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.

Here W T is a transposed weight matrix, whose (i, j)th element denotes the<br />

weight from transition tri to place pj. ‘Th’ is the threshold vector, the i-th<br />

component of which is a scaler denoting the threshold of transition tri. ‘U’ is a<br />

unit step vector, which applies step function to all its arguments. ‘N’ is the<br />

belief vector, whose i-th component is a scaler, representing the fuzzy belief of<br />

proposition di. ‘T’ is the fuzzy truth token (FTT) vector, whose i-th component<br />

is a scaler denoting the FTT of transition tri. ‘Q’ is a binary place to transition<br />

connectivity matrix, whose (i, j)-th element denotes the existence (or no<br />

existence) of connectivity from the place pj to transition tri.<br />

A new weight adaptation rule, presented below, is used for the encoding of the<br />

weights:<br />

dWij<br />

----- = - α Wij + ti (t) . nj (t) (16.16)<br />

dt<br />

where ti(t) <strong>and</strong> nj (t) denote the FTT <strong>and</strong> the belief of the transition tri <strong>and</strong> the<br />

place pj respectively. The above equation can be discretized into the following<br />

form<br />

Wij (t+1) = (1- α) Wij + ti(t) . nj(t). (16.17)<br />

The vector-matrix form of the above equation is given by<br />

W (t+1) = (1 - α) W (t) + [ ( N (t) . T T (t)) ^ P ] (16.18)<br />

where P is a binary connectivity matrix from transition to places in a FPN.<br />

The recall model, represented by expression 16.14 <strong>and</strong> 16.15, is stable, when<br />

N(t+1) = N (t) = N* at t = t* , the equilibrium time.<br />

Now, the above relation holds good for ( Q o N* C ) C > Th if<br />

N* ≥ [ W T o ( Q o N* C ) C ]. (16.19 )<br />

The steady-state values of the components of N* thus can be easily<br />

evaluated, with the known values of the Q <strong>and</strong> W matrix.<br />

The conditional convergence: 0 < α

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

Saved successfully!

Ooh no, something went wrong!