Repeated Bayesian updates can require repeated normalization or integration. Conjugacy makes certain updates reusable: when a prior and likelihood form a conjugate pair, the posterior stays in the prior’s distribution family, with data changing its parameters. section 7.4 section 7.4.1
Table 7.7 Common conjugate prior–likelihood pairs and their applications.
|
Likelihood
|
Conjugate prior
|
Posterior family
|
Typical applications
|
|
Binomial (number of successes in 𝑛 trials)
|
Beta
|
Beta
|
Proportions, defect rates, conversion
rates, treatment outcomes
|
|
Poisson (number of events in fixed time)
|
Gamma
|
Gamma
|
Arrival rates, failure counts, call volumes
|
|
Normal (unknown mean, known variance)
|
Normal
|
Normal
|
Measurement uncertainty, signal processing,
quality control
|
Common prior–likelihood pairs retain their distribution family in the posterior. section 7.4.1
Updating a machine’s defect rate
Represent an estimated 5% defect rate with a Beta(1, 19) prior. After inspecting 10 parts and finding 2 defects, update the parameters to obtain Beta(3, 27), whose mean is 0.10. section 7.4.2
The posterior can serve as the prior for the next update, supporting sequential learning. Conjugacy makes these calculations efficient and transparent, but applies to particular prior–likelihood combinations rather than removing the logic of Bayes’ Theorem. section 7.4.3 section 7.4.1




