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

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

18.2.3.1 Learning<br />

For adaptation of weights in the neural net, shown in fig.18.7, one simple<br />

method is to first map the patterns from different sentences onto the nodes of<br />

the neurons <strong>and</strong> then update the weights by employing the principles of bidirectional<br />

associative memory (BAM) [5]. For instance, suppose we have<br />

two sentences:<br />

1. A cat killed a rat.<br />

2. A man threw a stone.<br />

d0 d1 d2 d3 d4 d5 d6<br />

a0 a1 a2 a3 a4 a5 a6<br />

n0 n1 n2 n3 n4 n5 n6<br />

v0 v1 v2 v3 v4 v5 v6<br />

e0 e1 e2 e3 e4 e5 e6<br />

Start of the cat killed a rat End of<br />

Sentence (ART) (N) (V) (ART) (N) Sentence<br />

←activated node<br />

Fig. 18.7: A neural net with exhaustive connectivity (not shown for clarity)<br />

between nodes of each two successive layers.<br />

The activated nodes in the neural net of fig.18.7 for these patterns may<br />

be evaluated. The activated node i will have a node value ai=1 <strong>and</strong> the rest of<br />

the non-activated nodes in the net will have node value=-1. The weight wij<br />

from neuron ni to nj is now given by<br />

wij = ∑∀ pattern k (ai. aj)k (18.1)<br />

In this example we have only two patterns; so k has a maximum value 2. Thus<br />

18.2.3.2 Recognition<br />

wij = (ai.aj)1 + (ai.aj)2 (18.2)<br />

For recognizing the parts of speech of a missing word in a sentence, the<br />

following recognition rules may be employed.

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

Saved successfully!

Ooh no, something went wrong!