500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500

 5 R i u, p ic', I i ,

 0 I0I I 1 0 20I
 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500

 S1 5

 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500




 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500


Figure 5-6 Results: (a) Flow: desired in blue, predicted in red; (b) MSE criterion, full line
 is for expert 1, dotted line is for expert 2; (c) Winner; (d) Pressure.

 5.2 Results Comparison and Analysis

 First an important note to be made is that the results of the simulations, in this

chapter and the previous one, were somewhat erratic and not always consistent: for

example sometimes one type of expert performed really well and other times it exhibited

trends where one expert's criterion. Also, if the initial values of the networks are not

"chosen" to be near the desired ones, oftentimes the networks do not converge. Still the

results are encouraging for online segmentation from competitive experts: they have had

only one presentation of the data set, whereas the framework called for several epochs,

and segmentation is performed accurately for the three models.

5.2.1 Comparison of the Models

 RBFNs and MLPs are both examples of feedforward neural networks, and the

difference between them lies in the local versus global basis for approximation of the