2018/10/23 by Abubakar Abid, James Zou, Abid, Abubakar +1 · 1 citation
Computer Science · #Advanced Text Analysis Techniques #Anomaly Detection Techniques and Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Music and Audio Processing #Time Series Analysis and Forecasting
paper · pdf · doi:10.48550/arxiv.1810.10107
openalex publication_date 2018/10/23 · openalex created_date 2022/08/02 · openalex updated_date 2026/07/28
Measuring similarities between unlabeled time series trajectories is an\nimportant problem in domains as diverse as medicine, astronomy, finance, and\ncomputer vision. It is often unclear what is the appropriate metric to use\nbecause of the complex nature of noise in the trajectories (e.g. different\nsampling rates or outliers). Domain experts typically hand-craft or manually\nselect a specific metric, such as dynamic time warping (DTW), to apply on their\ndata. In this paper, we propose Autowarp, an end-to-end algorithm that\noptimizes and learns a good metric given unlabeled trajectories. We define a\nflexible and differentiable family of warping metrics, which encompasses common\nmetrics such as DTW, Euclidean, and edit distance. Autowarp then leverages the\nrepresentation power of sequence autoencoders to optimize for a member of this\nwarping distance family. The output is a metric which is easy to interpret and\ncan be robustly learned from relatively few trajectories. In systematic\nexperiments across different domains, we show that Autowarp often outperforms\nhand-crafted trajectory similarity metrics.\n