Image Classification using Deep Learning: An Experimental Study on Handwritten Digit Recognition

This paper presents an experimental study of the use of Deep Learning using Convolution Neural Networks (CNNs) for Image Classification. Specially, the problem being addressed here is of recognition of handwritten digits. The objective is to report variations in testing errors and accuracies with varying kernel size and varying number of feature maps. We performed handwritten digit classification using neural network and deep learning for a subset from the MNIST dataset, which contains 60,000 training images and 10,000 test images in all. It is observed that the accuracy and loss are stabilizing with minor change in the kernel size and the number of feature maps.