Control on landscapes with local minima and flat regions: A simulated annealing and gain scheduling approach

Landscapes containing local minima and ¿flat¿ regions are frequently encountered in multi-layer neural networks that employ sigmoid-like activation function in the hidden layers. Numerous techniques in the neural network community have been proposed to address these issues. In this note, we extend these ideas to the neural network control of nonlinear systems. We propose a solution which employs simulated annealing and a gain scheduled learning rate.