Applicability of Learning Automata Theory to Neural Network Design

这是一项初步的研究工作，旨在确定学习自动机理论的强大工具和概念是否可以用于解决神经网络中出现的各种优化问题。在拟议的研究的第一阶段，将对几个简单的互连自动机进行详细的研究。其主要目的是确定如何将已知结果扩展到广义自动机，从而扩展到神经网络。在第二阶段，重点将是使用学习自动机的方法来优化一般的等级系统，其中前面描述的视觉系统是一个典型的例子。在研究的两个阶段都考虑了简单网络和复杂网络的计算机模拟。对有趣的模拟行为的观察将提供新的理论焦点的来源以及新结果的线索。

This is a preliminary research effort to determine whether powerful tools and concepts of learning automata theory can be brought to bear on a variety of optimization problems which arise in neural networks. In the first phase of the proposed research a detailed study of several simple interconnected automata will be undertaken. The primary aim of this is to determine how known results can be extended to generalized automata and hence to neural networks. In the second phase the emphasis will be on the use of learning automata methods of optimization of general hierarchical systems of which the vision system described earlier is a typical example. Computer simulations of both simple and complex networks are contemplated during both phases of the research. Observation of interesting simulated behavior will provide both sources of new theoretical focus as well as clues to new results.