This research paper proposes the design and implementation of an Artificial Neural Network (ANN) development system which will provide the foundation and a tool for study and research in ANN architectures and algorithms. A system was developed which allows for the implementation of networks and for the modification of common and interesting network parameters and algorithms. This establishes a versatile and effective model from which to proceed with ANN study.

The system design is an initial prototype providing a generic and dynamic interface which allows the versatility to implement simple networks on a Personal Computer and modify their parameters, architectures, and algorithms. It also allows the monitoring of the internal conditions of the network, providing a basis for detailed data collection and research.

Several different neural nets were implemented and trained on the system. Various feedforward networks using the Delta Rule and the backpropagation learning algorithms were trained in the supervised mode to solve problems in pattern recognition. Modifications to these networks were then used to compare training and operational characteristics between the different architectures. A classic one layer Hopfield net using a recurrent feedback, associative memory architecture was also implemented and trained in the unsupervised mode. An unsupervised Hebbian learning algorithm was also implemented and tested on the system.

Several enhancements are proposed which will increase the versatility of the system and aid in the further study of Artificial Neural Network implementations and characteristics.