Using genetic algorithm to balance the D-index algorithm for metric search

The Distance Index (D-index) is a recently introduced metric indexing structure which capable of state-of-the-art performance in large scale metric search applications. In this paper we address the problem of how to balance the D-index structure for more efficient similarity search. A group of evaluation functions measuring the balance property of a D-index structure are introduced to guide the construction of the indexing structure. The optimization is formulated in a genetic representation that is effectively solved by a generic genetic algorithm (GA). Compared with the classic D-index, balanced D-index structures show a significant improvement in reduction of distance calculations while maintaining a good input-output (10) performance.