General Weight Matrix Formulation Using Optimal Control

摘要

Classical methods from optimal control theory are employed in deriving generalforms for neural network weights. The network learning or application task is encoded in a performance index of a general structure. Consequently, different instances of this performance index lead to special cases of weight rules, including some well-known forms. In particular comparisons are made with the outer product rule, spectral methods, and recurrent back-propagation. Simulation results and comparisons are presented. Specific Topics include neural network weights, outer product rule, spectral methods, and recurrent back-propagation.