COMP9418 Advanced ML — Anki deck

What it is

The card deck I built while taking UNSW's COMP9418 (Advanced Machine Learning — probabilistic graphical models): Bayesian and Markov networks, d-separation and graphoids, variable elimination, treewidth and jointrees, MAP/MPE, sampling and MCMC (with the variance analysis that tells you when an estimate is trustworthy), loopy belief propagation, HMMs and Kalman filtering with Gaussian factors, and graph neural networks.

What's inside

• 195 notes / 200 cards, tagged by topic so you can filter any slice (sampling, belief-propagation, kalman, gnn, jointree, …).
• ~36 implementation cards in Python, written against a from-scratch Factor/BayesNet codebase — 21 of them interactive type-in cards that diff the line of code you type against the answer.
• 12 network diagrams, shipped as pre-rendered SVGs (no LaTeX install needed); all other maths renders via Anki's built-in MathJax.
• Custom styled note types travel inside the .apkg and install on import.
• No review history — every card arrives fresh for your own schedule.

How the cards are written

One fact per card, interpretation first: cards ask what marginalisation does, why a convergence guarantee holds, which operation people forget — not just for symbol recall. Everything is phrased in my own words from my own study notes.

Pairs well with

My free probabilistic-graphical-models wiki notes at abaj.ai/wiki/ml/pgm — the cards drill what the notes explain.