Bias-Variance Decomposition
Squared-error prediction risk can be decomposed into bias, variance, and irreducible noise under stated assumptions.
-
-
Empirical Risk Minimization and Generalization
Training minimizes error on a sample. Everything hard is the gap to the population.
-
Loss Functions and Convex Surrogates
The loss function decides what a model estimates and how it treats large errors; decisions and costs come after.