Authors:
Roman Sergienko
1
;
Oleg Akhtiamov
2
;
Eugene Semenkin
2
and
Alexander Schmitt
1
Affiliations:
1
Ulm University, Germany
;
2
Siberian State Aerospace University, Russian Federation
Keyword(s):
Natural Language Processing, Classification, Artificial Neural Network, Ward Net, Error Backpropagation Algorithm, Genetic Algorithm, Supervised Learning.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Computational Intelligence
;
Evolutionary Computing
;
Genetic Algorithms
;
Hybrid Learning Systems
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Neural Networks Based Control Systems
;
Optimization Algorithms
;
Soft Computing
Abstract:
A novel approach to artificial neural network design using a combination of determined and stochastic optimization methods (the error backpropagation algorithm for weight optimization and the classical genetic algorithm for structure optimization) is described in this paper. The novel approach to GA-based structure optimization has a simplified solution representation that provides effective balance between the ANN structure representation flexibility and the problem dimensionality. The novel approach provides improvement of classification effectiveness in comparison with baseline approaches and requires less computational resource. Moreover, it has fewer parameters for tuning in comparison with the baseline ANN structure optimization approach. The novel approach is verified on the real problem of natural language call routing and shows effective results confirmed with statistical analysis.