Neural-Network Enhanced Visualization of High-Dimensional Data
Urška Cvek, Marjan Trutschl, John Cliffor · InTech eBooks · 2010
We provide a description of an approach for intelligent spatial placement of highdimensional records, based on a modified Kohonen SOM algorithm. SOM-augmented visualizations provide for increased visual scalability and offer higher intrinsic dimension than the classic visualizations. The resulting mapping efficiently alleviates crowding and occlusion, and emphasizes the relationships among the neighbouring multi-dimensional records. Utilizing these techniques, the user can efficiently approach perceptual ambiguities associated with occlusion and gain insight into multi-dimensional data sets using an informative visualization, instead of a series of linked visualizations. These algorithms were tested on high-dimensional biomedical data sets and have provided for meaningful associations in an interactive environment.