LLE Based Pivot Selection for Similarity Search of Biological Data

Distance-based indexing is a widely used technique for general purpose search. Pivot selection is the most crucial step of bulkloading a metric-space indexing tree. Current pivot selection methods are mainly based on linear methods. A non-linear method based on Locally Linear Embedding is proposed. Empirical results demonstrate that the performance of new method is superior to existing methods. Keywords-similarity search; metric-space indexing; pivot selection; locally linear embedding; dimensional reduction