Abstract
The topology-preserving representation of a rectangular part of space onto a square network of formal neurons is studied using the Kohonen algorithm. Linear stability analysis shows that there is a critical ratio for the sides of the rectangle. For larger ratios the map becomes unstable. The value of the critical ratio depends on the actual shape of the adjustment function. The problems cannot be scaled away in the case of inhomogeneously sampled input space. The results of the analysis are compared with computer simulations.