Transfer learning for multifidelity simulation-based inference in cosmology
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909807902982144 |
|---|---|
| author | Saoulis, Alex A. Piras, Davide Jeffrey, Niall Mancini, Alessio Spurio Ferreira, Ana M. G. Joachimi, Benjamin |
| author_facet | Saoulis, Alex A. Piras, Davide Jeffrey, Niall Mancini, Alessio Spurio Ferreira, Ana M. G. Joachimi, Benjamin |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_21215 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Transfer learning for multifidelity simulation-based inference in cosmology Saoulis, Alex A. Piras, Davide Jeffrey, Niall Mancini, Alessio Spurio Ferreira, Ana M. G. Joachimi, Benjamin Cosmology and Nongalactic Astrophysics Machine Learning 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. |
| title | Transfer learning for multifidelity simulation-based inference in cosmology |
| topic | Cosmology and Nongalactic Astrophysics Machine Learning |
| url | https://arxiv.org/abs/2505.21215 |