|Math Coffee: "New Patterns with a Paradigm Shift in Data Clustering" Colin Thomson '13
When: September 09,
Speaker: Colin Thomson '13Ticket Required: No
There are hundreds of clustering algorithms, each with its own strengths and weaknesses and each with an appropriate application. One approach is aggregate clustering, which combines multiple clustering methods to give a single, more conservative clustering. However, many some of the most important patterns in a data set are not found within the clusters, but within which data points were difficult to cluster. By taking advantage of the fact that different algorithms place different items in different clusters, the aggregate approach can reveal those items that are difficult to cluster.
This summer I designed an applet in MATLAB to view these "fringe" items, complete with a tool to identify which specific items had been placed in different clusters by a certain number of algorithms. As an example, I investigated handwriting samples, as it proved to illustrate the potential of the approach very well. The potential for biological data sets is also presented.
We will gather in Math Hall at 6:30 PM for light snacks.
Contact: Prof. Tim Chartier