Permutation-Based Pruning for Approximate K-NN Search

In this paper, we propose an effective indexing and search algorithms for approximate K-NN based on an enhanced implementation of the Metric Suffix Array and Permutation-Based Indexing. Our main contribution is to propose a sound scalable strategy to prune objects based on the location of the reference objects in the query ordered lists. We study the performance and efficiency of our algorithms on large-scale dataset of millions of documents. Experimental results show a decrease of computational time while preserving the quality of the results.