Speaker
Thomas Ng
Description
The Einstein Telescope will require parameter estimation methods that scale far beyond current CPU-based pipelines. I will discuss a GPU-native stack built on JAX for fast Bayesian inference. These tools combine differentiable waveform models, machine-learning-enhanced sampling, and GPU-friendly likelihoods to reduce runtimes for current-generation data analysis from hours to minutes on modern GPUs. The talk will give a brief overview of JAX and this ecosystem, and will focus on ongoing work, including benchmarking different GPU-based samplers using GWTC-4.0 data, as well as implementing state-of-the-art waveforms in JAX.