|
We know that if a Probability Distribution is part of the exponential family, is straight forward to obtain the sufficient statistics of said distribution. However, I have 2 questions, is there a general method to obtain the sufficient statistics for any distribution? And are there any other distributions, aside of those in the exponential family used for applications in Machine Learning? Thanks |
To the first question is Fisher-Neyman factorization general enough or is that too general for your needs? For the second the uniform distribution is not in the exponential family, but is used in machine learning particularly by Bayesians since its popular as an uninformed prior.