Real-time Online Learning for Event Streams

Abstract

A lot of interesting data comes in the form of event streams. Particularly for high volume streams, classical batch oriented approaches don't perform well. In this talk, I will discuss approaches for online learning and how to use techniques from stream mining like sketches and so-called heavy hitter algorithms to process large volume event streams with finite resources. Our experiences in particular with social media analysis lead to the development of streamdrill (http://streamdrill.com) and I'll discuss the major design decisions behind it.

Speaker Bio

Mikio Braun (http://blog.mikiobraun.de) is Chief Data Scientist and co-founder of TWIMPACT (http://twimpact.com), a startup focussing on real-time event analysis.