Utilizing Keras, Pytorch, and MXnet for Building CNN Binary Classification and Finding their Discrepancy in Performance

Utilizing Keras, Pytorch, and MXnet for Building CNN Binary Classification and Finding their Discrepancy in Performance

Poster 204

The image classification method has grown significantly from when it the idea was first introduced [1] to the point where now it is getting applied to most IoT devices. Python library is a prime example explaining this trend of openness and visibility of machine learning algorithm in a way that most people could directly view and implement its intricate and refined structure to their desire. In extension to the python library that provides basic concepts of machine learning, some of the well recognized open neural network library such as Tensor Flow, Keras , Pytorch , and Gluon MXnet adds significant amount of machine learning library to the python environment. Because of this open accessibility many papers that is related to machine learning have been inundating this research field, while image classification being one of them. While there are many papers published in this field utilizing deep learning algorithm applied to binary image classification. These paper’s findings are limited to changing the original structure of Convolutional Neural Network (CNN) to make it have better performance in image classification These papers do have some intriguing aspects to it in each of their special field, and some of them could possibly be the next paradigm for improving the CNN performance.