G001 A SYSTEMATIC APPROACH TO THE OPTIMAL DESIGN OF FEED FORWARD NEURAL NETWORKS APPLIED TO LOG-SYNTHESIS Abstract 1 Neural networks are increasingly used in geophysical applications. Optimizing neural networks is still a matter of experience and trial and error where network initialization and network size are the most challenging issues. We expanded conventional learning rules to a completely forward connected network including input neurons for automatic normalization of the data. In addition we developed a method for the network initialization based on the statistical properties of the input and output data generating an initial network state that ascertains a fast