Introduction to Neural Networks
October 18, 2016
Neural networks are capable of approximating arbitrary functions, and as such, have gained popularity in applications where methods to extract principled features from data are lacking. Here, we will discuss the basic structure of a neural network, and consider a few guidelines for training neural networks on straightforward supervised learning problems. Consider a classification task in which we wish to predict a vector of $P$ labels $\vec{y} = {\{}y^{(p)}{\}}_{p=1}^{P}$ given a matrix of $P$ examples $\vec{X}_0 = {\{}\vec{x}^{(p)}_0{\}}_{p=1}^{P}$.