Fast Sensitivity-Based Training of BP-Networks

Abstract

Sensitivity analysis allows to assess the influence of each neuron or weight on the final network output. This capability is crucial for various feature selection and pruning strategies.

In this paper, we present a new approximative sensitivity-based training algorithm yielding robust neural networks with generalization capabilities comparable to its exact analytical counterpart, yet much faster.

Keywords

• neural networks
• back-propagation
• internal representation
• sensitivity analysis
• feature selection
• pruning
• generalization