ML 4.5.2 Variational Bayesian Logistic Regression

Logistic Regression is a simple method linear classification, one where input features are provided with weights so that the weighted sum can be mapped to a class. Bayesian logistic regression starts with a Gaussian probability distribution over the weights. Prediction for a new input requires integrating over all values of weights where the integrand is a product of a sigmoid and Gaussian. The variational approximation makes the integration simpler by making the integrand a product of Gaussians.