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 2402.16270physics.ins-dethep-exphysics.comp-ph
收藏

用于簇计数的寻峰算法及域自适应

Peak finding algorithm for cluster counting with domain adaptation

Guang Zhao, Linghui Wu, Francesco Grancagnolo, Nicola De Filippis, Mingyi Dong, Shengsen Sun

AI总结:

本文提出一种半监督域自适应算法,利用最优传输对齐模拟与数据样本并结合连续小波变换部分标签进行学习,有效解决了漂移室簇计数的寻峰问题,在真实实验数据上表现优于传统算法且性能稳定。

摘要:

漂移室中的簇计数是粒子物理实验中粒子鉴别(PID)技术最有前景的突破。重建算法是簇计数的关键挑战之一。本文开发了一种半监督域自适应(DA)算法,并将其应用于簇计数中的寻峰问题。该算法使用最优传输(OT)来提供分布之间的几何度量,以对齐源(模拟)样本和目标(数据)样本之间的样本,并利用目标域中由连续小波变换(CWT)算法部分标记的样本进行半监督学习。该模型通过带标签的伪数据进行验证,其性能接近全监督模型。将该算法应用于在CERN使用180 GeV/c μ子束流获取的真实实验数据时,它显示出比传统的基于导数的算法更好的分类能力,并且对于不同径迹长度的实验数据样本,其性能保持稳定。

英文摘要:

Cluster counting in drift chamber is the most promising breakthrough in particle identification (PID) technique in particle physics experiment. Reconstruction algorithm is one of the key challenges in cluster counting. In this paper, a semi-supervised domain adaptation (DA) algorithm is developed and applied on the peak finding problem in cluster counting. The algorithm uses optimal transport (OT), which provides geometric metric between distributions, to align the samples between the source (simulation) and target (data) samples, and performs semi-supervised learning with the samples in target domain that are partially labeled with the continuous wavelet transform (CWT) algorithm. The model is validated by the pseudo data with labels, which achieves performance close to the fully supervised model. When applying the algorithm on real experimental data, taken at CERN with a 180 GeV/c muon beam, it shows better classification power than the traditional derivative-based algorithm, and the performance is stable for experimental data samples across varying track lengths.

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。

↑