In this study, we present a novel ranking model based on learning the nearest neighbor relationships embedded in the index space. Given a query point, a conventional nearest neighbor search approach calculates the distances to the cluster centroids, before ranking the clusters from near to far based on the distances. The data indexed in the top-ranked clusters are retrieved and treated as the nearest neighbor candidates for the query. However, the loss of quantization between the data and cluster centroids will inevitably harm the search accuracy. To address this problem, the proposed model ranks clusters based on their nearest neighbor probabilities rather than the query-centroid distances to the query. The nearest neighbor probabilities are estimated by employing neural networks to characterize the neighborhood relationships as a nonlinear function, i.e., the density distribution of nearest neighbors with respect to the query. The proposed probability-based ranking model can replace the conventional distance-based ranking model as a coarse filter for candidate clusters, and the nearest neighbor probability can be used to determine the data quantity to be retrieved from the candidate cluster. Our experimental results demonstrated that implementation of the proposed ranking model for two state-of-the-art nearest neighbor quantization and search methods could boost the search performance effectively in billion-scale datasets.READ FULL TEXT
We present PUFFINN, a parameterless LSH-based index for solving the
We study clustering algorithms based on neighborhood graphs on a random
Many algorithms for approximate nearest neighbor search in high-dimensio...
This paper is concerned with nearest neighbor search in distributional
The reverse k-nearest neighbor (RkNN) query is an established query type...
Hash based nearest neighbor search has become attractive in many
In this paper, we propose a new data structure for approximate nearest