PGMs are essential tools in computer science and artificial intelligence. They're versatile frameworks used to model complex systems involving uncertainty. Example applications include medical diagnosis and image segmentation. These applications share commonalities: a multitude of variables to reason about and inherent uncertainty, making it challenging to determine the correct answers.

Understanding Probabilistic Graphical Models

• Models: Declarative representations capturing interactions among variables. Stand-alone representations can be applied to various algorithms or learning methodologies separately.
• Probabilistic: PGMs handle uncertainty arising from partial knowledge, noisy observations, modeling limitations, and inherent stochasticity, using probability theory.
• Graphical: Involves ideas from computer science, using graphs to represent systems with numerous variables efficiently.

Key Components Covered in the Course

• Representation: Directed and undirected graphical representations, encompassing temporal and relational structures (plate models).
• Inference: Exact and approximate reasoning methods to derive probabilities and trade-offs between accuracy and computation.
• Decision Making: Utilizing PGMs for making decisions amidst uncertainty.
• Learning from Data: Learning model parameters and structures from historical statistical data, including scenarios with incomplete observations.

Representation

Bayesian Networks

Markov Networks

Utilities and Decision