Encyclopedia · 176 concepts

Math & Optimization · beginner · concept 39 of 176

Probability & Bayes' Theorem

Machine learning is applied probability: models output distributions, training maximizes likelihood, and Bayes' theorem tells you how to update belief when evidence arrives. It is also the antidote to the classic diagnostic fallacy, a positive test for a rare disease usually still means you are fine.

Key terms

Conditional probabilityPrior and posteriorLikelihoodDistributionsExpectation

Where you meet it in the real world

Spam filters, medical diagnosis, language model sampling, risk models, A/B test analysis

Guides and articles

Seeing Theory

Seeing Theory (Brown University)