4.3.3 Iterative Reweighted Least Squares - Pattern Recognition and Machine Learning

In this video, we apply the Newton-Raphson method to minimizing the cross-entropy loss of logistic regression and find that the parameter updates are a weighted least sqaures projection of a mysterious vector that can be interpreted as a linear approximation to the logistic sigmoid. Why the updates have these particular properties will be discussed using supplementary notes in the next video.