Statistical Relational Artificial Intelligence (StaRAI)

Overview

In systems that exhibit artificial intelligence (AI), an agent at its centre has to learn and represent a model of its environment, reason about it, and decide on its actions. A possible approach to realising such a rational agent lies in probabilistic graphical models, which allows for modelling uncertainty. Another approach lies in first-order logics, which allows for modelling objects and relations. Methods from the field of statistical relational AI (StaRAI) combine both aspects, dealing with uncertainty and relations. This lecture focusses on the beginnings as well as the newest developments within StaRAI, covering the different aspects of the tasks an agent has to solve.

This new lecture will be heavily based on the lecture part on Probabilistic Graphical Models in the module Intelligent Agents given at the University of Lübeck, linked here.

For this course, a good understanding of probability theory and logics is beneficial. We will cover some basics again but it will probably be too fast without any prior knowledge of these. Although we will also cover probabilistic graphical models to the extent that we need them for StaRAI models, a very detailed introduction into probabilistic graphical models can be found in the course Introduction to Artificial Intelligence here (course material in German only; literature in English).

Please register in the Learnweb course for more details and material once it exists.

Presentation Material

Topics and slides:

  1. Introduction (pptx, pdf)
  2. Foundations:
    1. Logic (pptx, pdf)
    2. Probability theory (pptx, pdf)
    3. Probabilistic graphical models (pptx, pdf)
  3. Probabilistic relational models (pptx, pdf)
  4. Lifted inference
    1. Exact inference
      1. Lifted variable elimination (pptx, pdf)
      2. Lifted junction tree algorithm (pptx, pdf)
      3. First-order knowledge compilation (pptx, pdf)
    2. Approximate inference (pptx, pdf)
  5. Lifted learning (pptx, pdf)
  6. Lifted sequential modelling and inference (pptx, pdf)
  7. Lifted decision making (pptx, pdf)
  8. Continuous space and lifting (pptx, pdf)
  9. End (pptx, pdf)

Literature

The lecture is heavily based on research papers and PhD theses on the topics mentioned above and will be referenced on the slides. For more a more general introduction and details on propositional modelling and inference, you can refer to the following books:

  • Modelling and Reasoning with Bayesian Networks, Adnan Darwiche
  • Probabilistic Graphical Models, Daphne Koller and Nir Friedman
  • Artificial Intelligence - A Modern Approach (3rd ed.), Stuart Russel and Peter Norvig

All sources should be available (i.e., find-able) online.