SGER: Efficient Support for Mining Queries in Data Stream Management Systems

University Of California-Los Angeles, Los Angeles CA

Investigators

Abstract

Efficient support for continuous queries on massive data streams is critical in many application areas, including publish/subscribe, traffic monitoring, sensor networks, and algorithmic trading. Thus, designing data stream management systems (DSMS) to support such queries efficiently and reliably represents a vibrant area of current research. Unfortunately, DSMS cannot yet support efficiently the very complex mining queries required to extract new patterns and knowledge from data streams-- although they are needed in important applications, such as intrusion detection and other security tasks. DSMS designed to support mining tasks are called Inductive DSMS (since they induce new knowledge from data). This project's objective is to develop the enabling technology for Inductive DSMS by (i) designing faster data stream mining algorithms, and (ii) extending DSMS to support efficiently mining tasks expressed in the DSMS query language, and (iii) building an Inductive DSMS prototype and evaluating it on data mining testbeds. Data mining technology is having a major impact on diverse applications domains, including business, security, and science. However, mining the massive data streams that represent the lifeblood of the information age has proven very difficult: this is the first project addressing this challenge. A broad range of scientific, educational, and economic activities will benefit greatly once the vision of Inductive DSMS becomes reality. Project funds will support PhD students pursuing research on DSMS and data mining. The new technology will enrich several graduate courses. Dissemination is through publications, reports, and demos available from: http://wis.cs.ucla.edu/idsms.