Questions: Factor Graphs and Inference

5 questions to test your understanding

Score: 0 / 5
Question 1 Multiple Choice

A Markov random field has a clique of three variables {A, B, C} with a single joint potential ψ(A,B,C). How does this appear in a factor graph?

AAs a triangle of three variable nodes with edges between them — the same structure as the MRF clique
BAs one factor node connected by edges to three variable nodes (A, B, C)
CAs three separate factor nodes, one per variable, each connected to the others
DAs a single variable node labeled ABC representing the joint state
Question 2 Multiple Choice

On a tree-structured factor graph, what does the sum-product algorithm guarantee?

AApproximate marginals that converge after enough iterations
BExact joint distribution over all variables in the graph
CExact marginal distributions for all variables, computed in a single forward-backward message-passing pass
DThe most probable assignment of all variables via dynamic programming
Question 3 True / False

Loopy belief propagation on a factor graph with cycles usually fails to converge and can seldom produce useful results.

TTrue
FFalse
Question 4 True / False

A factor graph can represent any distribution expressible as either a Bayesian network or a Markov random field.

TTrue
FFalse
Question 5 Short Answer

What ambiguity in Markov random field representations do factor graphs resolve, and how do they resolve it?

Think about your answer, then reveal below.