Extensions of self-organizing maps

We present two novel methods we developed that take advantage of the large power of the Self-Organizing Map (SOM) technique developed by Teuvo Kohonen. SOM is an unsupervised neural network mapping a set of n-dimensional vectors to a two-dimensional topographic map. We first present a method that combines the analytic SOM with the scatter plot into an interpolated model. Second, we introduce an interactive technique that intelligibly organizes occluded or overlapped points, a self-organizing map-based Smart Jittering algorithm. Large and high-dimensional data sets mapped to low-dimensional visualizations often result in perceptual ambiguities. We address this ambiguity and present a method to systematically organize the occlusion.