Kernel methods have emerged as a versatile mechanism for data classification, clustering and pattern recognition. Understanding of the mapping performed by a kernel simplifies the choice and the design of kernels as well as the fine-tuning of kernel parameters, thus improving the effectiveness of kernels in specific applications.
In this work, the authors present a novel technique able to map data from a kernel defined feature space to a visual space. The technique, named Kernel-based Linear Projection (Kelp), has low computational cost and enables interactive resources for users to dynamically interact with the resulting layout. These desirable properties render Kelp an attractive visualization tool in different scenarios, such as interactively changing the position of sample points in the visual space to realize the SVM mental model.
A. Barbosa, F. V. Paulovich, A. Paiva, S. Goldenstein, F. Petronetto, and L. G. Nonato. 2016. Visualizing and Interacting with Kernelized Data. IEEE Transactions on Visualization and Computer Graphics 22, 3 (March 2016), 1314-1325. 10.1109/TVCG.2015.2464797