Extended Cascade-Correlation for Syntactic and Structural Pattern Recognition

Automatic inference is one of the main problems that syntactic and structural pattern recognition must solve for successful applications. Neural networks are artificial intelligence tools which already support automatic inference for successful applications of statistical pattern recognition. In this paper, we suggest that neural networks, and specifically Cascade-Correlation, can be used for automatic inference in syntactic and structural pattern recognition, as well. An extended version of a standard neuron which is able to deal with structures is presented and the Cascade-Correlation algorithm generalized to structured domains. The computational complexity of the proposed algorithm as well as experimental results obtained on problems involving logic terms are presented.