[论文解读] Advancing Tools for Simulation-Based Inference

Henning Bahl, Víctor Bresó+2 more|arXiv (Cornell University)|Oct 9, 2024

Simulation Techniques and Applications被引用 5

一句话总结

论文开发并测试先进的 SBI 工具以约束 LHC 粒子相互作用，引入具形态变换感知的估计、导数学习、一种新的分数模糊化技术，以及等变网络以改进似然学习，在玩具模型和 pp→WZ 的 LHC 上演示。

ABSTRACT

We study the benefit of modern simulation-based inference to constrain particle interactions at the LHC. We explore ways to incorporate known physics structures into likelihood estimation, specifically morphing-aware estimation and derivative learning. Technically, we introduce a new and more efficient smearing algorithm, illustrate how uncertainties can be approximated through repulsive ensembles, and show how equivariant networks can improve likelihood estimation. After illustrating these aspects for a toy model, we target di-boson production at the LHC and find that our improvements significantly increase numerical control and stability.

研究动机与目标

• Motivate the use of modern SBI to extract high-dimensional information from LHC events beyond traditional histograms.
• Integrate known physics structures into likelihood estimation to improve stability and accuracy.
• Develop and evaluate methods (morphing-aware estimation, derivative learning, fractional smearing, L-GATr) within SBI.
• Demonstrate the approaches on a toy model and on di-boson production at the LHC (pp → WZ).

提出的方法

• Derive and compare likelihood-learning strategies through classifier-based (CARL/ALICE) and likelihood-regression approaches.
• Incorporate physics structure via morphing to interpolate p(z_p|θ) across SMEFT-like parameter spaces.
• Introduce derivative-learning techniques to obtain R_i and R_ij derivatives of the reco-level likelihood.
• Propose fractional smearing to efficiently cover sparse high-sensitivity phase-space regions.
• Employ Lorentz-equivariant L-GATr networks to enhance likelihood learning in higher-dimensional spaces.

实验结果

研究问题

• RQ1 Can morphing-aware sampling and derivative learning provide more stable likelihood-ratio estimates than traditional SBI in high-dimensional LHC contexts?
• RQ2 How can physics structure be exploited to improve training efficiency and numerical stability of SBI methods for SMEFT-like parameter spaces?
• RQ3 Does fractional smearing enhance learning in sparsely populated, high-sensitivity regions of phase space?
• RQ4 What is the impact of using Lorentz-equivariant networks on likelihood estimation in reconstruction-level LHC data?
• RQ5 How do the proposed methods perform on a realistic LHC process like pp→WZ with SMEFT operators?

主要发现

• Morphing-aware likelihood estimation yields accurate likelihood ratios in the toy model, especially for non-local phase-space settings.
• Derivative-learning alone can underperform when the training basis points inadequately cover relevant phase space, highlighting the value of morphing in multi-parameter problems.
• Fractional smearing significantly improves learning in high-sensitivity, sparsely populated regions by balancing the target derivative distributions.
• L-GATr equivariant networks provide improvements in higher-dimensional reconstruction-level analyses.
• Applied to pp→WZ with SMEFT operators, the methods reveal clear impacts of Wilson coefficients on high-invariant-mass regions and improve stability of likelihood estimation.
• The combination of these techniques yields better numerical control and stability in SBI for LHC-like problems.