Justification of STL-10 dataset using a competent CNN model trained on CIFAR-10
Hardik Singh, Sweta Swagatika, Raavi Sai Venkat, Sanjay Saxena · 2019 3rd International conference on Electronics, Communication and Aerospace Technology (ICECA) · 2019
Deep Learning is a subfield of machine learning concerned through algorithms stimulated by the edifice and purpose of the brain called ANN (artificial neural networks). A convolutional neural network (CNN) is a class of deep neural networks, utmost generally applied to examining painterly images. It uses a distinction of multilayer perceptrons intended to necessitate nominal preprocessing. The CIFAR-10 dataset contains 60000 32×32 color images distributed in 10 classes, having 6000 images in each class and it is recurrently used for training, testing and validation of CNN. In this article, we have designed a well-organized model with several convolutions a hidden layers. 90.91% training accuracy and 83.69 validation accuracy are obtained. Further, it is validated on STL 10 data set which is inspired by CIFAR 10 data set and significant testing accuracy has been obtained. This article will help the researchers who are working on CNN deep models.