Inference

Table of contents

Inference corresponds to using the distribution to answer questions about the environment.

E.g.:

  • What is the probability \(p(x = 4 \mid y=1,z=2)\)?
  • What is the most likely joint state of the distribution \(p(x,y)\)?
  • What is the entropy of the distribution \(p(x,y,z)\)?
  • What is the probability that this example is in class 1?
  • What is the probability the stock market will go down tomorrow?

Why do we use inference? Inference is in general computationally very expensive and we wish to characterise situations in which inferences can be computed efficiently. For singly-connected graphical models (trees), and certain inference questions, there exist efficient algorithms.

We are going to look at: