7–9 November 2016: Conference & Tutorials
9–10 November 2016: Training
Amsterdam, The Netherlands

Anomaly detection: A hateful lovestory

Pieter Buteneers (CoScale)
16:10–16:50 Monday, 7/11/2016
Metrics/monitoring Automation Auditorium (Ground + Balcony) Audience level: Beginner
Average rating: ***..
(3.30, 20 ratings)

Prerequisite knowledge

  • A basic understanding of server monitoring

What you'll learn

  • Gain a new perspective on anomaly detection
  • Learn some dos and don'ts to get started on your own anomaly detection system

Description

Anomaly detection. Many people speak about it, some use it, some build it, some love it. . .but many hate it. Pieter Buteneers offers an overview of this strange beast and explains why you should use it.

Pieter shares his experiences developing an anomaly detection system for CoScale, outlining the methods he used, tested, and dismissed and covering what worked, what didn’t, and why. Pieter explores in detail the difference between statistical significance and relevance, why the univariate approach is the way to go, why most multivariate techniques don’t work, and why you should keep it simple, stupid.

Some of you will try more complex techniques. And you should. But not before you know how to use them without them misleading you.

Photo of Pieter Buteneers

Pieter Buteneers

CoScale

Pieter Buteneers is a data strategist and machine learning consultant at CoScale. Pieter is relatively new to the DevOps field. Previously, he was a postdoc at Ghent University, where he did research on AI, machine learning, and deep learning. Pieter has also given a TEDx talk (on a somewhat unrelated subject).