NSK, an object-oriented simulator kernel for arbitrary feedforward neural networks
Cédric Gégout, Bernard Girau, Fabrice Rossi · 2002
An object-oriented neural network simulator kernel is presented. It as based on a general mathematical model for arbitrary feedforward nets. We propose a C++ implementation of this model which satisfies the following requirements: expandability (allowing an easy implementation of a new neural model), portability and efficiency (the kernel does not increase significantly its computation times for classic models, compared to a direct object-oriented implementation). Learning algorithms such as gradient-based ones can be written for arbitrary nets and are therefore directly available for every particular model.>