Field-level weak lensing cosmology with 60 simulations using multifidelity simulation-based inference
astro-ph.CO, cs.AI
Submitted: 2026-06-22
Updated: 2026-09-07
Comments: 19 + 7 pages, 13 + 4 figures
Code: https://github.com/asaoulis/glass_gower_transfer
License: http://creativecommons.org/licenses/by/4.0/
The gist: We perform a realistic KiDS-Legacy mock analysis with field-level neural compression and simulation-based inference using fewer than 100 N-body simulations.
Terminology
Abstract
We perform a realistic KiDS-Legacy mock analysis with field-level neural compression and simulation-based inference using fewer than 100 N-body simulations. The weak lensing shear field encodes substantially more cosmological information than standard two-point summary statistics such as the power spectrum. Field-level inference can fully exploit this information, but physical realism at the field-level requires very high-fidelity simulations. This poses a major challenge for simulation-based inference (SBI): accurate empirical density modelling and deep-learning-based neural compression require many training simulations, but achieving physical realism at the field level makes each simulation extremely costly. We demonstrate that multifidelity SBI can alleviate this tension by substantially reducing the number of high-fidelity simulations needed for accurate cosmological inference. We pre-train neural inference models on realistic KiDS-Legacy-like shear mocks using fast log-normal GLASS simulations and fine-tune them on a small set of high-fidelity N-body simulations. We show that between 60 - 100 high-fidelity simulations are sufficient to obtain informative and well-calibrated cosmological posteriors, enabling an order-of-magnitude reduction in simulation cost for accurate field-level inference in a realistic setting.
Sources
- Dark Energy Survey Year 6 Results: Cosmological Constraints from Cosmic Shear
- How many simulations do we need for simulation-based inference in cosmology?
- Low-Budget Simulation-Based Inference with Bayesian Neural Networks
- The Llama 3 Herd of Models
- ${\rm S{\scriptsize IM}BIG}$: Mock Challenge for a Forward Modeling Approach to Galaxy Clustering
- Gaussian Error Linear Units (GELUs)
- Multilevel neural simulation-based inference
- Training Compute-Optimal Large Language Models
- Cosmological Analysis with Calibrated Neural Quantile Estimation and Approximate Simulators
- Scaling Laws for Neural Language Models
- Dark Energy Survey Year 3 Results: Multi-Probe Modeling Strategy and Validation
- Multifidelity Simulation-based Inference for Computationally Expensive Simulators
- Optimal Neural Summarisation for Full-Field Weak Lensing Cosmological Implicit Inference
- Learning Optimal and Interpretable Summary Statistics of Galaxy Catalogs with SBI
- Deep Ensembles Secretly Perform Empirical Bayes
- Decoupled Weight Decay Regularization
- Euclid preparation. CIV. Impact of galaxy intrinsic alignment modelling choices on Euclid 3x2pt cosmology
- KiDS-Legacy: Constraining dark energy, neutrino mass, and curvature
- Domain Adaptive Graph Neural Networks for Constraining Cosmological Parameters Across Multiple Data Sets
- Simulation-Efficient Cosmological Inference with Multi-Fidelity SBI
Related papers
- Angular clustering and bias of photometric quasars in the Kilo-Degree Survey Data Release 4
- A Novel kinetic Sunyaev-Zel'dovich Estimator for Electron-Electron Correlations
- Magnetic fields at the dawn of structure formation I. The CARLA J1510+5958 proto-cluster
- Dark Energy Survey Year 6 Results: Weak Lensing and Galaxy Clustering Cosmological Analysis Framework
- Exploring the Impact of Systematic Bias in Type Ia Supernova Cosmology Across Diverse Dark Energy Parametrizations
- Non-Gaussian Galaxy Stochasticity and the Noise-Field Formulation