Dr. Rina Dechter @ UCI

Dr. Rina Dechter - University of California at Irvine
home \| publications \| book \| courses \| research	Revised on Nov. 11, 2024


CS 276 - Causal and Probabilistic Reasoning with Graphical Models 2023-24, Winter (Q2)

[ main \| software \| references ]

Course Reference

Instructor:

TA:

Days/Time:

Room:

Zoom:

https://uci.zoom.us/j/93822218749

Ed Discussion:

https://edstem.org/us/courses/49867/discussion/

Office hour:

TA Office hour:

Course Description

One of the main challenges in building intelligent systems is the ability to perform causal inference under uncertainty. Graphical models, which include Bayesian networks and their extension into structural causal models offer a powerful and successful methodology of operationalizing causal inference in a wide spectrum of applications. Intelligent systems based on Bayesian networks are being used in a variety of real-world applications including diagnosis, sensor fusion, on-line help systems, credit assessment, bioinformatics and data mining. Causal reasoning is at the core of human reasoning and thus should play a central part in artificial intelligence and in machine learning.

The objective of this course is to provide an in-depth exposition of causal reasoning under uncertainty using structural causal models and Bayesian networks. Both theoretical underpinnings and practical considerations will be covered, with a special emphasis on exploring reasoning within the ladder of causation that include 1. Association, 2. Intervention, 3. Counterfactual.

Prerequisites

Familiarity with basic concepts of probability theory.
Knowledge of basic computer science, algorithms and programming principles.
Previous exposure to AI is desirable but not required.

Target Students:

This course is intended for PhD students in the area of AI, Machine Learning, and Statistics, and also appropriate for students from other disciplines with interest in causality and having relevant AI/ML/Statistics background.

Course material

The course will be based on the following primary sources:

Additional sources:

Course Topics

Probabilistic Graphical Models, Structural causal models,The Causal Hierarchy.
Bayesian and Markov Networks: Representing independencies by graphs. d-seperation.
Algorithms for probabilistic reasoning (Bucket-elimination for summation and optimization, Join-trees, The induced-width.).
Sampling schemes for graphical models (MCMC and Importance Sampling).
Structural Causal Models; Identification of Causal Effect; The problem of confounding.
The Back-Door and Front-Door Criteria and the Do-Calculus.
Linear Causal Models.
Counterfactuals.
Algorithms for identification using c-components. The ID algorithm.
Learning Bayesian networks and Causal graphs (causal discovery).

Homework Assignments:

There will be four homework, each given roughly a week to complete.

Course Project:

Each student will also be engaged in a project based on papers from recent literature. The project will involve learning about and preseting an assigned paper/literature in class and writing a project report.

Project Information
Project Sign-Up Sheet

Grading Policy:

Homework (70%), Course Project (15% presentation + 15% report = 30% total)

Syllabus

s

Week	Topic	Lectures	Slides	Homework	Reading	Date

Week 1	Probabilistic Graphical Models and Causal Bayesian Networks Bayesian and Markov Networks: Representing independencies by graphs	Lec 1 Lec 2	Slides 1 Slides 2		[Darwiche] Ch. 1, 3, 4 [Russell-Norvig] Ch. 13	M 1/8 W 1/10

Week 2	Bayesian and Markov Networks: Representing independencies by graphs The notion of d-separation	Lec 3	Slides 3		[Darwiche] Ch. 4	M 1/15 (No Class!!!) W 1/17

Week 3	Variational algorithms for probabilistic graphical models (Bucket-elimination for summation and optimization) Join-trees; Belief propagation; The induced-width	Lec 4 Lec 5	Slides 4 GenIE Slides 5		[Dechter] Ch. 4, 5.1-5.2 [Darwiche] Ch. 6	M 1/22 W 1/24

Week 4	Structural Causal Models; Identification of Causal Effect; The problem of confounding	Lec 6 Lec 7	Slides 6 Slides 7		[Dechter] ch. 3.4 [Primer] ch. 1, ch 2	M 1/29 W 1/31

Week 5	The Back-Door Criterion	Lec 8 Lec 9	Slides 8 Slides 9		[Primer] Ch. 3 [causality] 1.3, 3.1-3,3	M 2/5 W 2/7

Week 6	Front-Door Do-Calculous	Lec 10 Lec 11 Lec 12	Slides 10		[Primer] ch. 3.4-3.6 [Causality] 3.4	M 2/12 W 2/14

Week 7	Algorithms for identification using c-components	Lec 12	Slides 11		[Primer] Ch. 4 [Causality] Ch. 7	M 2/19 (No Class!!!) W 2/21

Week 8	Linear Structural Causal Models Counterfacturals	Lec 13 Lec 14	Slides 12 Slides 13		_	M 2/26 W 2/28

Week 9	Learning graphical models	Lec 15 Lec 16	Slides 14		_	M 3/4 W 3/6

Week 10	Student Project Presentations	Lec P1 Lec P2	Markov Equivalence in Cyclic Directed Graphs Causal Information Splitting Neural Probabilistic Logic Programming in Discrete-Continuous Domains Causal Relation Learning Causal Effect Estimation Counterfactual Image Editing			M 3/11 W 3/13

Finals Week	Project Reports Due					F 3/22, 8-10am