daren@umbc5.umbc.edu (10/29/90)
I am attempting to implement the self-organizing map neural nets described in Ch. 5 of Neurocomputing (Robert Hecht-Nielsen). Most of the details are pretty straight forward, but I am confused by the description of training these nets. The book states: At the beginning of each trial, a point (u,v) is selected at random. The feeler mechanism is then moved to this position ... The joint angles THETA and PHI are then read out and supplied to all processing elements ... What type of values do (u,v) take on? Are they essentially just the 2 angle measures that define the point at (u,v) (i.e. (u,v) = (THETA, PHI))? Though this would go against the definition of the mapping done by self-organizing maps, whereby it maps elements of B (THETA, PHI) contained in N-dimensional space, into elements of C (u,v) in M-dimensional space, by means of Y examples in C. As anyone can tell, I am quite confused! Any help would be greatly appreciated! Thanks, Daren