to fit multi-wavelength light curves of GRB,230307A across different emission models, highlight its efficiency in managing high-dimensional parameter space and show that SBI yields posterior distributions consistent with those from likelihood-based methods, while significantly reducing computational time from 110 hours to just 5 hours. This is a significant saving that can allow for more exhaustive model comparisons in future works.