Preferencje help
Widoczny [Schowaj] Abstrakt
Liczba wyników
2004 | nr 1011 Pozyskiwanie wiedzy i zarządzanie wiedzą | 210--223
Tytuł artykułu

Algorytm genetyczny z optymalizacją wielokryterialną w sensie Pareto w pozyskiwaniu reguł z sieci neuronowej

Warianty tytułu
Genetic Algorithm with Pareto Optimization in Rule Extraction from Neural Network
Języki publikacji
W pracy została zaprezentowana metoda GENPAR, która pozyskuje reguły z sieci neuronowej. Bazuje ona na algorytmie genetycznym i optymalizacji wielokryterialnej w sensie Pareto.
In the paper the method of rule extraction from a trained neural network based on genetic algorithm is presented. That problem can be seen as a multiobjective optimization, because acquired set of rules describing behavior of a trained neural network has to satisfy the objectives impose by the user, which can include for example the small as possible the number of rules describing behavior of neural network with the as high as possible fidelity. Such approach is used in the paper by applying multiobjective optimization in Pareto sense. The evaluation of each objective separately gives the possibility for the user to decide which objective is the most important. The proposed method of rule extraction - GENPAR, treats neural network as black box. Owing to this fact it does not impose any constraints on the neural network architecture. It does not require special training procedure of neural network, as well. It proceeds all types of neural network attributes. The effectiveness of the method was tested in the experimental study with using benchmark data sets from UCI repository. They allowed to test its efficiency and scalability. The obtained results of experimental study are presented in the paper and discussed. They show that GENPAR has comparable results to other known methods, but it has bigger elasticity offering to the user ability to choose which objective is more important for him. Presented version of the method consider only two objectives, but in the future after small modification it is possible to use other objectives in the evaluation process of extracted set of rules. (abstract original)
  • Andrews R.: Survey and Critique of Techniques for Extracting Rules from Trained Artificial Neural Networks, Knowledge-Based Systems, Volume: 8, Issue: 6, December (1995), pp. 373-389.
  • Andrews R., Geva S.: (1995): Rulex & СЕВР Networks as the Basis for Rule Refinement System,
  • Blake C.C., Merz C.: UCl Repository of Machine Learning Databases, University of California, Irvine, Dept. of Information and Computer Sciences (1998).
  • Darbari A.: Rule Extraction f rom Trained ANN:A survey. Technical Report Institute of Artificial intelligence, Dep. of Сотр. Science, TU Dresden (2000).
  • McMillan C.: Rule Induction through Integrated Symbolic and Subsymbolic Processing,
  • Neumann J.: Classification and Evaluation of Algorithms for Rule Extraction from Artificial Neural Networks, PhD Summer Project ICCS, Division of Informatics, University of Edinburgh, August 1998.
  • Setiono R., Liu H.: NeuroLinear: A System for Extracting Oblique Decision Rules from Neural Networks, European Conference on Machine Learning, pp. 221-23, (1997).
  • Setiono R.: Extracting Rules from Neural Networks by Pruning and Hidden-Unit Splitting, Neural Computation, vol. 9 (1), pp. 205-225, (1997).
  • Setiono R.: Extracting M of N Rule f rom Trained Neural Networks,
  • Taha I., Ghosh J.: Symbolic interpretation of Artificial Neural Networks, Tech. Rep. TR-97-01-106, The Computer and Vision Research Center, University of Texas, Austin (1996).
  • Thrun S.B.: Extracting Provably Correct Rules from Artificial Neural Network, Technical Report, Institut für Informatik, Universität Bonn, Bonn (1994).
  • Zitzler E., Thoele L.: An Evolutionary Algorithm for Multiobjective Optimization: The Strength Parete Approach,
Typ dokumentu
Identyfikator YADDA

Zgłoszenie zostało wysłane

Zgłoszenie zostało wysłane

Musisz być zalogowany aby pisać komentarze.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.