射流子结构分析的自动发现
Automated Discovery of Jet Substructure Analyses
摘要:
对来自夸克和胶子或喷流的准直粒子的子结构的研究,有望揭示色荷如何与RHIC和LHC等对撞机中产生的QCD等离子体介质相互作用的细节。传统的喷流子结构观测量是使用专业知识构建的,并且在很大程度上是从高能物理学中移植的,未经修改,其目标主要是研究增强的强子衰变。描述了一种新的神经网络体系结构,该体系结构能够检查理论模型,并构造对内部模型特征敏感的分析过程。该架构与符号回归相结合,进一步允许从学习结果中提取封闭形式的代数表达式-使自动构建的射流子结构分析能够随后被人类理解和再现。然后,该系统的任务是构建一个分析,从观察射流推断等离子体温度,这是使用JEWEL和线性玻尔兹曼传输模型证明,并在一个现实的残余的等离子体,或潜在的事件,测量必须克服的存在。在一个演示中,算法如何能够产生与人类专家直接竞争的原创性研究,所得到的射流子结构变量和分析能够通过分析1200- 2500个射流来确定等离子体介质的初始温度,这是现有的手动设计的分析中看不到的性能。一个偶然发现的观察与现有文献的比较进一步表明,所描述的系统能够检查模型相空间的细节,至少可以与当前领域的人类专家相媲美。
英文摘要:
The study of the substructure of collimated particles from quarks and gluons, or jets, has the promise to reveal the details how color charges interact with the QCD plasma medium created in colliders such as RHIC and the LHC. Traditional jet substructure observables have been constructed using expert knowledge, and are largely transplanted, unmodified, from the high-energy physics, where the goal is primarily the study of boosted hadronic decays. A novel neural network architecture is described that is capable of examining theoretical models, and constructs, on its own, an analysis procedure that is sensitive to the internal model features. This architecture, in combination with symbolic regression, further allows the extraction of closed-form algebraic expressions from the learned result -- enabling the automatically constructed jet substructure analysis to be subsequently understood and reproduced by humans. This system is then tasked to construct an analysis that infers the plasma temperature from observing jets, which is demonstrated using both JEWEL and the Linearized Boltzmann Transport model, and at the presence of a realistic remnant of the plasma, or underlying event, that the measurement has to overcome. In a demonstration how algorithms can produce original research in direct competition to human experts, the resulting jet substructure variables and analyses are capable of determining the initial temperature of the plasma medium from analyzing 1200--2500 jets, a performance not seen in existing, manually designed analyses. Comparison of an incidentally discovered observable with the existing literature further indicates that the system described is capable of examining the model phase spaces to a detail at least comparable to the current field of human experts.
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。