• ☰
  • Home
  • Introduction
  • Generative models
  • Conditioning
  • Causal and statistical dependence
  • Conditional dependence
  • Social cognition
  • Interlude - Bayesian data analysis
  • Interlude - Algorithms for inference
  • Rational process models
  • Learning as conditional inference
  • Learning with a language of thought
  • Hierarchical models
  • Occam's Razor
  • Mixture models
  • Learning (deep) continuous functions
  • Appendix - JavaScript basics
  • Appendix - Useful distributions

1. Discussion of MCMC

@T.L.Griffiths:2008:dd194 - Sec. 5.0 Markov Chain Monte Carlo (pp. 31-34)

Reading questions:

a) Under what conditions is it not necessary to use an approximate sampling method to solve a Bayesian equation?

b) What are the major differences between Gibbs sampling and Metropolis-Hastings sampling?

2. Particle filters

Particle Filters Explained without Equations

Viewing questions:

a) As the number of particles increases, what happens to a particle filter’s accuracy? What happens to its run-time? Would you want an infinite number of particles? Why or why not?

b) Describe a phenomenon that particle filters be particularly good for modeling. Why do you think a particle filter would be helpful?

Extras

Extra math

Algorithms for Inference For a somewhat longer, mathier disucssion of MCMC algorithms, see @andrieu2003introduction.