Math - part 5

• ML Estimation for unconditional distributions

  A) log

  B) max

  C) derivative

  D) 0

  
• Maximum likelihood estimate for

  A) Bernoulli distribution

  B) binary

  C) x0

  D) x0

  E) x1

  F) occurences

  
• Conditional probabilities

  A) discrete

  B) conditional

  C) xi|y

  D) events

  E) D

  
• ML estimate for the parameters

  A) Normal distribution

  B) Gaussian PDF

  C) Joint Mulitvariable Normal

  D) Multivariate Gaussian PDF

  
• Problem with the ML estimate :

  • ignores amount of evidence
  • bases its decision on likelihood
  • single estimate
• Bayesian Parameter estimation :

  • learns a probability distribution over all possible parameter values
  • computes posterior distribution
• Posterior distribution combines :

  • strength of evidence
  • subjective expectations
  • P(D) is hard to calculate integral
• Bayesian paramter estimation

  A) Maximum likelihood

  B) Maximum A posteriori

  C) posterior mean

  
• How to choose a final single estimate θ

  A) Maximum Likelihood

  B) Maximum A Posteriori

  C) Posterior Mean

  
• Bayesian Model averaging :

  • result of parameter estimation a probabiliy distributino over all possible θ
  • we could use all possible parameter settings to perform inference
  • "true bayesian approach" -> to complicated