Priors, likelihood and updating on evidence. The maths behind medical test confusion, spam filters and why good models say how unsure they are.
3 units, about 3 hours. Free.
Apply Bayes rule to a screening problem and explain why a 99 per cent accurate test can still be usually wrong.
2Implement a naive Bayes classifier and explain the independence assumption and smoothing.
3Distinguish confidence from correctness and explain calibration and the two kinds of uncertainty.