CPSC 532 - Topics in AI:
Probabilistic graphical models, knowledge graphs and statistical-relational AI

David Poole

January-April 2021

Announcements

The class will now be held on Wednesdays and Fridays 9:30-11:00.
First class Wednesday January 13th.
Assignment 0 due on Thursday January 14th.
See Wiki for February Assignment/Project. Due on February 12.

This is an advanced AI course on mixing uncertainty and relational models: what should an agent do based on its ability, its beliefs, its perceptions and its values/goals in an uncertain environment that consists of individuals (things, objects) with relationships among them. For example, in a medical diagnosis system, we want to make a probabilistic prediction of the effect of a treatment on a patient, conditioned on the patient's electronic health record (history of doctor's visits, tests, treatments, fitness data, etc). In geology, we might condition on the description of a geological area including sensing data to predict earthquakes,

Background

We will assume the following background (which you can get from courses, MOOCs, books):

Basic AI. The idea of representations, inference, learning.
Probability, including graphical models, to the level of CPSC 322. We will assume you know what a Bayesian network (belief network) is.
Machine learning: you need to know some of the basics, e.g., as covered in CPSC 340.
First-order logic. Preferably some logic programming.

Structure of the Course

Instructor: David Poole, poole@cs.ubc.ca.

There will be 3 hours of in-class interaction per week. The classes will be a mix of lectures on the foundations, student presentations, discussion of research papers and problem solving. This is a participatory class; everyone will be expected to participate fully, to have read the reading material before class, and come ready to discuss and critically analyze it.

The classes will be held:

Wednesdays and Fridays: 9:30-11:00. The first class will be on Wednesday, January 13. See Canvas for Zoom links.

Grades and zoom links will be on Canvas .

Topics

The topics we will cover include semantics, inference and learning for many representations with the goal of inderstanding how they can be combined.

Graphical Models: representations (directed and undirected), inference (exact and approximate), learning, causality.
Relations, knowledge graphs. Learning and predicting relations and properties.
Combinations of probabilistic models and first-order logic.
Relational, identity, number and existence uncertainty
Representations for models and data that semantically interact

Inference: ground and lifted inference; exact, bounded and approximate inference

Learning: learning about individuals and general knowledge, learning probabilities, structure learning

Preferences and utilities. Actions. Relational reinforcement learning (time permitting)

Textbook

The following is a good for the background, but we will go into more detail in some parts and less in others, and also cover more recent topics:

Statistical Relational Artificial Intelligence: Logic, Probability, and Computation Morgan and Claypool, 2016. Full text is available for download from UBC (you may have to use a VPN if off-campus).

For background reading, see CPCS 522 Wiki from some previous versions, which contains background material.

Assessment

The course assessment will be based on presentations, 3 assignments/projects, and reviewing. The 3 assignments/projects will be monthly,using the UBC Wiki (old) as a collaborative teaching/research platform. More details to come.

Last updated: 2021-01-06, David Poole

CPSC 532 - Topics in AI: Probabilistic graphical models, knowledge graphs and statistical-relational AI