Simulation-Efficient Cosmological Inference with Multi-Fidelity SBI

TL;DR

This paper introduces a multi-fidelity approach for cosmological inference that reduces simulation costs by combining different fidelity simulations, improving posterior quality especially under limited computational resources.

Contribution

It presents a novel multi-fidelity inference method using feature matching and knowledge distillation tailored for cosmological simulations.

Findings

Enhanced posterior accuracy with fewer high-fidelity simulations

Effective reduction in computational costs for inference

Improved performance in challenging inference scenarios

Abstract

The simulation cost for cosmological simulation-based inference can be decreased by combining simulation sets of varying fidelity. We propose an approach to such multi-fidelity inference based on feature matching and knowledge distillation. Our method results in improved posterior quality, particularly for small simulation budgets and difficult inference problems.