• Occam's Razor: when given a choice between models that model the data "equally well", choose the least complex

• Use data-driven methods to determine the optimal trade-off between model complexity and the model's ability to accurately represent the data

• Typically based on Bayesian statistics - posterior probabilities