Beamsynthesis

波束合成是由高能粒子加速器的直线加速器 (线性粒子加速器) 部分的模拟构造的电流波形的360时间序列数据的简单数据集。数据集包含两个地面真实数据生成因子: 表示可以采用三个值 (10,15，20) 之一的粒子波形频率的分类随机变量，以及由120波形占空比值 ∈ [0.2，0.8)。两个数据生成因子的笛卡尔积形成了一组观察值。解开这个数据集的挑战来自于这样一个事实，即波形的频率和占空比都会影响每个波的 “开启” 周期的长度。我们可视化不同算法的完整潜在空间并评估算法在使用该数据集提取正确数量的地面真实数据生成因子时的可靠性。

This simple Beam Synthesis dataset comprises 360 time-series samples of current waveforms, which are simulated from the linear accelerator (linac, linear particle accelerator) component of a high-energy particle accelerator. The dataset contains two ground-truth data generation factors: a categorical random variable representing the particle waveform frequency, which takes three discrete values: 10, 15, and 20; and a set of 120 duty cycle values for the waveforms, ranging from 0.2 (inclusive) to 0.8 (exclusive). The Cartesian product of these two factors forms the complete set of observational samples. The challenge of disentangling this dataset stems from the fact that both the waveform frequency and duty cycle affect the length of the "on" period of each individual waveform. We visualize the full latent spaces of different algorithms and evaluate their reliability when extracting the correct number of ground-truth data generation factors using this dataset.

搜集汇总

背景与挑战

背景概述

Beamsynthesis数据集包含由高能粒子加速器模拟生成的360个时间序列数据，聚焦于粒子波形频率和占空比两个数据生成因子，用于评估算法在提取这些因子时的可靠性。该数据集由杜克大学和橡树岭国家实验室于2022年发布。