计算数据流上的小波密度估计

COMPUTING WAVELET DENSITY ESTIMATORS OVER DATA STREAMS

许多科学和商业应用需要对瞬时数据流做立即分析,许多分析技术的核心操作是使用密度估计寻找未知的连续数据分布,近期的研究证明基于小波的方法优于传统的密度估计技术。由于计算资源的限制,基于小波的方法无法直接用于无限的数据流。提出一种数据流上计算概率阈值小波密度估计的新方法,只需要确定数量的内存,能以在线的方式更新估计,实验证明了方法的有效性。

Many scientific and commercial applications rely on an immediate analysis of transient data streams. A core operation of many analysis techniques is the exploration of an unknown continuous data distribution by using density estimation. A recently examined approach based on wavelets promises to be superior to traditional density estimation techniques. For possibly infinite data streams, however, this approach is not feasible due to the limited resources, e. g. memory. A new technique for computing wavelet density estimators with probabilistie threshold over data streams that only requires a fixed amount of memory is proposed in this paper. Our estimators are updated in an online manner such that a continuous analysis of data streams is supported during runtime. A thorough experimental evaluation demonstrates the efficacy of our estimators.