Simulation-based inference (SBI) enables cosmological parameter estimation when closed-form likelihoods or models are unavailable. However, SBI relies on machine learning for neural compression and density estimation. This requires large training datasets which are prohibitively expensive for high-quality simulations. We overcome this limitation with multifidelity transfer learning, combining less expensive, lower-fidelity simulations with a limited number of high-fidelity simulations. We demonstrate our methodology on dark matter density maps from two separate simulation suites in the hydrodynamical CAMELS Multifield Dataset. Pre-training on dark-matter-only N-body simulations reduces the required number of high-fidelity hydrodynamical simulations by a factor between 8 and 15, depending on the model complexity, posterior dimensionality, and performance metrics used. By leveraging cheaper simulations, our approach enables performant and accurate inference on high-fidelity models while substantially reducing computational costs.
@article{arxiv.2505.21215,
title = {Transfer learning for multifidelity simulation-based inference in cosmology},
author = {Alex A. Saoulis and Davide Piras and Niall Jeffrey and Alessio Spurio Mancini and Ana M. G. Ferreira and Benjamin Joachimi},
journal= {arXiv preprint arXiv:2505.21215},
year = {2025}
}