Incremental Pattern Discovery on Stream, Graphs and Tensors
by Jimeng SUN

Abstract

Data stream has received tremendous attention in recent years due to numerous emerging applications such as network traffic monitoring, environmental sensor network, web click stream mining, social network analysis, financial fraud detection. Many challenges arise in analyzing and mining streams: How to find patterns (main trends, clusters, anomalies) in real-time? How to summarize/compress data streams incrementally? How to efficiently update the old patterns when new data arrive?

In this talk, we first present a powerful data model tensor stream (TS), which covers diverse data formats including traditional stream/time series, time-evolving graphs, and data cubes. Second, we present an online algorithmic framework for TS, namely, incremental tensor analysis. We discuss three different variants of incremental tensor analysis and the tradeoff among them. Finally, we show several success stories of the proposed methods in real applications including social network analysis, network anomaly detection and data center monitoring.

About the speaker

Jimeng Sun is a PhD candidate in Computer Science Department at Carnegie Mellon University. Previously, Jimeng obtained a B.S. and Master in computer science from Hong Kong university of Science and Technology in 2002 and 2003. His research interests include data mining on streams, graphs and tensors, anomaly detection and spatio-temporal databases. He has been actively applying data mining techniques for different applications such as environmental sensor monitoring, financial fraud detection, data center monitoring, network anomaly detection.