Adaptive quadratic neural nets

The authors present the theory and some results of a new algorithm for artificial neural networks that behaves well on complex data sets. They consider quadratic neural nets, and use dynamic methods for adapting the state of the net. The algorithm uses adaptive quadratic forms as discriminant functions and is very fast compared with backpropagation. The algorithm was applied to the well-known double-spiral problem, and it was shown that good solutions are attainable in times many orders of magnitude faster than conventional neural nets. >