Simulation-Efficient Cosmological Inference with Multi-Fidelity SBI

2025-07-01

AI Summary

Bayesian inference in cosmological simulations suffers from high computational cost and degraded posterior accuracy under limited simulation budgets. Method: We propose a multifidelity Bayesian inference framework that jointly leverages low- and high-fidelity simulation outputs. It aligns representations across fidelity levels via feature matching and employs knowledge distillation to transfer posterior information from high-fidelity simulations to a low-fidelity surrogate model, enabling efficient inference. Contribution/Results: Our approach establishes an end-to-end multifidelity simulation-based inference (SBI) pipeline. It significantly improves posterior accuracy and stability under constrained simulation budgets—particularly for high-dimensional, nonlinear cosmological parameter inference. Experiments demonstrate a 30–50% reduction in posterior error compared to single-fidelity baselines at equivalent computational cost, offering a scalable new paradigm for resource-constrained astrophysical inference.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Probabilistic Inference
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• Reducing simulation cost in cosmological inference
• Combining varying fidelity simulations efficiently
• Improving posterior quality with limited budgets

Innovation

• Combines varying fidelity simulation sets
• Uses feature matching and knowledge distillation
• Improves posterior quality efficiently