Efficient Context-Aware K-Nearest Neighbor Search.

We develop a context-sensitive and linear-time K-nearest neighbor search method, wherein the test object and its neighborhood (in the training dataset) are required to share a similar structure via establishing bilateral relations. Our approach particularly enables to deal with two types of irregularities: (i) when the (test) objects are outliers, i.e. they do not belong to any of the existing structures in the (training) dataset, and (ii) when the structures (e.g. classes) in the dataset have diverse densities. Instead of aiming to capture the correct underlying structure of the whole data, we extract the correct structure in the neighborhood of the test object, which leads to computational efficiency of our search strategy. We investigate the performance of our method on a variety of real-world datasets and demonstrate its superior performance compared to the alternatives.