$29
1. [4 points] Briefly compare neurons used in a) perceptron-type networks and b) RBF-type networks. [Hint: concentrate on what they compare, how tot is calculated, and how neuron output is determined]
2. [6 points] Consider the modified Hebbian learning rule
w
new
= w
old
(1 − α)+ hx
o
j
ij
ij
i
and assume the following values xi = oj = 1, learning rate = 1, forgetting factor = 0.1, and initial
old
= 0
.
weight wij
a) Compare standard ( learning steps (i.e. plot
= 0) and modified Hebbian learning by plotting weights over 30 subsequent
new
as a function of time) using the parameters provided.
wij
b) Determine maximum value of weight w that can be obtained by modified Hebbian learning using the
parameters provided [Hint: in the limit case, the values of
w
new
ij
and
w
old
ij
would be identical].