Abstract

The talk provides a quick introduction to streaming and online data mining algorithms. These algorithms are required to summarize, process, or act upon an arbitrary sequence of events (data records). At every point in time, future events/data are unknown and past event are too numerous to store. While this computational model is severely restricting, it is, de facto, the working model in many large scale data systems. This talk introduces some classic and some new results in the field and show how they apply to email threading, news story categorization, clustering, regression, and factor or principal component analysis.