by Leander Thiele, Adrian E. Bayer,
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.
Reference:
Leander Thiele, Adrian E. Bayer, Naoya Takeishi:Simulation-Efficient Cosmological Inference with Multi-Fidelity Simulation-Based Inference, In The 3rd Machine Learning for Astrophysics Workshop at ICML, Vancouver, 2025. arXiv:2507.00514
Bibtex Entry:
@conference{thieleML4ASTRO2025,
title = {Simulation-Efficient Cosmological Inference with Multi-Fidelity Simulation-Based Inference},
author = {Leander Thiele and Adrian E. Bayer and Naoya Takeishi},
labauthor = {Naoya Takeishi},
url = {https://arxiv.org/abs/2507.00514},
year = {2025},
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.},
booktitle = {The 3rd Machine Learning for Astrophysics Workshop at ICML, Vancouver},
note = {arXiv:2507.00514},
lang = {en}
}