GA-Based Filter Selection for Representation in Convolutional Neural Networks

Junbong Kim, Minki Lee, Jongeun Choi, Kisung Seo

Abstract: One of the deep learning models, a convolutional neural network (CNN) has been very successful in a variety of computer vision tasks. Features of a CNN are automatically generated, however, they can be further optimized since they often require large scale parallel operations and there exist the possibility of overlapping redundant features. The aim of this paper is to use feature selection via evolutionary algorithms to remove the irrelevant deep features. This will minimize the computational complexity and the amount of overfitting while maintaining a good quality of representation. We demonstrate the improvement of the filter representation by performing experiments on three data sets of CIFAR10, metal surface defects, and variation of MNIST and by analyzing the classification performance as well as the variance of the filter.