Recursive learning rules for SOMs

Applications of Different Self‐Organizing Map Variants to Geographical Information Science Problems

“…This resembles the way nodes are pulled in a fisherman's net, and thus this update rule was dubbed 'fisherman's rule'. It has been shown (Lee et al, 2001) that this will improve the convergence speed in the first iterations of the learning phase. The main reason for this is that, at each learning step, the different units will not be attracted in exactly the same direction (the direction of the input pattern), but will instead be pulled in a direction that depends on their immediate neighbors.…”

“…One alternative is to move the unit in the direction of the nearest unit, as was proposed by Lee et al (2001). This resembles the way nodes are pulled in a fisherman's net, and thus this update rule was dubbed 'fisherman's rule'.…”

“…The main reason for this is that, at each learning step, the different units will not be attracted in exactly the same direction (the direction of the input pattern), but will instead be pulled in a direction that depends on their immediate neighbors. It is suggested (Lee et al, 2001) that the fisherman's rule be used in the first iterations of the training algorithm, being replaced by the standard rule in the last iterations.…”

Parallel Programmable Asynchronous Neighborhood Mechanism for Kohonen SOM Implemented in CMOS Technology

“…Here we present selected results for 8 × 8 and 16×16 neurons for two example 2-D data sets. The results for 2-D sets have been selected for a better illustration [8], [21], [24], [26]. In both sets, data are divided into P classes (centers), where P equals the number of neurons in the map.…”

Hierarchic Fault Diagnosis by Pattern-Recognition Approaches Applied to DAMADICS Benchmark

“…PCA is an optimal linear dimension reduction technique (in the meansquare sense); LDA finds the linear projection that maximizes the separation between classes, relative to the dispersion of data within classes. CCA (Demartines et al, 1997, Lee, 2000, 2002, on the contrary, is relatively recent. It may be seen as a non-linear generalization of PCA, but is actually a topology preserving method: it searches for vectors in the transformed space that reproduce the distances in the input space.…”