PROJECT TITLE :

Efficient Algorithms for Kernel Aggregation Queries

ABSTRACT:

Kernel functions provide assistance for a wide variety of application types, including those that require activities such as density estimation, classification, regression, or the detection of outliers. When dealing with these types of tasks, one of the online operations that is frequently performed is to compute the weighted aggregation of kernel function values with respect to a set of points. Nevertheless, scalable aggregation methods for typical kernel functions (such as the Gaussian kernel, polynomial kernel, sigmoid kernel, and additive kernels) and weighting schemes have not yet been discovered. In this paper, we present a novel and efficient bounding technique that leverages index structures in order to speed up the computation of kernel aggregation. The technique can be found in the introduction. In addition to this, we extend our method to additive kernel functions such as the JS and Hellinger kernels, the 2 kernel, the intersection kernel, and the JS kernel, which are all widely used in a variety of fields such as computer vision, medical science, and geoscience, amongst others. In order to manage the additive kernel functions, we have further developed the novel and efficient bound functions in order to evaluate the kernel aggregation in the most time-effective manner. Experimentation and research on a large number of real datasets have shown that our proposed solution, KARL, is at least one order of magnitude faster than the current state of the art when it comes to the execution of various kernel function types.