Classification, Similarity Search, and Labeling of Time Series Data

1.开发了使用动态时间规整(DTW)距离的时间序列数据核。我们使用半定规划来保证核矩阵的正定性。我们使用时间序列分类和时间序列嵌入两个应用程序进行相似性搜索，以验证我们的方法。我们提出了HHCRF(分层隐藏CRF)。实验表明，随着训练集规模的增大和数据源的非马尔可夫性，HHCRF在状态序列估计方面的性能优于HHMM(Hierarchical Hidden MarkovModels)。

1. We have developed kernels for time series data using dynamic time warping (DTW) distances. We use semidefinite programming to guarantee the positive definiteness of a kernel matrix. We use two applications, time series classification and time series embedding for similarity search to validate our approach.2. We have proposed HHCRFs (hierarchical hidden CRFs). In the experiment, we show that HHCRFs perform better than HHMMs (hierarchical hidden Markov models) in state sequence estimation, as the training set size becomes larger and the data source becomes non-Markovian.