Neurocomputing in Space - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 120

Fig. 6.1 a Training patterns mapping shows scaling by factor 1 / 2, angle of rotation ˀ/ 2and
displacement by (0

, , .

0 0 2); b The testing over a 3D object

Example 6.1 A neural network based on 3D vector-valued neurons has been trained
for the composition of all three transformation. Training input-output patterns are
shown in Fig. 6.1 a. The generalization ability of such a trained network is tested over
sphere (1681 data pints). Figure 6.1 b presents the generalization ability of trained
network. All patterns in output are contracted by factor 1

/ 2, rotated over ˀ/ 2 radians

clockwise and displaced by . This example demonstrate the generalization

ability of considered network in interpretation of object motion in 3D space.

( 0 , 0 , 0 . 2 )