Large Scale KNN-Graph Approximation
Mohamed Riadh Trad, Alexis Joly, Nozha Boujemaa · 2012
Efficiently constructing the K-Nearest Neighbor Graph (K-NNG) of large and high dimensional datasets is crucial for many applications with feature-rich objects, such as images or other multimedia content. In this paper we investigate the use of high dimensional hashing methods for efficiently approximating the K-NNG in distributed environments. We first discuss the importance of balancing issues on the performance of such approaches and show why the baseline approach using Locality Sensitive Hashing (LSH) does not perform well. Our new KNN-join method is based on RMMH, a recently introduced hash function family based on randomly trained classifiers. We show that the resulting hash tables are much more balanced and that the number of resulting collisions can be greatly reduced without degrading quality. We further improve the load balancing of our distributed approach by designing a parallelized local join algorithm. We show that our method outperforms state-of-the-art in centralized settings and that it is efficiently scalable given its inherently distributed design. We finally present a distributed implementation of our method using a MapReduce framework and evaluate its performance on a large dataset.