...constant:
In the low-dimensional LISSOM, a more complex length normalization is necessary to avoid favoring long input vectors [15]. As the dimensionality of the input increases, the necessity for strict length normalization decreases, and the sum-of-weights normalization gradually becomes a valid approximation.

...retina.
This rather remote possibility is similar to the input abstraction in LISSOM and in self-organizing maps.

Joseph Sirosh Tue Mar 14 16:56:39 CST 1995