A Fast Generative Framework for High-dimensional Posterior Sampling: Application to CMB Delensing

TL;DR

This paper presents a rapid deep generative framework for high-dimensional Bayesian inference, significantly accelerating posterior sampling and effectively applied to CMB delensing to recover the unlensed CMB power spectrum.

Contribution

It introduces a novel, fast generative method for high-dimensional Bayesian inference, outperforming diffusion-based approaches in speed and robustness, with specific application to cosmological data analysis.

Findings

Achieves an order of magnitude faster sampling than diffusion models.

Successfully recovers the unlensed CMB power spectrum from simulated data.

Demonstrates robustness to shifts in cosmological parameters.

Abstract

We introduce a deep generative framework for high-dimensional Bayesian inference that enables efficient posterior sampling. As telescopes and simulations rapidly expand the volume and resolution of astrophysical data, fast simulation-based inference methods are increasingly needed to extract scientific insights. While diffusion-based approaches offer high-quality generative capabilities, they are hindered by slow sampling speeds. Our method performs posterior sampling an order of magnitude faster than a diffusion baseline. Applied to the problem of CMB delensing, it successfully recovers the unlensed CMB power spectrum from simulated observations. The model also remains robust to shifts in cosmological parameters, demonstrating its potential for out-of-distribution generalization and application to observational cosmological data.