•   When: Wednesday, February 17, 2016 from 11:00 AM to 12:00 PM
•   Speakers: Jessica Lin
•   Location: Nguyen Engineering, Room 4201
•   Export to iCal

Abstract

Massive amounts of data are generated daily at a rapid rate. As a result, the world is faced with unprecedented challenges and opportunities on managing the ever-growing data, and much of the world's supply of data is in the form of time series. Time series data mining has thus attracted an enormous amount of attention from researchers and practitioners in the past two decades. This talk will focus on the discovery of novel and non-trivial patterns in time series data, including frequently encountered patterns (motifs) and rare patterns (anomalies). The ability to efficiently detect frequent and anomalous patterns in time series allows for the exploration, summarization, and compression of data. In addition, such information is crucial to a variety of application domains where these patterns convey critical and actionable information. In recent work, we demonstrate that grammar induction, the process of learning rules of a formal language from a set of observations, allows the discovery of hierarchical structures and regularities from input time series. We propose several algorithms based on grammar for efficient discovery of co-existing variable-length approximate motifs and anomalies without any prior knowledge about their length, shape, or minimal occurrence frequency. Based on time series motifs, we further propose an algorithm to identify class-specific representative patterns for time series classification. The key motivation is that the identication of a small set of distinctive and interpretable patterns of each class allows us to exploit their key characteristics for discriminating against other classes. We present GrammarViz, an interactive tool for grammar-driven mining and visualization of variable-length time series patterns.