Fast and credible likelihood-free cosmology with truncated marginal neural ratio estimation

Abstract Sampling-based inference techniques are central to modern cosmological data analysis; these methods, however, scale poorly with dimensionality and typically require approximate or intractable likelihoods. In this paper we describe how Truncated Marginal Neural Ratio Estimation ( tmnre ) (a new approach in so-called simulation-based inference) naturally evades issues, improving the i efficiency, ii scalability, iii trustworthiness of inference. Using measurements Cosmic Microwave Background (CMB), show that can achieve converged posteriors using orders magnitude fewer simulator calls than conventional Markov Chain Monte Carlo mcmc methods. Remarkably, examples required number samples is effectively independent nuisance parameters. addition, a property called local amortization allows performance rigorous statistical consistency checks not accessible sampling-based promises become powerful tool for analysis, particularly context extended cosmologies, where timescale methods converge greatly exceed simple models such as ?CDM. To perform computations, use an implementation via open-source code swyft .[ available at https://github.com/undark-lab/swyft . Demonstration on simulators used https://github.com/a-e-cole/swyft-CMB .]