Incorporating epistemic uncertainty has been a subject of many recent research. I’d like to show a method that is in line with the Jayensian view of probability.
With advanced sampling method (e.g., HMC), assigning prior distributions with arbitrary marginal is no longer a hurdle. However, what if we want to assign correlation between multiple random variables, each with marginal distributions very different to each other? Here I provide a general trick, which can also be applied to simply help the sampler in difficult regions (e.g., near the boundaries of bounded distributions), regardless if the marginals are independent, semi-dependent, or fully dependent.
