Tuning Neural Networks by Both Connectivity and Size

This paper presents a new tuning algorithm for a neural network which not only considers the weighting but also the size and connectivity of the network. The approach is done by two parts: the addition of the switch-based hidden nodes and the application of a modified fitness function. The new model is tested by using two simple logic functions. The results show that the modifications lead to the creation of simpler networks, without sacrificing any accuracy or training time in the process. In addition, the lessened human interaction aspect of the new algorithm is also significant in real-time applications.