COBEM99

SR21 Redes Neurais, Algoritmos Genéticos e Lógicas Nebulosa/Neural

Title:

SMART TRAINING IN NEUROCOMPUTING MODELING OF ENGINERING SYSTEMS

Summary :

IN THIS PAPER,``SMART TRAINING IS INTRODUCED TO INCREASE THE ACCURACY OF NEUROCOMPUTING MODELING AND REDUCE THE TIME REQUIRED TO DEVELOP NEUROCOMPUTING APPLICATIONS. IT IS SHOWN THAT BOTH OBJECTIVES ARE ATTAINED BY (I) PARTITIONING THE TRAINING DATA INTO CONTIGUOUS SUBSETS WHOSE RESPONSE BELONGS TO A SIMILAR CLASS, AND (II) DESIGNATING DIFFERENT NEURAL NETWORKS TO FITTING PARTITIONED DATA. RESPONSE-BASED PARTITIONING IS PERFORMED BASED ON QUANTIFICATION OF THE RESPONSE CHARACTERISTIC BY AN AUTOMATED CLUSTERING ROUTINE. THE ROUTINE USES A RULE-BASED FUZZY LOGIC INFERENCE ENGINE TO ADAPT DATA PARTITIONING TO THE CHANGING RESPONSE CHARACTERISTIC AND TO MAKE THEM SUITABLE FOR TRAINING. THE NEUROCOMPUTING MODELING WITH RESPONSE-BASED CLUSTERING IS APPLIED TO TWO DISTINCTLY DIFFERENT PROBLEMS. THE FIRST PROBLEM ADDRESSES THE NEED FOR CONTINUOUS REPRESENTATION OF DATA PRODUCED BY A FINITE ELEMENT ANALYSIS, AND SPECIFICALLY, DISPLACEMENT FIELDS IN THE UPPER SKIN OF A WING BOX ARE MODELED. IN THE SECOND PROBLEM, SIMULATED DATA REPRESENTING POWER LOSS OF TRANSMITTED SIGNALS TO BE USED IN DESIGN OF A WIRELESS COMMUNICATION SYSTEM IN PRAGUE ARE MODELED. KEYWORDS: NEURAL NETWORKS, SMART TRAINING, PROGRESSIVE DEFORMATION, POWER LOSS MODELING

Author :

Macdonald, Andrew J.
Szewczyk, Z. Peter

Paper View :