This document summarizes key points from Chapter 3 of the book "Pattern Recognition and Machine Learning" by Christopher M. Bishop. It discusses linear regression, Bayesian linear regression, and model comparison. The main points are:
1) Linear regression finds the best fitting linear relationship between inputs and outputs. Bayesian linear regression places prior distributions over the weights and finds the posterior distribution.
2) The prior in Bayesian linear regression acts as an intrinsic regularization. As more data is added, the posterior variance decreases while the noise variance remains.
3) Model evidence can be used to perform Bayesian model comparison by finding which model best explains the data. Approximations are required to evaluate the evidence.