Field-level weak lensing cosmology with $<100$ simulations using multifidelity simulation-based inference

The article presents a method for field-level weak lensing cosmology that utilizes fewer than 100 high-fidelity N-body simulations through a multifidelity simulation-based inference (SBI) approach. By pre-training neural inference models on fast log-normal GLASS simulations and fine-tuning them with a limited number of high-fidelity simulations, the authors demonstrate that this method can achieve accurate cosmological posteriors while significantly reducing computational costs. This advancement is crucial for practitioners as it enables efficient and realistic cosmological analyses with reduced simulation requirements, enhancing the feasibility of large-scale weak lensing studies.