Chapter 4 Summary - Pattern Recognition and Machine Learning

Описание: In this video I summarize some of the key ideas in the Chapter. We start with the idea of linear discriminant functions, then touch on the three ad-hoc approaches to finding the discriminant: classification as regression, the perceptron learning algorithm, and Fisher's linear discriminant. We then discuss the probabilistic approach and discuss the generative vs. discriminative perspectives, arriving at logistic regression, the associated cross-entropy loss, iterative reweighted least-squares for estimating its parameters. We finish with the Bayesian approach to logistic regression, which requires evaluating complicated integrals over the parameters, motivating Laplace's approximation.