Seonglae ChoEmpirical Risk Minimization
A generalization of the Maximum Likelihood principle MLE: replace the log likelihood with any other loss function
When a loss function is computationally difficult to minimize, it is often replaced with convex upper bounds