Pattern classification using updated fuzzy hyper-line segment neural network and it's GPU parallel implementation for large datasets using CUDA

Abstract

Fuzzy hyper-line segment neural network (FHLSNN) is a hybrid system of fuzzy logic and neural network and is used for pattern classification. It learns patterns in terms of n-dimensional hyper line segment (HLS). Modified fuzzy hyperline segment neural network (MFHLSNN) is a modified version of FHLSNN that improves the quality of reasoning and recall time per pattern using modified fuzzy membership function. However, for the large training patterns MFHLSNN creates a large number of HLS, which increases training and recall time and thus, is limited only to smaller dataset on sequential machines/implementations. In this paper, we proposed a new fuzzy membership function for MFHLSNN. Using new membership function architecture is called Updated FHLSNN (UFHLSNN). We also proposed GPU (Graphics Processing Unit) parallel implementation of UFHLSNN, for larger pattern datasets, using NVIDIA's CUDA (Compute Unified Device Architecture). The updated membership function is found superior than the original one, in terms of number of arithmetic operations. The maximum speed-up achieved in training and recognition phases are 12× and 29×, respectively for the used datasets. Thus, we strongly recommend GPU parallelization of UFHLSNN using CUDA, for larger pattern recognition datasets.