ABSTRACT
Many machine learning models learn a relationship between variates in a dataset by introducing a parametric class of functions and identifying the values of the parameters that minimize a specified loss value. The core idea of deep learning is to enhance simple parametric models by composing several of them. Such compositions are called neural networks, in reference to their historical role in efforts to model biological systems. In addition to its role in image and language processing, deep learning has led to monumental improvements in computer board game play. The pace of achievements in deep learning over the past decade has been unparalleled in other areas of machine learning and has cemented its status as an important resource in the data scientist's toolkit. The authors also develop the ideas necessary to implement a plain-vanilla neural network called the multilayer perceptron. They discuss convolutional neural networks, which are designed for handling image data.
