| Demo 15.7. Self-organization in LISSOM and GLISSOM.
This animation of Figure 15.7 shows
that self-organizing a gradually growing map using GLISSOM matches
ordinary self-organization in LISSOM. The OR preferences of the
neurons are color coded using the key on top.
The GLISSOM map starts with 36 × 36 neurons and is gradually scaled to
144 × 144, i.e. the same size as the LISSOM map. At each iteration,
the features that emerge in the GLISSOM map are similar to those of
LISSOM, except for discretization differences (Figure 15.9 shows that results match
even more closely when larger initial networks are used). GLISSOM can
therefore be used to reduce the memory and CPU requirements of
self-organization simulations, making it possible to simulate very
large networks, such as the entire human V1, at the columnar level.
Previous demo
|
