ABSTRACT
This chapter focuses on the pruning algorithms and reviews published techniques used in pruning algorithms. Since the 1990s, the feedforward neural network has been universally used to model and simulate complex nonlinear problems, including the supposedly unknown mathematical relationship existing between the input and output data of applications. To date, the choice remains arbitrary and intuitive, which makes the identification of the structure a fundamental problem to be solved, and has a huge effect on the remaining steps. The number of existing neurons in the input layer and output layer is fixed by the number of inputs and outputs, respectively, of the system to be modelled. The power of neural network architectures strongly depends on the used architecture. Commencing with a multilayer neural network of large structure, the task for the pruning algorithm is to optimize the number of layers and the number of neurons needed to model the desired function or application.
