I am looking into Bayesian learning for the first time ever and am just wondering why we look to have a conjugate prior to carry out our estimation with.


1 Answer 1


You do not have to have a conjugate prior and indeed, you should not have a conjugate prior unless it fits your prior knowledge. Many conjugate prior distributions are good approximations of actual knowledge. Some can be problematic, like the inverse Wishart, when used in a way that is not representative of information or as a diffuse prior.

Conjugate priors permit fast Bayesian updating, which can be valuable in high dimension problems.

Conjugacy only exists for a fraction of likelihood functions. You cannot always use one.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.