COMP9418 — Advanced Topics in Statistical Machine Learning at UNSW

I took COMP9418 in 2026 T2 as part of my Masters of Statistics. Despite the broad name, the course is really one long story about probabilistic graphical models: how to encode a joint distribution as a graph, how to answer queries against it exactly and approximately, and how to learn it from data. If you are deciding whether to enrol, or you are mid-term and drowning, this page collects everything I wish I had on day one.

what the course actually covers#

The lecture sequence follows Darwiche’s book fairly closely:

1. probability refresher, then Bayesian networks — factorisation, conditional probability tables, d-separation.
2. exact inference — variable elimination, elimination orders and treewidth, the jointree (junction tree) algorithm, MPE/MAP queries.
3. Markov networks — Gibbs distributions, factor graphs, moralisation, conditional random fields.
4. sequence models — Markov chains, hidden Markov models, the forward and Viterbi algorithms, particle filtering, dynamic Bayesian networks.
5. approximate inference — rejection sampling, likelihood weighting, Gibbs sampling / MCMC, loopy belief propagation.
6. learning — maximum-likelihood estimation of CPTs, smoothing, EM for missing data, and structure learning (Chow–Liu trees onward).

I wrote up each of those blocks as a standalone explainer in my wiki while studying, and they double as revision notes for the exam:

• probabilistic graphical models — the section index
• bayesian networks
• exact inference
• markov networks
• markov chains and hidden markov models
• approximate inference
• learning graphical models

textbooks#

The prescribed text is Modeling and Reasoning with Bayesian Networks (Adnan Darwiche, Cambridge, 2009). It is the one book you actually need — the tutorials and the exam lean on its notation and its exercises.

The recommended list, for depth or a second angle:

• Probabilistic Graphical Models: Principles and Techniques. Daphne Koller and Nir Friedman. MIT Press, 2009.
• Probabilistic Graphical Models: Principles and Applications. Luis Enrique Sucar. Springer, 2015.
• Bayesian Reasoning and Machine Learning. David Barber. Cambridge, 2012.
• Machine Learning: A Probabilistic Perspective. Kevin P. Murphy. MIT Press, 2012.
• Pattern Recognition and Machine Learning. Christopher M. Bishop. Springer, 2006.

assessment and the open-book final#

The piece of admin that should shape your whole term: in the 2026 offering the final exam was 3 hours, in-person, and open book — the textbook, the lecture slides, and all tutorial and past-exam solutions were available on the exam machine — with a 40% hurdle on the final. Check the current outline before relying on this, formats drift year to year.

An open-book exam changes the optimal strategy. You are not memorising the jointree algorithm; you are making sure you can execute it quickly and find things fast. My advice:

• work every tutorial to completion during term — the tutorials are theory problem sets, and the exam questions rhyme with them heavily.
• do the past papers under time pressure, but note that only the most recent papers match the open-book format.
• organise your materials so that any definition or algorithm is findable in under thirty seconds. Condensed notes (like the wiki pages above) beat raw slide decks.

should you take it?#

If you want the probabilistic foundations under modern machine learning — the machinery that Kalman filters, HMM taggers, and diffusion-adjacent latent-variable models all stand on — yes. It rewards mathematical maturity more than coding chops: the assignments involve programming, but the hard part is always the probability. If you are after a survey of deep learning, this is not that course.

flashcards#

the 195-card COMP9418 Anki deck built alongside these notes — probabilistic graphical models, exact and approximate inference, HMMs/Kalman, with interactive python code cards — is for sale on the bazaar (A$5.99, .apkg, no LaTeX install needed).