UIUC, CS498, Section EA - Autumn 04 - Final Sample (3 hours)
Given by: Eyal Amir
December 13, 2004
1 Short Questions (20%)
Circle the most appropriate answer
1. The Markov Blanket of a variable X within a Bayes Net consists exactly of
(a) parents of X
(b) children of X
(c) (1a) and (1b)
(d) (1a) and (1b) and parents of children of X
2. Let hX1, ..., Xnibe a total ordering of the variables in a Bayes Net, where for all 1 ≤i≤nall the
members of P a(Xi) appear before Xiin the ordering. The following defines the semantics of the Bayes
Net:
(a) P(Xi) = P(P a(Xi))
(b) P(Xi) = P(Xj) for all Xjin P a(Xi)
(c) P(Xi) = P(Xj) for all Xjin the Markov Blanket of Xi
(d) P(X1, ..., Xn) = Qn
i=1 P(Xi|P a(Xi))
3. In a stationary Dynamic Bayes Net with nbinary state variables and a transition model linking every
variable at time t+ 1 to exactly one variable at time t, the number of parameters required for a belief
state at time twith no observations (starting from a fully factored belief state at time 0) is
(a) n
(b) n2
(c) 2n−n
(d) 2n
4. A multi-variate gaussian with nvariables takes the following time to compute a marginal over m
variables
(a) O(mn)
(b) O(m2n)
(c) O(n2m)
(d) O(n3)
(e) other
5. Gibbs sampling may not converge
1
