Graphical Models Quiz
Check your understanding of probabilistic graphs — what the DAG encodes, when colliders block or open paths, why the MRF partition function is the hard part, and which HMM algorithm answers which question.
Exercise
What does a Bayesian network's DAG structure encode?
Need a hint?
Missing edges are the informative part.
Exercise
In the collider structure A → C ← B, when does information flow between A and B?
Need a hint?
This is the 'explaining away' pattern — it's backwards from chains and forks.
Exercise
In a Markov Random Field, computing the unnormalized score of one configuration is cheap. What makes probabilities, marginals, and likelihood gradients hard?
Need a hint?
What does the denominator of P(x) = (1/Z)∏ψ sum over?
Exercise
What two assumptions define a hidden Markov model?
Need a hint?
One about state transitions, one about emissions.
Exercise
And the forward algorithm computes…
Need a hint?
It marginalizes over all possible hidden paths.
Exercise
What does the Viterbi algorithm find?
Need a hint?
Compare with the forward algorithm, which sums over paths.
Exercise
Encoding conditional-independence assumptions as a graph structure (a Bayes net DAG or MRF) is a choice in which slot of the ML project loop?
Need a hint?
The edges define which variables can interact — the family of distributions the model can represent.
Exercise
Transfer test — a technique this course did NOT teach: adding an L1 penalty to a loss to encourage sparse weights. Which slot of the ML project loop does it modify?
Need a hint?
It changes the quantity being minimized.