Control systems in neural networks with local memory

AbstractBased on a new paradigm of neural networks consisting of neurons with local memory (NNLM), we discuss the representation of a control system by neural networks. Using this representation, the basic issues of complete controllability and observability for the system are addressed. A separation principle of learning and control is presented for NNLM. The result shows that the weights of the network will not affect its dynamics. Some results about local linearization via a regular static feedback and nonlinear transformation are also given