On applications of parameterized hyperplane partitioning

The efficient similarity search in metric spaces is usually based on several low-level partitioning principles, which allow filtering of non-relevant objects during the search. In this paper, we propose a parameterizable partitioning method based on the generalized hyperplane partitioning (GHP), which utilizes a parameter to adjust "borders" of the partitions. The new partitioning method could be employed in the existing metric indexes that are based on GHP (e.g., GNAT, M-index). Moreover, we could employ the parameterizable GHP in the role of a new multi-example query type, defined as a partition determined by an available query object and several "anti-example" objects. We believe that both applications of parameterizable GHP can soon take place in metric access methods and new query models.