Dr. Rina Dechter @ UCI

Dr. Rina Dechter - University of California at Irvine
home \| publications \| book \| courses \| research	Revised on Sep. 16, 2020


CompSci-276 Winter 2020, Reasoning in Graphical Models

[ main \| software \| references ]

Course Project

The project information is available at: Project Information
The deadline for proposals is February 19th.

Course Reference

Instructor:

Days:

Time:

Room:

Office hours:

Course Description

One of the main challenges in building intelligent systems is the ability to reason under uncertainty, and one of the most successful approaches for dealing with this challenge is based on the framework of Bayesian networks, also called graphical models. 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.

The objective of this class is to provide an in-depth exposition of knowledge representation and reasoning under uncertainty using the framework of Bayesian networks. Both theoretical underpinnings and practical considerations will be covered, with a special emphasis on dependency and independency models, on construction Bayesian graphical models and on exact and approximate probabilistic reasoning algorithms. Additional topics include: causal networks, learning Bayesian network parameters from data and dynamic Bayesian networks.

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 essential.

Course material

The course will be based mostly on three sources:

Additional sources:

Russell and Norvig, Artificial Intelligence: A Modern Approach
Daphne Koller and Nir Friedman, Probabilistic Graphical Models: Principles and Techniques

A longer list including secondary references.

Some links to software and tools.

Course Topics

Introduction: probabilistic graphical models.
Bayesian and Markov networks: Representing independencies by graphs.
Building Bayesian networks.
Inference in Probabilistic models: Bucket-elimination (summation and optimization), Tree-decompositions, Join-tree/Junction-tree algorithm.
Graph properties: induced-width, tree-width, chordal graphs, hypertrees, join-trees.
Search in Graphical models: AND/OR search Spaces for likelihood, optimization queries.
Learning graphical models.
Approximate Bounded Inference: weighted Mini-bucket, belief-propagation, generalized belief propagation.
Approximation by Sampling: MCMC schemes, Gibbs sampling, Importance sampling.
Causal graphical models.

Syllabus

Week	Date	Topic	Readings	Slides/HW
Week 1	1/6	Introduction and Background.	(a) Pearl 1-2 (b) Darwiche 1-3 (c) Russell-Norvig 13 (d) Darwiche. Bayesian Networks	Slides 1
	1/8	Bayesian and Markov networks: Representing Independencies by graphs.		HW 1
Week 2	1/13			Slides 2
	1/15	Building Bayesian networks.	Darwiche Ch. 5	Slides 3
Week 3	1/20	>>> No Class: MLK Holiday <<<
	1/22	Probabilistic Inference: Bucket-elimination (summation, optimization).		Slides 4a HW 2
Week 4	1/27	Local structures CPTs and Induced width algorithms.		Slides 4b
	1/29	Probabilistic Inference: Tree-decompositions: Join-tree/Junction-tree algorithm. Cluster tree elimination.	Dechter Ch. 4, Darwiche Ch. 6
Week 5	2/3		Dechter Ch. 5, 7.1, Darwiche Ch. 7-8	Slides 5 HW 3
	2/5	Probabilistic Inference by search: AND/OR search spaces.	Dechter Ch. 6-7
Week 6	2/10	>>> No Class <<<
	2/12			slides 6 HW 4
Week 7	2/17	>>> No Class: President's Day <<<
	2/19	Learning graphical models.	Darwiche Ch. 17	Slides 7
Week 8	2/24	Approximate algorithms by Bounded Inference.	Dechter Ch. 8-9 Darwiche Ch. 14	Slides 8 HW 5
	2/26
Week 9	3/2	Approximate Algorithms by Sampling: MCMC schemes.	Darwiche Ch. 15 Paper: Cutset-Sampling	Slides 9b
	3/4			Slides 9b
Week 10	3/9
	3/11	Project presentations
	3/13	Project presentations

Assignments:

There will be homework assignments and students will also be engaged in projects.

Grading Policy:

Homework (70%), class project (30%)