Knowledge Extraction with Neural Networks: Significant Patterns and their Representation in Back-Propagation Networks

Abstract

This book provides new means for knowledge extraction with neural networks of the Back-Propagation type: to understand and interpret correctly the knowledge extracted by the network, detect significant, e.g. novel input patterns, identify their characteristic features and assess their future development. Two included case studies are devoted image classification and analysis of economical data provided by the World Bank.

Keywords

• knowledge extraction
• neural networks
• back-propagation
• internal knowledge representation
• regularization
• pruning