Conjugate Priors
Pairs of distributions where the math stays in the family.
When the posterior is in the same family as the prior, the prior is called conjugate to the likelihood. Conjugate analysis gives closed-form posteriors and turns Bayesian updating into mechanical parameter arithmetic.
The most common conjugate pairs:
Beta is conjugate to Bernoulli/Binomial.Gamma is conjugate to Poisson and Exponential rates.Normal-Inverse-Gamma is conjugate to Normal with unknown mean and variance.Dirichlet is conjugate to Multinomial.
Why bother? Closed-form updates are fast and sanity-checkable, no MCMC needed. Even when your final model uses flexible priors and numerical sampling, a conjugate baseline gives you intuition for what the data are saying and a check on the more complex code.
Conjugate models are especially handy in production systems where Bayesian updates need to happen on every event — A/B test dashboards, click-through rate estimators, multi-armed bandits.