A16: Neural Networks

As in Activities 14 and 15, we implement object classification but this time with the use of neural networks. Using the Artificial Neural Networks (ANN) toolbox for Scilab, we attempt classification (as in Mr. Cole Fabros' code) of the objects used in the previous classification activities.

Figure 1.

Figure 2.

Figure 3. Left to right: learning rate = 0.01, 0.1, and 0.5.

Figure 3 shows how in the training set, increasing the learning rate also increase the disparity between the two classes. So, we can expect that learning rate would also increase correct classification. However, it did not help in this data set.

Having three different methods fail at giving 100% correct classification, we realize that feature extraction is crucial in classification. No matter how good the algorithm is, if the features do not distinguish the classes well, correct classification cannot be assured.

I give myself a grade of 9 because the classification implemented is better than random classification.

I would like to thank Mr. Cole Fabros for the Scilab code from which the whole class based scripting.