GPGPU Implementation of Fuzzy Rule-Based Classifiers

This paper presents a parallel implementation of fuzzy-rule-based classifiers using a GPGPU (General Purpose Graphics Processing Unit). There are two steps in the process of fuzzy rule-based classification: Fuzzy-rule generation from training data and classification of an unseen input pattern. The proposed implementation parallelizes these two steps. In the step of fuzzy-rule generation from training patterns, the membership calculation of a training pattern for available fuzzy sets is simultaneously processed. On the other hand, the membership calculation of an unseen pattern for the generated fuzzy if-then rules is simultaneously processed in the step of the classification of the pattern. The efficiency of the parallelization is evaluated through a series of computational experiments. The effect of the parallelization is evaluated for each step of fuzzy rule-based classifiers. The results of the computational experiments show that the proposed implementation successfully improve the speed of fuzzy rule-based classifiers.