Advanced algorithms for neural networks

This practical guide contains a wide variety of state-of-the-art algorithms that are useful in the design and implementation of neural networks. All algorithms are presented on both an intuitive and a theoretical level, with complete source code provided on an accompanying disk. Several training algorithms for multiple-layer feedforward networks (MLFN) are featured. The probabilistic neural network is extended to allow separate sigmas for each variable, and even separate sigma vectors for each class. The generalized regression neural network is similarly extended, and a fast second-order training algorithm for all of these models is provided. The book also discusses the recently developed Gram-Charlier neural network and provides important information on its strengths and weaknesses. Readers are shown several proven methods for reducing the dimensionality of the input data.

Advanced Algorithms for Neural Networks also covers advanced multiple-sigma PNN and GRNN training, including conjugate-gradient optimization based on cross validation, the Levenberg-Marquardt training algorithm for multiple-layer feedforward networks, advanced stochastic optimization, including Cauchy simulated annealing and stochastic smoothing, data reduction and orthogonalization via principal components and discriminant functions, economical yet powerful validation techniques, including the jack-knife, the bootstrap, and cross validation and includes a complete state-of-the-art PNN/GRNN program, with both source and executable code.