Abstract :
The era of modern astronomical surveys has ushered in an unprecedented deluge of cosmic data, unveiling new frontiers in our understanding of the Universe. This rich tapestry of information demands novel numerical and information processing techniques to unlock its full potential, particularly in the realm of cosmology. Bayesian inference, a cornerstone of cosmological and astrophysical analyses, now faces unprecedented challenges due to the high dimensionality and nonlinearity of the theoretical models, as well as the sparsity of the observational data. Traditional sampling techniques struggle to keep pace, presenting a crisis and an opportunity for scalable algorithms. In this talk, we introduce Preconditioned Monte Carlo (PMC), a cutting-edge algorithm that leverages geometric insights to enable robust and efficient sampling of high-dimensional, nonlinear posteriors. We demonstrate empirically that PMC offers substantial improvements in posterior sampling efficiency and evidence estimation, while favorably scaling to higher dimensions. PMC presents a promising solution for Bayesian inference in cosmology, offering a new tool to tackle the complexities of contemporary and forthcoming datasets. Furthermore, we discuss extensions and connections to related methods, and introduce pocoMC, our open-source PMC implementation tailored for the astronomical community.