An eigenvalue-based pivot selection strategy for efficient indexing and searching in metric spaces

Pivots are used widely during indexing and searching in metric spaces. We maintain the distances from pivots to data objects to be indexed so the pre-computed distances can be used to prune unpromising objects during the search process. The search efficiency depends on the pivots used, but choosing good pivots is a challenging task. In this paper, we propose a new pivot selection method that incrementally chooses pivots using an eigenvalue-based uncorrelatedness scoring function. We also present a GPU implementation for computing the uncorrelatedness score in order to accelerate the pivot selection process. Our experimental results demonstrated that the proposed method performed better than other previously described pivot selection methods.