06.03.2013 Views

Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

Artificial Intelligence and Soft Computing: Behavioral ... - Arteimi.info

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

P4 = [ 0 1 1 1 0 0 1 ]<br />

Compute the next instances of each pattern by using the rules of binary<br />

Hopfield nets, set threshold = 0.<br />

Repeat the above process by replacing 0 in the patterns by –1 ; do you get<br />

any changes in the results?<br />

Can you call these 4 patterns stable states (states that do not change)? If<br />

so, in which of the two cases?<br />

Take any other arbitrary patterns, comprising of 1 <strong>and</strong> 0s. Are they also<br />

stable states?<br />

2. Suppose the 4 patterns are only given. Can you compute the weights in<br />

the<br />

network by using the formula given in the text?<br />

3. Show that the learning rule in Widrow-Hoff’s model, given by the<br />

following expression, is stable, when 0 < α < 2, unstable for α > 2 <strong>and</strong><br />

oscillatory when α = 2.<br />

Learning rule:<br />

∈<br />

k<br />

∆∈k<br />

∆ w = α k X / | X |<br />

k ∈ k k<br />

~<br />

; X = T<br />

d − k k k<br />

= − T<br />

X k ∆w<br />

~ ~<br />

.<br />

= −α<br />

∈<br />

k<br />

= −α<br />

∈<br />

k<br />

X<br />

~<br />

~ ~<br />

w<br />

k<br />

k<br />

/ |<br />

X<br />

~<br />

~<br />

where ∆Wk = change in weight vector, Xk = the input vector, whose<br />

components are x1, x2, …xn.<br />

[ Hints:<br />

So,<br />

Now, ( E –1 + α) ∈k = 0,<br />

or, ∈k = (1 - α) k .<br />

2<br />

|<br />

k<br />

Now, put 0 < α < 2, α > 2 <strong>and</strong> α = 2 to prove the results.]<br />

4. Derive the gradient descent learning rule for the following error function:<br />

E= (1 / 2) ∑ ( tk – Outk) 2 + ∑ wj i 2<br />

~<br />

2<br />

,

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!