Career: Adaptation and Learning in Distributed Systems Using Neural Networks

分布式信息共享计算机网络和汽车系统是工程和计算系统中越来越多的应用领域的两个例子，这些应用领域需要物理分布的决策者和控制器之间的协作交互。由于局部决策对总体目标的影响和系统状态信息的不完全等不确定性，这种分布式应用经常会引起非线性分布式决策和控制问题。人工神经网络在非线性决策和控制规则的自适应实现方面已被证明是有效的。为了将它们应用到分布式应用中，需要新的互连模型以及适应和学习方法来应对上述分布式不确定性源。利用大系统理论的成果和当前对多个神经网络的研究，本研究将探讨几个密切相关的神经网络分布式互连模型。将找到建设性的方法来确定局部测量完备性不同条件下整体性能函数的局部近似值。这些局部性能函数，反过来，将产生自适应的方法来实现非线性决策和控制规则使用神经网络。该模型和方法将应用于计算机网络信息共享和汽车控制等应用领域的问题。

9623971 Mukhopadhyay Distributed information-sharing computer networks and automotive systems are two examples of an increasing number of applications areas in engineering and computing systems which require collaborative interaction between physically distributed decision-makers and controllers. Such distributed applications frequently give rise to nonlinear distributed decision and control problems in the presence of uncertainties such as the effects of the local decisions on the over-all objective and incomplete system state information. Artificial neural networks in the past have proven effective in adaptive realization of nonlinear decision-making and control rules. In order to apply them to distributes applications, new interconnection models as well as adaptation and learning methods are needed to cope with distributed sources of uncertainty such as those mentioned above. Utilizing results form large-scale systems theory and current research on multiple neural networks, the proposed research will investigate several closely-related distributed interconnection models of neural networks. Constructive methods will be found to determine local approximations to an overall performance function under different conditions on the completeness of local measurements. These local performance functions, in turn, will yield adaptive methods for realization of nonlinear decision and control rules using neural networks. The model and methods derived will be applied to problems in the application areas of information sharing over computer networks and automotive control.