CS Faculty Candidate Talk
Tuesday, March 23rd
10:30am,  ST2 Room 430a

Similarity Search and Indexing Techniques for Multidimensional Time-Series

Michalis Vlachos
Computer Science Department
UC Riverside

In this talk we will investigate techniques for estimating the similarity between multidimensional time-series and, in particular, data trajectories. Object trajectories are very prevalent nowadays, especially in environmental applications, animal mobility experiments, video tracking/surveillance data, motion capture data etc. The discovery of objects with specific motion patterns is a challenging task, therefore our similarity model must be robust to noise and support elastic and imprecise matches. Our primary similarity measure is based on the Longest Common Subsequence (LCSS) paradigm that offers enhanced robustness. However, the index that we employ is also able to accommodate other distance measures as well, including the ubiquitous Euclidean distance and the increasingly popular Dynamic Time Warping. A major contribution of this work is the ability to support all these measures into a single index without any need for reconstruction or adjustment. The proposed framework guarantees no false dismissals and can also be tailored to provide much faster response time at the expense of slightly reduced precision/recall. Finally, we will demonstrate the high applicability of the described techniques on many real world problems, such as motion-capture (MOCAP) matching, OCR recognition, image classification etc. (This work has been published in IEEE ICDE'02, ACM SIGKDD'03, ACM SIGGRAPH'03) Michalis Vlachos is a PhD candidate and a member of the Database Lab at University of California, Riverside. His research interests expand on the areas of data-mining, databases, time-series, clustering & classification of multimedia data. He has received his BS in Informatics with Highest Honors from Aristotle University in Greece, and he is a recipient of the Fulbright Foundation scholarship for graduate studies.