Growing Cascade Correlation Networks in Two Dimensions - A Heuristic Approach

Abstract

Dynamic neural network algorithms are used for automatic network design in order to avoid time-consuming search for finding an appropriate network topology with trial and error methods. Cascade Correlation Network (CCN) is one of the constructive methods to build network architecture automatically. CCN faces problems such as large propagation delays and high fan-in. In this paper, we present a Heuristic Pyramid-Tower (HPT) neural network designed to overcome the shortcomings of CCN. Benchmarking results for the three real-world problems are reported. The simulation results show that a smaller network depth and reduced fan-in can be achieved using HPT as compared to the original CCN.