Analysis of Activation Functions for Convolutional Neural Network based MNIST Handwritten Character Recognition
7 Pages Posted: 28 Jan 2019
Date Written: January 14, 2019
This research work has deployed a promising approach to recognize handwritten English alphabets and neural digits. In this research work, the deep neural network based Convolution Neural Network model is analyzed and studied for its performance on the MNIST dataset. This standardized dataset comprises handwritten text including numeric and alphabets. During various experimentations, parameters of CNN were configured at multiple instances to analyze its performance in different environments. Though the prime configuration set included the error function, activation function, number of hidden layers, number of epochs, various optimization techniques to resolve the convex and non-convex optimizable objective function. The experimental outcome proves its promising worth comparable to existing arts. The concluded performances of the discussed mode achieve the highest classification rate of 99.65 at sigmoid activation function with cross-entropy error function and foremost the latency in performance is evaluated as the measure of different number of hidden layers.
Suggested Citation: Suggested Citation