From Machine Learning to Deep Learning

Abstract

This chapter provides a thorough grounding in the fundamental ematical concepts of deep learning. It is first shown how a simple linear classifier can be defined based on the equation for a straight line. A more general scheme for optimization of the parameters of classifiers is introduced, based on gradient descent and its variants. We then see how the basic classifier model can be extended to produce simple artificial neural networks such as the perceptron and logistic regression. Next, these models are taken further to show how multiclass classification problems and nonlinearly separable data can be handled. Finally, the idea of a convolutional neural network is introduced and we see how this leads to the idea of deep learning. A practical tutorial is provided to give the reader some practical experience of developing classification models using Python.