Educational Interactive Software as a Support to the Teaching of Artifical Neural Network Methodology Applied to a Classification Problem
Marko Čupić, Jan Šnajder, Bojana Dalbelo Bašić · 2003
As artificial neural networks (ANNs) become increasingly popular, significant efforts are made to teach students this attractive paradigm. Today, teaching artificial neural networks is inevitable in machine learning courses. Having this in mind, we designed an educational interactive software system to foster students’ understanding of the topic. By developing this system, we were guided by the following objectives: (i) To demonstrate how artificial neural network works ; (ii) To enable experimentation with different test samples ; (iii) To illustrate training of the neural network using backpropagation algorithm ; (iv) To exemplify how weights obtained from different training phases influence the network performance ; (v) To show the robustness of the neural network approach. In order to fulfill these objectives, we have designed ie-ABC (interactive educational ANN-based Banknotes Classifier), a system that deals with a rather complex task of banknotes classification. There are two components to the system: a neural network training module and a neural network demonstration module, both designed with a user-friendly graphical interface. The training module enables students to study the training process and repercussion of different parameter values. The demonstration module, being composed from a neural network simulator and a sample generator, allows the processing of input samples by previously trained network, generation of sample damage and mutilation and network performance examination during stages of underfitting, overfitting and optimal training. It is our experience, based on the usage of this system in the educational process that the system contributes to students’ better understanding. It also encourages them to deeper exploration of the machine learning techniques.