part two in the series on dualities in statistics

Part of the reason I like these dualities is that they can justify some techniques that seem very improvised or ad hoc. Especially machines learning is full of this, often having many hyperparameters