An Industrial Strength Numerical Modeling Research & Development Tool

NODElib is an open source programming library that can be used to rapidly develop complex neural network simulations. In NODElib, all neural architectures are considered special cases of a general feedforward model; thus, advance numerical routines (such as calculating the Hessian, optimizing the Jacobian, performing quasi-Newton, conjugate gradient, backprop, or Leverberg-Marqardt, etc.) are all available regardless of the model type (MLP, SMLP, CNLS, RBFN, HONN, etc.). Similarly, the models and optimization routines are blind to the data containers as well, which allows NODElib to work on memory-mapped gigabyte files with no extra coding. Since NODElib is OS-neutral, it can be trivially embedded into an existing application or used for research purposes. This paper describes the main design of NODElib with an emphasis on features that allow for novel research questions to be pursued.