Applicability of Neural Network Techniques to Underwater Naval Tactics

In this paper several key issues in supervised learning using artificial neural networks are addressed within an ASW barrier operation framework. Training data distribution, input space representation, parameters variations, signal excess fluctuations and agents behaviors have a great influence on the neural network controlled submarine performance and tactics. A qualitative sensitivity analysis of all these factors is carried out using a simulated battlefield. The ultimate goal of this study is to assess the capabilities and the limitations of neural network controlled intelligent agents for solving more general problems.