By Dr. Nikolaos A. Laskaris

The term ‘‘pattern’’, currently, encompasses the notion of a variety of data-forms the machines have to tackle with. Despite the fact that in early days it was used mostly for pictorial information, i.e. 2D-signals, now the same term stands almost for any output from a data-source. For instance, any digital-signal can be considered as an 1D-pattern, a grey-scale image as a 2D-patterm, a video-sequence as a (temporal) multi-dimensional pattern etc.

Here the author presents a general purpose framework for dealing with patterns and discuss simple algorithms with a wide-range of applications (from novelty-detection and prototyping in databases to the full organization of a library of patterns). The main characteristic of the framework and simultaneously its great benefit is its Geometrical character. This enables the direct conceptualization of the employed ideas and promotes the easy understanding of the described algorithmic steps.

The Tutorial includes the following sections:

  • Relating topological descriptors of point sets with the data.
  • Unmasking Outliers
  • Hierarchical Clustering
  • Partitioning Clustering Algorithms
  • Subtractive Clustering
  • Multidimesional Scaling
  • Manifold Learning & ISOMAP
  • Minimal Spanning Tree
  • Vector Quantization

The tutorial is available here in pdf format.

Further details including Matlab M-Files and exemplary data sets may be found at the author’s home page.

Post to Twitter Post to Yahoo Buzz Post to Delicious Post to Digg Post to Facebook Post to Google Buzz Post to LinkedIn Post to Slashdot Post to StumbleUpon Post to Technorati