Add Completion and Some Robustness
- Add an extra link from each neuron to the one below it (remember
toroidal topology) at 0.7. Reduce the side excitatory links to 0.35
- This enables completion (as long as one element of the column is
active), but noise turns everything on.
- Replace the one link below with one above and one below with both
weights at 0.35. This gives some completion (down to 50/80)
and some robustness.
- Occasionally stable floating states also form.
- This is a really simple topology and it does several things that
we'd like a CA to do (e.g. completion, robustness, and persistence).
- However, it doesn't abide by several things we know neurons do
(e.g. Gray's Law, decay, fatigue, and shutting off).
- (Shutting Off: CAs are short term memory items that can
remain active after external stimulation has ceased, but
eventually become inactive. [Kaplan])
- (Fatigue may solve the shutting off problem.)
