Performance Analysis of CNN Model for Digit Recognition

Convolutional Neural Networks (CNN) are the key constructors of deep neural networks, which mimics the human brain activity. CNN models rely on convolution operation performed on images with different filters or kernels, that helps to identify certain key aspects of an image. However, in order to activate such a convolution process, activation functions are required and optimizers are then used to further optimize the operation. There are several activation functions and optimizers available, based on which the scale of efficiency of the model varies. In this paper, analysis of different combinations of optimizers and activation functions for the CNN model has been done and resulted in a most suitable and efficient combination, which produced a promising result of 99.2% accuracy on a general architecture.