Speaker
Description
The unprecedented precision of recent galaxy surveys and CMB experiments is transforming cosmology but at a cost. As modelling frameworks grow more sophisticated, the number of nuisance parameters has surged, while emerging tensions with the standard cosmological model motivate extensions in cosmological parameters as well. Combined with computationally expensive models, these developments make standard Bayesian sampling increasingly inefficient, especially for joint analyses, where the real constraining power lies.
To address this, we introduce an adaptive emulation framework that targets the high-likelihood regions of parameter space, dramatically accelerating inference without sacrificing accuracy. Built on the mature ecosystem of CosmoSIS and its extensive library of likelihoods, our approach integrates a novel sampling strategy designed to bypass key computational bottlenecks.
In this talk, I will showcase the performance of this method across multiple case studies, including analysis pipeline optimisation and multi-probe combinations. The aim is simple: make state-of-the-art cosmological inference faster, lighter, and less dependent on large HPC resources.