Heuristic and hybrid methods for finding the global minimum of the error function in artificial neural networks

This paper introduces heuristics for the random optimization methods which result in better performance than random optimization method. A hybrid method is also demonstrated which is a combination of standard back-propagation and heuristic random optimization method, and which performs better than its constituents in that it results in a faster learning rate than these methods and does not get trapped in a local minimum. The above results are demonstrated through extensive simulation experiments, which demonstrate that heuristic and hybrid approaches are simple to implement, computationally effective, and robust over wide ranges of parameter values.