Quantum theory according to the Copenhagen interpretation holds that, when a quantum interaction is observed (i.e., “measured”), the observer’s measuring devices temporarily become a part of the quantum system. Rel...Quantum theory according to the Copenhagen interpretation holds that, when a quantum interaction is observed (i.e., “measured”), the observer’s measuring devices temporarily become a part of the quantum system. Relativity theory holds that the event clock of the absorbed or emitted photon or graviton is frozen in time relative to all clocks outside the observed system. If we harmonize both theories, this would appear to imply that time continuity must be interrupted at each instant of observed photon or graviton interaction with matter. It is as if a segment of space-time is clipped out during each such observed interaction. If so, we must dispense with the notion of an absolutely smooth and continuous space-time and replace it with an observation-dependent, discontinuous, relativistic/quantum space-time. Mathematical physicists should be able to model this hypothesis (call it a “time-jump hypothesis”) and its inherent discontinuous space-time in their further efforts at unification.展开更多
从时间序列流中获取事件是对时间序列流处理的基础.目前的研究大多采用传统的阈值确定方法对数据点进行查询,以获取时间序列流中存在的事件信息.在真实场景中,事件通常被定义为在连续一段时间内包含多种信息的异常,然而现有方法无法快...从时间序列流中获取事件是对时间序列流处理的基础.目前的研究大多采用传统的阈值确定方法对数据点进行查询,以获取时间序列流中存在的事件信息.在真实场景中,事件通常被定义为在连续一段时间内包含多种信息的异常,然而现有方法无法快速定位和充分获取这些异常.针对现有方法执行效率低、准确性差的问题,本文提出了一种基于可变多级时窗的时间序列流事件获取方法.具体来说,该方法首先使用中值滤波器对原始数据进行预处理,在一定程度上提高了事件获取的准确性;然后提出了一种基于短/长时窗平均值(STA/LTA)的事件触发算法来定位异常的触发点和终止点的近似范围;最后基于AIC(Akaike information criterion)法则对异常的起止点进行准确定位,从而获得异常的完整信息,即时间序列流事件.实验结果表明,与现有方法相比,该方法在执行效率和准确性方面具有显著优势.展开更多
储备池计算(Reservoir computing,RC)是一种高效处理时序数据的神经网络框架,属于递归神经网络的简化变体。与传统神经网络相比,RC仅需训练输出层的线性权重,从而显著降低计算复杂度,具有低训练成本和高计算效率,特别适合处理复杂的非...储备池计算(Reservoir computing,RC)是一种高效处理时序数据的神经网络框架,属于递归神经网络的简化变体。与传统神经网络相比,RC仅需训练输出层的线性权重,从而显著降低计算复杂度,具有低训练成本和高计算效率,特别适合处理复杂的非线性时间序列预测任务。本文将霍德里克-普雷斯科特(Hodrick-Prescott,HP)滤波器应用于分组储备池结构,并将储备池改进为循环跳跃储备池(Cycle reservoir with jumps,CRJ)结构,设计了基于HP滤波器的分组循环跳跃储备池(Grouped cycle reservoir with jumps based on HP filter,HP-GroupedCRJ)模型。该模型通过HP滤波器将复杂的输入信息分解成多个不同分量,分别输送到分组储备池结构中,使每个子储备池提取不同特征,从而增强模型处理复杂任务的性能。同时,储备池内部结构改进为固定循环跳跃形式,可以降低传统储备池结构因随机性导致的性能波动,显著提升模型结构的稳定性。在实验部分,本文对于模型的记忆容量进行了评估,并且在三种数据集(Sante Fe激光强度、共享单车租赁数量和每日心血管住院患者人数)开展时间序列预测实验。实验结果表明,HP-GroupedCRJ模型在预测方面显著优于其他比较模型,其归一化均方根误差(Normalized root mean square error,NRMSE)在所有数据集中均达最小值。展开更多
文摘Quantum theory according to the Copenhagen interpretation holds that, when a quantum interaction is observed (i.e., “measured”), the observer’s measuring devices temporarily become a part of the quantum system. Relativity theory holds that the event clock of the absorbed or emitted photon or graviton is frozen in time relative to all clocks outside the observed system. If we harmonize both theories, this would appear to imply that time continuity must be interrupted at each instant of observed photon or graviton interaction with matter. It is as if a segment of space-time is clipped out during each such observed interaction. If so, we must dispense with the notion of an absolutely smooth and continuous space-time and replace it with an observation-dependent, discontinuous, relativistic/quantum space-time. Mathematical physicists should be able to model this hypothesis (call it a “time-jump hypothesis”) and its inherent discontinuous space-time in their further efforts at unification.
文摘从时间序列流中获取事件是对时间序列流处理的基础.目前的研究大多采用传统的阈值确定方法对数据点进行查询,以获取时间序列流中存在的事件信息.在真实场景中,事件通常被定义为在连续一段时间内包含多种信息的异常,然而现有方法无法快速定位和充分获取这些异常.针对现有方法执行效率低、准确性差的问题,本文提出了一种基于可变多级时窗的时间序列流事件获取方法.具体来说,该方法首先使用中值滤波器对原始数据进行预处理,在一定程度上提高了事件获取的准确性;然后提出了一种基于短/长时窗平均值(STA/LTA)的事件触发算法来定位异常的触发点和终止点的近似范围;最后基于AIC(Akaike information criterion)法则对异常的起止点进行准确定位,从而获得异常的完整信息,即时间序列流事件.实验结果表明,与现有方法相比,该方法在执行效率和准确性方面具有显著优势.
文摘储备池计算(Reservoir computing,RC)是一种高效处理时序数据的神经网络框架,属于递归神经网络的简化变体。与传统神经网络相比,RC仅需训练输出层的线性权重,从而显著降低计算复杂度,具有低训练成本和高计算效率,特别适合处理复杂的非线性时间序列预测任务。本文将霍德里克-普雷斯科特(Hodrick-Prescott,HP)滤波器应用于分组储备池结构,并将储备池改进为循环跳跃储备池(Cycle reservoir with jumps,CRJ)结构,设计了基于HP滤波器的分组循环跳跃储备池(Grouped cycle reservoir with jumps based on HP filter,HP-GroupedCRJ)模型。该模型通过HP滤波器将复杂的输入信息分解成多个不同分量,分别输送到分组储备池结构中,使每个子储备池提取不同特征,从而增强模型处理复杂任务的性能。同时,储备池内部结构改进为固定循环跳跃形式,可以降低传统储备池结构因随机性导致的性能波动,显著提升模型结构的稳定性。在实验部分,本文对于模型的记忆容量进行了评估,并且在三种数据集(Sante Fe激光强度、共享单车租赁数量和每日心血管住院患者人数)开展时间序列预测实验。实验结果表明,HP-GroupedCRJ模型在预测方面显著优于其他比较模型,其归一化均方根误差(Normalized root mean square error,NRMSE)在所有数据集中均达最小值。
