Jeffreys Prior

The Jeffrey’s prior is an easy-to-compute reference prior that is invariant to transformation, used in Bayesian Inference. If the model only has a univariate parameter $θ$ , the prior is given by:

$p (θ) \propto \sqrt{I (θ)}$

where $I (θ)$ is the expected Fisher information in the model.

If $θ$ is multi-dimensional, then the Jeffrey’s prior is given by:

$p (θ) \propto \sqrt{\det {l (θ)}}$

where I is the Fisher information matrix. When the number of dimensions is large, this method becomes cumbersome. A common approach is to obtain non-informative priors for the parameters individually, and form the joint prior as a product of these individual priors.

Jethro's Braindump

Jeffreys Prior

Links to this note