Unsupervised Feature Learning from Time Series / 2322
Qin Zhang, Jia Wu, Hong Yang, Yingjie Tian, Chengqi Zhang
In this paper we study the problem of learning discriminative features (segments), often referred to as shapelets [Ye and Keogh, 2009] of time series, from unlabeled time series data. Discovering shapelets for time series classification has been widely studied, where many search-based algorithms are proposed to efficiently scan and select segments from a pool of candidates. However, such types of search-based algorithms may incur high time cost when the segment candidate pool is large. Alternatively, a recent work [Grabocka et al., 2014] uses regression learning to directly learn, instead of searching for, shapelets from time series. Motivated by the above observations, we propose a new Unsupervised Shapelet Learning Model (USLM) to efficiently learn shapelets from unlabeled timeseries data. The corresponding learning function integrates the strengths of pseudo-class label, spectral analysis, shapelets regularization term and regularized least-squares to auto-learn shapelets, pseudo-class labels and classification boundaries simultaneously. A coordinate descent algorithm is used to iteratively solve the learning function. Experiments show that USLM outperforms search-based algorithms on real-world time series data.