Algorithms and computational-architectures for change detections in large-scale data

在本研究中，我们研究了变化检测问题，即仅从观察到的时间序列数据中检测数据生成过程中的系统性变化的问题。特别是，我们开发了明确考虑数据规模的高效算法。为此，我们开发了一种基于最近提出的密度比估计技术的算法。并且，我们推导了该算法的在线实现，使得变化检测算法更加高效。由于变化检测是一种非常适合包括工程系统中的故障诊断在内的应用的技术，因此我们的结果作为此类应用的基础技术也将具有重要意义。

In this research, we study the change detection problem, i.e., the problem of detecting systematic changes in data generating processes only from observed time-series data. Especially, we develop efficient algorithms that explicitly take the scale of data into consideration. To this end, we developed an algorithm based on recently proposed density-ratio estimation techniques. And, we derived an online implementation of this algorithm, which make the change detection algorithm much more efficient. Since change detection is a technology well suited for applications including fault diagnosis in engineering systems, our results would be significant also as a fundamental technology for such applications.