arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.04887hep-ph
收藏

用于矢量玻色子融合的曲率感知深度学习:微分几何、物理启发特征与量子方法局限性

Curvature-Aware Deep Learning for Vector Boson Fusion: Differential Geometry, Physics-Inspired Features, and Quantum Method Limitations

Faculty of Physics, University of Warsaw(华沙大学物理学院)

Alibordi Muhammad

AI总结:

针对粒子物理分类忽略对撞数据弯曲统计结构的问题,提出结合物理启发特征与乘积流形神经网络的几何框架,统一多种几何表示捕捉非线性关联,相比平坦基线有可测量提升,证明几何结构可增强高能对撞数据区分能力。

摘要:

粒子物理分类通常假设平坦几何,忽略对撞数据的弯曲统计结构。我们提出一种用于Vector Boson Fusion希格斯粒子分类的几何框架,将物理启发的可观测变量与乘积流形神经网络相结合。该方法统一了欧氏、双曲和球面表示,以捕捉运动学特征间的非线性关联。几何嵌入相比平坦基线带来了可测量的提升,表明曲率感知架构能恢复标准方法中丢失的信息。本研究凸显了引入几何结构可增强高能对撞数据中的区分能力。

英文摘要:

Particle physics classification often assumes flat geometry, ignoring the curved statistical structure of collision data. We present a geometric framework for Vector Boson Fusion Higgs classification that combines physics-inspired observables with product manifold neural networks. The method unifies Euclidean, hyperbolic, and spherical representations to capture nonlinear correlations among kinematic features. Geometric embedding yields measurable improvements over flat baselines, demonstrating that curvature-aware architectures recover information lost in standard approaches. The study highlights how incorporating geometric structure enhances discrimination power in high-energy collision data.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑