This lecture generalizes an idea that we saw in the previous one: Learning can be framed as an optimization problem and can be declaratively stated as minimizing empirical risk.
The lecture ends with a discussion of model selection, the bias-variance tradeoff and cross validation.
- Chapter 7 from The Elements of Statistical Learning, by Hastie, Tibshirani and Friedman. (Available online)
- Pedro Domingos, A Unified Bias-Variance Decomposition