Methods for rapid learning in artificial neural networks

The slow convergence rate of the fixed step size backpropagation learning algorithm used for training artificial neural networks (ANNs) is discussed. The role of numerical methods in accelerating the learning process is discussed along with some observations about the parallels between some new acceleration methods described recently by ANN researchers and well known methods in the mathematical literature. The PARTAN algorithm is introduced to the ANN learning problem. The results show that PARTAN has excellent convergence properties, even when compared to other accelerated methods, converging hundreds of times faster than simple fixed step backpropagation methods