11. Objective Bayes inference and MCMC Flashcards
What are the problems of: Bayesian inference
- In the absence of relevant past experience, the prior may have an unwanted
subjective role. - Numerical calculations often involves intricate high-dimensional integrals.
Recall the bayesian setting:
given a parametric model:
How to choose g(µ)?
A: We’ll focus on uninformative priors
Q: How to evaluate integrals over the posteror such as E{µ|x}?
A: Markov Chain Monte Carlo (MCMC).
What is a flat prior?
For a finite space it is 1/k, all samples have the same prob
Fisher claims a flat prior does not have a transfomration invariance, what was his solution?
Jeffreys prior
What are cunjgate priors?
They are priors such that they will a posterior from the same family.
Explain what n+ and x+ mean.
x+ = Its the update of the mean of the observation, its a fraction of the hypothetical observations of all of those, that is the weight of x0 and the mean of x