When: November 12, 2009   6:45 pm -  7:45 pm

Some clustering algorithms return a locally optimal solution for a particular (often random) starting point. To avoid accepting a bad clustering derived from a poor starting point, one can run the clustering algorithm a large number of times and then use the resulting solutions to arrive at a single, robust clustering. This process is known as consensus clustering. This talk will briefly cover the idea of clustering and then develop a process for consensus clustering the results from multiple runs of the non-negative matrix factorization algorithm. Results from using this methodology on North Carolina county-by-county election data will be shared.

We will gather in Mathematics Hall at 6:30 PM for our traditional light snacks.

Contact: Prof. Donna Molinek