CINF 20 |
| Spectral clustering utilizes matrix decompositions to transform a dataset of n-dimensions to a lower dimensional subspace within which clustering can be performed. The most common decomposition used is the SVD and it has been shown that the SVD of a data matrix represents a clustering. We investigate the use this approach in the clustering of an Ames mutagenecity dataset and an aqueous solubility dataset. We also investigate the use of the fast SVD algorithm which approximates the SVD of a matrix. Our results indicate that the approximation algorithm leads to an order of magnitude speedup. Furthermore the clustering results are similar to those obtained using traditional patritional clustering algorithms. |
|
Chemistry Applications Involving Data Analysis and Visualization
9:00 AM-11:30 AM, Monday, March 26, 2007 Hyatt Regency McCormick -- 12 C/D, Oral
Division of Chemical Information |