previous: connectionism. home: connectionism. next: connectionism: patterns.
each node in a connectionist network has an activation
level. this is generally a floating point number between
0 and 1. the activation level varies with time. a
connection (which, as you recall, is one-way) from node Alfred
to node Arnie has a weight attached to it. the weight is
another floating point number; this time it's generally
between -1 and 1. Arnie's activation level for the next
time tick is found using his activation from the current tick
and the product of Alfred's activation level and the
Alfred-to-Arnie connection weight.
(italics indicate subscript...) in a simple
connectionist model, Arnie's activation level is this:
aarnie = f(Saalfredwalfred->arnie)
..where a is activation level, f is some function, often
a function whose shape is sigmoidal, S is a sigma, and w
is a connection weight.