Low-cost neural summaries for cosmological inference: an ensemble of convolutional neural networks trained on approximate simulations
Abstract
Machine learning can build near-optimal summary statistics for extracting cosmological information from the large-scale structure of the Universe. Training them, however, requires large numbers of costly simulations. We aim to build a neural summary trained on fast, approximate low-resolution simulations that remains informative when applied to higher-resolution ones, reducing the cost of the training set. We train an ensemble of convolutional neural networks with different architectures to compress the matter overdensity fields of low-resolution particle-mesh simulations. The outputs of a subset of networks are concatenated into a summary statistic, which we use to constrain Ωm and σ8 from high-resolution particle-mesh simulations, each about a thousand times more costly than their low-resolution counterparts. Theoretical predictions are provided by a Gaussian process emulator built from the high-resolution simulations. We test three strategies for selecting the smallest informative subset of networks. Constraints are compared with those from a reference power spectrum analysis. We also consider dark matter halo catalogues from independent high-resolution N-body simulations. The neural summary extracts information beyond the binned power spectrum. Using multiple networks is essential: a single network can be less informative than the power spectrum. A greedy search based on posterior volume provides the tightest constraints with the fewest networks. The neural summary also outperforms the power spectrum for dark matter haloes, at different number densities and redshifts, despite being trained on low-resolution matter fields at z=0. This work demonstrates that an ensemble of networks trained on approximate simulations yields a summary statistic that remains informative on higher-fidelity data, including dark matter haloes. (Abridged)
Related
- Rates of tidal disruption events from constrained cosmological simulations of the local Universe: population properties and implications for transient surveys
- Coupling between stellar and HI lopsidedness in Milky Way--type galaxies from the Auriga Superstars cosmological simulations
- Bridging perturbation theory and simulations: initial conditions and fast integrators for cosmological simulations
- Agentic Systems for Calibrating Cosmological Simulations
- OpenCosmo: Community Portal and Analysis Framework for Flagship Cosmological Simulations
BibTeX
@misc{saezcasares2026lowcost,
title = {{Low-cost neural summaries for cosmological inference: an ensemble of convolutional neural networks trained on approximate simulations}},
author = {Iñigo Sáez-Casares and Davide Bianchi and Matteo Calabrese and Luigi Guzzo},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.19891},
eprint = {2610.19891},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.19891},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Low-cost neural summaries for cosmological inference: an ensemble of convolutional neural networks trained on approximate simulations
AU - Iñigo Sáez-Casares
AU - Davide Bianchi
AU - Matteo Calabrese
AU - Luigi Guzzo
PY - 2026
DA - 2026/10/14
PB - arXiv
DO - 10.48550/arxiv.2610.19891
UR - https://arxiv.org/abs/2610.19891
AB - Machine learning can build near-optimal summary statistics for extracting cosmological information from the large-scale structure of the Universe. Training them, however, requires large numbers of costly simulations. We aim to build a neural summary trained on fast, approximate low-resolution simulations that remains informative when applied to higher-resolution ones, reducing the cost of the training set. We train an ensemble of convolutional neural networks with different architectures to compress the matter overdensity fields of low-resolution particle-mesh simulations. The outputs of a subset of networks are concatenated into a summary statistic, which we use to constrain Ω_m and σ_8 from high-resolution particle-mesh simulations, each about a thousand times more costly than their low-resolution counterparts. Theoretical predictions are provided by a Gaussian process emulator built from the high-resolution simulations. We test three strategies for selecting the smallest informative subset of networks. Constraints are compared with those from a reference power spectrum analysis. We also consider dark matter halo catalogues from independent high-resolution N-body simulations. The neural summary extracts information beyond the binned power spectrum. Using multiple networks is essential: a single network can be less informative than the power spectrum. A greedy search based on posterior volume provides the tightest constraints with the fewest networks. The neural summary also outperforms the power spectrum for dark matter haloes, at different number densities and redshifts, despite being trained on low-resolution matter fields at z=0. This work demonstrates that an ensemble of networks trained on approximate simulations yields a summary statistic that remains informative on higher-fidelity data, including dark matter haloes. (Abridged)
ER -