Associative-Projective Neural Networks (APNNs) - Neural Networks and Micromechanics

Biomedical Engineering Reference

In-Depth Information

j

j + k

j +2 k

00000

01000

00100

00010

00001

10000

00000

i

00000

01000

00100

00010

00001

10000

00000

i + k

00000

01000

00100

00010

00001

10000

i +2 k

01000

00100

00010

00001

10000

00000

Fig. 5.17 Matrix of synaptic weights

After this, the code's center of gravity again was calculated. The results for several

positions of the object are given in Table 5.1 . Thus, the coordinates of the center of

gravity of the code for the object located at the coordinates X = -20, Y = 10 are

determined as X cg = 26,66, Y cg = 11,26.

To determine the coordinates of an object using the position of center of gravity,

it is necessary to use the following formula:

X cg Þ;

Y cg Þ;

X

¼

2

ð

X cg

Y

¼

2

ð

Y cg

(5.14)

where X cg , Y cg are the coordinates of the center of gravity of the initial code

(object in the initial position), and X * cg , Y * cg are the coordinates of the center of

gravity of the code in the new position (after shift). Using the tabular values, we will

obtain X = -19,6, Y = 10,58, which represents the approximation of shift values.

In order to decode the name of the object, an additional associative field is

introduced. In this field, the ensembles corresponding to the separate features and

the names in the different positions are formed. The obtained code of the object is

applied to the additional associative field, where its name and features in the

appropriate position will be recognized. Then the codes of the recognized feature

are centered and input to the decoder.

Neural Networks and Micromechanics

Search WWH ::

Custom Search

Home