Simulation-Efficient Cosmological Inference with Multi-Fidelity SBI
This addresses simulation efficiency for researchers in cosmology, but it is incremental as it builds on existing multi-fidelity and SBI methods.
The paper tackles the high simulation cost in cosmological simulation-based inference by proposing a multi-fidelity approach using feature matching and knowledge distillation, resulting in improved posterior quality, especially for small simulation budgets and difficult problems.
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.