Probabilistic Graphical Models

By the end, you can

• Explain why the full joint distribution is intractable and how a PGM compresses it.
• Identify the parents of a node in a Bayesian network and state what its CPT encodes.
• Apply the chain-rule factorization to compute a joint probability from a set of CPTs.
• Use the law of total probability to compute a marginal prior from CPT values.
• Describe inference as backward propagation from observed evidence to hidden causes.
• Explain the explaining away effect and identify the collider structure that causes it.
• Distinguish Bayesian networks, Markov random fields, and hidden Markov models by their edge directionality and typical applications.
• State the computational complexity of exact inference and name the two main approximate alternatives.