From my understanding, some miRNA prediction and discovery progrms, like miRDeep, utilize a probabilitic model to base their scoring on how likely a sequence segment is part of a miRNA precursor. What are some ways in which these programs choose their model and parameters? What are some of the general methods these programs use estimate the model parameters?

I have tried reading their papers. Some omit the discussion on the topic, some were a bit too technical for me to follow :( An introductory overview with reference reading materials would be greatly appreciated!
Thank you in advance!