Mining Useful Patterns
by Dr. Jilles Vreeken
Abstract: Pattern mining is a powerful tool for exploratory data analysis, aimed at identifying interesting local structure. While highly promising, the traditional approach does typically not provide very useful results: by answering the question 'find me all potentially interesting patterns' typically far too many results are returned - and many of which will be redundant.
Instead, in this talk I will show that you should ask for the set of patterns that describes your data best. That is, to use information theory to identify the optimal set of patterns. This set has many desirable properties: it is small, captures the most important structure in your data, while being neither redundant nor overfit. Moreover, these patterns are useful. As an example, I will discuss a wide range of data mining tasks, include classification, one-class classification, anomaly detection, missing value estimation, and clustering, in which these patterns have been shown to obtain top-notch and highly interpretable results, without the need of any parameters.
Bio: Jilles Vreeken is a post-doctoral
researcher at the
In 2009, he defended his PhD thesis 'Making
Pattern Mining Useful' at the