You are on page 1of 2

In-class Activity

ME 570 - Prof. Tron


2020-11-24

Problem 1: Conditional statistical independence


Consider a probability space (Ω, F, P ) where P (A) = µ(A)
µ(Ω)
, µ is a measure such that one
square has unit area, that is, µ( ) = 1, and consider also the events (sets of outcomes)
A, B, C ∈ F as represented in the figure (note that B is a subset of A).

B
C

In the following, when asked to “verify” a relation, first state the formula, then check with
the numerical values.
1) Compute the following probabilities

Event P (Event) Event P (Event)

A A|B

C C|B

A, C A, C | B

1
2) Verify that A and C are not statistically independent (A 6⊥ C), but they are statistically
independent given B (A⊥C | B).

3) Show that A⊥C | B implies P (A | B, C) = P (A | B) (i.e., we can “forget” C if we


have B).

4) Assume that P (A, B, C) = P (A|B)P (B|C)P (C). This can be represented via the
following Directed Acyclic Graph (DAG):

C B A

Show that this assumption implies A⊥C | B.

5) Assume that A can assume values in the set {ai }ni=1 . Write Bayes’ rule for computing
the posterior P (A = a | B = b) from the prior P (A = a) (where a ∈ {ai }ni=1 ) given the
transition probabilities P (B = b|A = a).

You might also like