Querying time-series streams

Index trees created using distance based indexing are difficult to maintain online since the distance function involved is often costly to compute. This problem is intensified when the database we are dealing with, is frequently updated, as only limited time is available to perform the maintenance. In this paper, we propose a novel tree maintenance mechanism for the problem of answering approximate k-Nearest Neighbor queries with a probabilistic guarantee on timeseries streams. When the underlying data change, we may choose to defer updating the tree as long as the probabilistic guarantee of answering queries is high. To prolong such deferment, we present innovative techniques that maintain the utility of the tree by migrating its pivots and by partially reconstructing it. As the probabilistic guarantee decays with time and crosses the minimum guarantee threshold, all of the deferred updates are performed. In essence, our work offers an elegant compromise between the accuracy guarante...

• Type: Conference
• Year: 2008
• Where: EDBT
• Authors: Vivekanand Gopalkrishnan
• Keywords: Accuracy Guarantee, Database, EDBT 2008, Minimum Guarantee Threshold, Probabilistic Guarantee Decays