Speaker
Description
We present a new method for fast inference of pulsar emission and interstellar medium (ISM) parameters directly from frequency-resolved pulse profiles. Accurate parameter estimation is challenging due to strong degeneracies, and classical Bayesian likelihood-based fitting becomes computationally expensive for high-dimensional parameter spaces.
We build a physically motivated simulator that generates frequency-resolved pulses. The model includes multi-component profiles, spectral evolution of amplitude and width, dispersion measure, and scattering.
Using simulation-based inference (SBI) we are able to recover posterior distributions of dispersion measures and scattering parameters, without explicit likelihood evaluation.
Once trained, SBI enables faster inference with reduced computational costs.
This technique may be particularly useful for pulsar timing arrays, gravitational-wave astronomy and fast radio burst studies. In these studies Bayesian models are common and high dimensional and vast amounts of data are present, making efficient posterior estimation important.
| Talk category | Splinter 6: Data science |
|---|---|
| Second preference | Splinter session:5 if it covers Machine Learning, otherwise Splinter session:2 |
