Probabilistic Graphical Models

Studiewijzer 2025/2026

Globale inhoud

In almost every practical situation in which a person or automated system needs to reason they face uncertainty---that is, the available information is not sufficient for them to reach a unique conclusion about what might be true in the world or how to act. Probabilistic graphical models (PGMs) offer a general framework that can be used to allow a computer system to support reasoning and decision making under uncertainty. In PGMs, we use a declarative, graph-based representation to build a transparent model of the system about which we would like to reason. This model encodes our knowledge of how the system works in a computer readable form, which can be manipulated by various algorithms that can answer questions based on the model. Moreover, this representation can be optimised to reflect data and prior experience. PGMs find wide applicability across domains (e.g., computer vision, natural language processing, medical informations, etc.) and their unmeasurable impact in machine learning and artificial intelligence (from forescasting, to unsupervised learning, to decision making, to causal reasoning) extends from the earliest days of the theory to modern-day research.

This course will cover the foundations of Representation, Inference and Learning in PGMs. This is not a programming course, but it does have a programming component, where we demonstrate the theory as well as applications in different domains of AI. 

Textbook: Koller and Friedman’s Probabilistic Graphical Models – Principles and Techniques.

Leerdoelen

• The student can illustrate the wide applicability of PGMs and how they are relevant to a broad number of ideas and applications in machine learning and artificial intelligence.
• The student can interpret the relation between statistical independence and the graphical representations thereof, and is able to draw conclusions from this relation.
• The student can represent domain knowledge in the form of (directed and undirected) probabilistic graphical models over collections of continuous and discrete random variables.
• The student can use a probabilistic graphical model to evaluate probability queries (i.e., perform inference) involving conditioning and marginalisation.
• The student can obtain data-driven estimates for the parameters of a given probabilistic graphical model using complete and/or partial observations.
• The student can implement (i.e., represent, perform inferences with and learn parameters for) several probabilistic graphical models (e.g., naive Bayes models, Markov models, generalised linear models, mixture models, hidden Markov models, Markov random fields, conditional random fields, mixed membership models) across a variety of applications (e.g., classification, segmentation, forecasting) using a programming language (Python).
• The student can compare several probabilistic graphical models across a variety of applications (e.g., classification, segmentation, forecasting) and discuss their advantages and shortcomings.
• The student can augment PGMs (representation, inference and learning) with unobserved random variable.
• The student can impose causal semantics on a directed graphical model (i.e., attach causal significance to directed dependence).
• The student can use probabilistic inference to support decisions under uncertainty using the principle of maximisation of expected utility.
• The student can extend PGMs (representation, inference and learning) to include prior knowledge in the form of distributions over its parameters.

Onderwijsvormen

• Hoorcollege
• Werkcollege
• Laptopcollege
• Zelfstudie

HCs: we develop the main theory planned for the week and solve some exercises; sometimes there's reading that's necessary before or after the class.

LCs: we translate the main ideas in PGMs to Python code; because PGMs deal with graph structure, probabilistic inference and statistical learning, programming exercises will often help you consolidate your knowledge of the theory. 

WCs:  we work on exam-like exercises, discuss feedback, and sometimes cover some side topics that help you get a good hold of the material. 

Self-study: we indicate reading material, videos (we will try to have our classes recorded too, so this can be of help in self-study), and lists of exam-like exercises.

Verdeling leeractiviteiten

Aanwezigheid

Aanwezigheidseisen opleiding (OER-B Artikel B-4.10):

• Voor sommige studieonderdelen geldt een aanwezigheidsplicht. Indien er een aanwezigheidsplicht geldt, dan staat dit aangegeven in de studiegids. De onderbouwing voor, en invulling van, deze aanwezigheidsplicht kan per vak verschillen, en is opgenomen in de studiewijzer. Wanneer studenten niet voldoen aan deze aanwezigheidsplicht kan het onderdeel niet met een voldoende worden afgerond.
  

Aanvullende eisen voor dit vak:

There is no mandatory attendance in this course, but attendance is highly encouraged and live HCs can contain exercises that are worth bonus points towards the final grade.

We monitor (but do not require) attendance to LCs and WCs. Besides, we keep a rough count of how many people (not whom) are attending HCs. This information helps us estimate the level of engagement with the different learning activities. 

Toetsing

The final grade is made of 2 components:

• exam: average of midterm and endterm exams 
• programming: average of weekly programming assignments

The resit grade substitutes the Final grade as a whole. 

Exams are on-paper and on-site. They are graded on ANS and ANS is also used to discuss the grade and feedback.

Independent assignments and exams are all scaled between 0 and 10 and rounded to a single decimal point. Components (e.g., combination of exam, combination of assignments) are not further rounded. The final grade is rounded and clipped for SIS (where the minimum grade is 1, the maximum grade is 10, and the precision is of half point except between 5 and 6 where no decimals is allowed).

Further details about assignments and/or bonus points will be officially documented and announced on Canvas.

Inzage toetsing

All feedback will be available on ANS, where discussions are possible for a period of time after publication of results. 

Fraude en plagiaat

Dit vak hanteert de algemene 'Fraude- en plagiaatregeling' van de UvA. Hier wordt nauwkeurig op gecontroleerd. Bij verdenking van fraude of plagiaat wordt de examencommissie van de opleiding ingeschakeld. Zie de Fraude- en plagiaatregeling van de UvA: http://student.uva.nl

Weekplanning

Honoursinformatie

NA

Contactinformatie

Coördinator

• W. Ferreira Aziz