Function complexity estimation and its application to the optimum tie of geophysical data using Anns

Based on the learning convergence responses of BP neural networks to the complexity of functions underlying their learning data, we suggest a method to estimate the relative complexity of the approximated functions and apply it to the optimum solutions of geophysical data by searching the relative simplest function in some function space without needing exactly to know the mapping function. The numerical analysis and a real case verify the efficiency of the method.