CSCI 573 Advanced Artificial Intelligence

Prof. Ram Nevatia

Phone :

(213) 740-6427

Email :

nevatia AT usc DOT edu

Class Time :

11:00 A.M. -12:20 P.M.

Class Location :

OHE100D

Office Hours:

2:00 P.M. -3:30 P.M. MW (and By appointment)

Office Location:

PHE 204

TBD

Phone :

(213)740-xxxx

Email :

csci573@aludra.usc.edu

Office Hours :

TBD – TBD

Office Location :

TBD

Required:

1. Artificial Intelligence : A Modern Approach, S. Russell and P. Norvig , Prentice-Hall, Second Edition
2. Some papers that will be made available to the class.

1.

CSCI 455 or equivalent - Data Structures, good programming skills. Ability to convert informal descriptions into computer algorithms. Students must be able to program in C++.

2.

A basic course undergraduate or graduate in Artificial Intelligence, such as CSCI 460 or CSCI 561 is required.

3.

The course makes extensive use of probability theory. It is highly desirable that the students have had a course in probability theory even though we will cover the basic concepts in class; however, the pace of development will be fast not as rigorous as in a mathematics/statistics department class.

The objective of this course is to cover the basic topics in uncertain reasoning and machine learning that are commonly used in modern artificial intelligence. Emphasis will be on concepts and algorithms and not on "cookbook" techniques or current commercial systems.

The Web page for this course is http://www-scf.usc.edu/~csci573/. It will be replaced by http://den.usc.edu/ at the start of the semester.

Distance Education Network Web Site

There will be two examinations:

1. First exam, during approximately 7-9 week of the term.
2. Second exam, on the last day of the class (subject to approval of the students in the class).

Each exam will count for about 30% of the course grade. The rest of the grade will be determined by the written and programming assignments. All assignments are considered an integral part of the course and MUST be completed. Not completing assignments may result in a grade of "F" even if the student performance is good in the examinations.

The USC Student Conduct Code prohibits plagiarism. All USC students are responsible for reading and following the Student Conduct Code, which appears on our school Scampus;

In this course we encourage students to study together. This includes discussing general strategies to be used on individual assignments. However, all work submitted for the class is to be done individually, unless an assignment specifies otherwise.

Some examples of what is not allowed by the conduct code: copying all or part of someone else's work, and submitting it as your own; giving another student in the class a copy of your assignment solution; consulting with another student during an exam. If you have questions about what is allowed, please discuss it with the instructor.

Violations of the Student Conduct Code will be filed with the Office of Student Conduct, and appropriate sanctions will be given.

All programming assignments may be completed using C++. In some cases, we will use public domain software systems, these will be made available for installation on campus computers or student's personal computers.

The primary topics to be covered are chapters 13 through 21 in the textbook. Mostly, we will follow the material covered in the text book but in some cases, the material will be supplemented by newer methods, which are available only in the form of research papers or tutorial articles. Following is a list of topics expected to be covered, in anticipated order, and with expected time to be spent on them. This list is intended to be only indicative, the actual topics, the order and the time may vary somewhat depending on various factors including student interests and preparation.

1. Introduction (1 lecture)
   Background, requirements, topics to be covered, conduct of the class

2.      Introduction to Probability Theory (1 week)
Probability definitions, Bayes rule and its applications (chapter 13)

3.      Probabilistic Reasoning (2.5 weeks)
Bayesian networks: representation and inference, Belief Propagation, MCMC algorithm, other methods (chapter 14 and additional material)

4.      Probabilistic Reasoning over Time (2.5 weeks)
Hidden Markov Models, Dynamic Bayesian networks (chapter 15 and additional material)

5.      Probabilistic Reasoning over Time (1 week)
Utility theory, Decision networks (chapter 16)

6.      Making Complex Decisions (1.5 week)
Sequential decision problems, Partially observable Markov decision problems (POMDPs) (chapter 17, augmented by additional material)

7.      Learning from Observations (1 week)
Inductive learning, decision trees, ensemble learning (chapter 18)

8.  Statistical Learning (2.5 weeks)
Complete data, Hidden nodes (EM method), Instance based learning, Neural networks (chapter 20)

9.  Reinforcement Learning (1 week)
Passive and active (Chapter 21)