UNDERTOW: Multi-Level Segmentation of Real-Valued Time Series

The discovery of meaningful change points, finding segments, in both categorical and real-value data time series is a well-studied problem. Prior segmentation algorithms and tasks operate under overly restrictive assumptions (e.g., a priori knowledge of the number of segments, trivial inputs) and in singular domains (e.g., finding common regions in images, speaker change detection). We introduce a domain-independent algorithm, UNDERTOW, which discovers segment boundaries in real-valued time series and constructs hierarchies of segments to form macro segments.

• Type: Conference
• Year: 2007
• Where: AAAI
• Authors: Tom Armstrong, Tim Oates
• Keywords: AAAI 2007, Data Time Series, Intelligent Agents, Meaningful Change Points, Time Series