Speakers
Description
The statistical characterization of Fast Radio Burst (FRB) populations is fundamentally constrained by selection biases and the non-linear coupling of intrinsic source physics with extrinsic propagation effects. A primary hurdle is the DM-z degeneracy, where unknown host galaxy dispersion ($DM_{host}$) obscures cosmological distances. This is further compounded by morphological smearing; multi-path scattering. Consequently, current catalogs provide a biased representation of the underlying population, necessitating a robust framework to disentangle these signatures.
To address these challenges, we implement a computational pipeline that integrates a physical model of FRB emission with a stochastic transport module for the intervening medium, ultimately convolved with a deterministic telescope response function. The first stage of the project establishes a minimal physical model of FRB to define baseline parameter correlations. Subsequent stages will expand this framework by integrating multi-parameter population synthesis tools, such as \texttt{frbpoppy}, alongside more sophisticated models of burst morphology and DM-z relations. The machine learning module is structured as a stochastic inverse mapping of the above mentioned forward-modeled populations.
By employing a Deep Neural Network (DNN) as an estimator, we test the feasibility of recovering intrinsic burst parameters by mapping observables from the detection domain back to the progenitor’s source model parameters. This process requires the network to account for the degenerate effects of the transport medium and the telescope response function. We evaluate the performance of this pipeline in yielding estimations for critical burst parameters by testing it with the CHIME/FRB catalog, providing a preliminary assessment of whether such a framework can reliably constrain intrinsic burst geometry or contribute to the use of FRBs as cosmological probes.