Image recognition and study of hyperparameter optimization of convolutional neural networks using tensorflow and keras frameworks

Deep Learning is a branch of Machine Learning which involves complex learning with massive data sets. It is implemented with Artificial Neural Networks (ANNs). The ANNs, in turn, are inspired by the biology of the Human Brain. ANNs mimic the functioning of biological neurons in the human brain, by a combination of hardware (physical circuitry and processors) and software (algorithms implemented with special ANN supporting frameworks), to perform human like tasks. Different Deep Learning models are used for different purposes. The model used for image recognition tasks is called Convolutional Neural Networks (CNNs). CNNs are also inspired from the Human Brain’s ability to recognize images through its Visual Cortex. The goal of this research work is firstly to contrast the effectiveness of Deep Learning models over traditional Machine Learning models for Image Recognition distinguishing between cat and dog images. Secondly, it details ways to improve the accuracy of the Convolutional Neural Network by using an optimizing method through which, all non-self-adjusting network parameters are varied to yield an optimized model. The thesis helps to validate the concepts of Machine Learning being able to successfully emulate the Human Brain for Visual Cognition. The ANN and Deep Learning models used in this research have been implemented with popular frameworks of Tensorflow and Keras with Python. The results across 70 experiments are summarized. The key insights are presented in the conclusion.