Fall 2008 Graduate CSE Courses

Last Update: July 29, 2008

Note: NEW or UPDATED material is highlighted

THIS PAGE ONLY LISTS THOSE COURSES FOR WHICH INSTRUCTORS HAVE SENT ME COURSE DESCRIPTIONS.

FOR THE FULL LIST OF C.S.E. COURSES FOR Fall 2008, link to MyUB.

PLEASE CONTACT THE INSTRUCTOR FOR FURTHER INFORMATION ABOUT ANY OF THESE COURSES

CSE 663

TOPIC: ADVANCED TOPICS IN KNOWLEDGE REPRESENTATION & REASONING
INSTRUCTOR: William J. Rapaport
DAY & TIME: MWF 11:00-11:50 a.m.
DESCRIPTION: This course is a sequel to Prof. Shapiro's CSE 563 from the Spring 2008 semester. It will be a survey of issues and techniques of representing knowledge, belief, and information in a(n artificially intelligent) computer system and of the syntax and semantics of various representational formalisms. Classic papers will be read and current research issues discussed. I will begin with a brief review of logic and automated theorem proving (unification and resolution) and of the SNePS knowledge-representation, reasoning, and acting system. Remaining topics will include some or all of the following, as well as others as time permits: ontologies, semantic networks, production systems, frames, description logics, inheritance networks, default reasoning, and modal and epistemic logics.
PREREQUISITES: Official:
Graduate standing and either CSE 563 (Knowledge Representation) or CSE/LIN 567 (Computational Linguistics); or else permission of instructor.

Unofficial:
Knowledge of first-order logic, and some familiarity with resolution and unification (such as might have been obtained in a previous AI course, CSE 563, or—for unification, at least—in CSE 567). If you did not take CSE 563 in Spring 2008 and/or have no background in first-order logic, including unification and resolution theorem proving, then please see Prof. Rapaport before registering.
WEB PAGE: Will eventually be available here; till then, see the website for the previous incarnation of the course.

CSE 694: PROBABILISTIC ANALYSIS AND RANDOMIZED ALGORITHMS

TOPIC: How probability theory and randomness help solve problems in various areas of Computer Science
INSTRUCTOR: Hung Q. Ngo
DAY & TIME: Tue-Thur, 9:30-10:50, 260 Capen
DESCRIPTION: Probabilistic analysis and randomized algorithms have become an indispensible tool in virtually all areas of Computer Science, ranging from combinatorial optimization, machine learning, approximation algorithms analysis and designs, complexity theory, coding theory, to communication networks and secured protocols. This course has two major objectives: (a) it introduces key concepts, tools and techniques from probability theory which are often employed in solving many Computer Science problems, and (b) it presents many examples from three major themes: machine learning, randomized algorithms, and combinatorial constructions and existential proofs.
In addition to the probabilistic paradigm, students are expected to gain substantial discrete mathematics problem solving skills essential for computer scientists and engineers.
PREREQUISITES: Basic Algorithm course (CSE 531), a good sense of discrete mathematics thinking, and rudimentary knowledge of probability theory.
WEB PAGE: http://www.cse.buffalo.edu/~hungngo/classes/2008/694.

NEWCSE 702: Seminar in Pattern Theory

Instructor: Jason Corso
Day and Time: TBA
Description: This seminar will focus on Grenander's Pattern Theory from a practical, contemporary perspective. Pattern Theory is the study of patterns from a representational perspective rather than a recognition one. Miller and Grenander write "Pattern theory attempts to provide an algebraic framework for describing patterns as structures regulated by rules, essentially a finite number of both local and global combinatory operations. Pattern theory takes a compositional view of the world, building more and more complex structures starting from simple ones. The basic rules for combining and building complex patterns from simpler ones are encoded via graphs and rules on transformations of these graphs." We will explore various theoretical aspects of modern pattern theory (e.g., probabilistic graphical models, grammars, matrix groups, information measures, manifolds, Markov processing and sampling) in the context of practical problems in computer vision and medical imaging. Students will be required to give one or two (depending on seminar size) prepared lectures during the semesters. Grading is S/U; letter grading is available as an option and requires a project.
PREREQUISITES: A working knowledge of computer vision, pattern recognition, and machine learning is suggested. Students are expected to know material in courses 555, 573, 574 and 672.
WEB PAGE:http://www.cse.buffalo.edu/~jcorso/t/2008F_702

NEWCSE 704: EXPANDERS, PROPERTY TESTING AND THE PCP THEOREM - I

TOPIC: Expanders and Property Testing
INSTRUCTORS: Hung Ngo and Atri Rudra.
DAY & TIME:TBD
DESCRIPTION: One of the crown jewels of theoretical computer science is the PCP theorem. Roughly speaking, the theorem says that there is a way to write a (mathematical) proof which allows readers to confidently verify the correctness of the proof by just reading a few words from the proof. (It is not necessary to go through hundreds of notation-heavy pages to be sure that Andrew Wiles did have a valid proof of Fermat's last theorem!)

More rigorously, the PCP theorem states that, for any NP-language L the membership for any input in L can be verified by inspecting only a *constant* number of positions in a "proof". (By contrast, recall that in the "traditional" definition of NP, the membership of any input in L is "witnessed" by a proof, e.g. a satisfying assignment for a 3SAT formula. However, "verifying" the proof would entail inspecting the *entire* proof, which could be polynomially large in the input length.)

In addition to be being a philosophical shift in how we view the complexity class NP, the PCP theorem brought about a revolution in proving NP-hardness of even *approximating* NP-hard problems. (For example, a work of Johan Håstad showed that in general it is NP-hard to come up with an assignment that satisfies even 7/8 fraction of clauses. Note that the Cook-Levin theorem just states that it is NP-hard to satisfy *all* the clauses.)

The original proof of the PCP theorem was due to the seminal work of Sanjeev Arora, Carsten Lund, Rajeev Motwani, Madhu Sudan (who was a distinguished speaker in our department in Spring 2008) and Mario Szegedy (they co-won the Goedel prize in 2001). The proof was highly intricate and very algebraic and it is virtually impossible to cover the complete proof even in a semester dedicated to the proof. However, in 2006 Irit Dinur gave a much simplified and combinatorial proof of the PCP theorem (her paper won the best paper award at STOC 2006).

Dinur's proof uses "expanders" and "property testing." Expander are sparse graphs that have good connectivity properties. Expanders have numerous applications from networks to pseudorandomness. Property testing is the following algorithmic paradigm. Fix some property P, and design an algorithm that (ideally) with a constant number of queries decide whether the input satisfies P. Given that the algorithm cannot even inspect the entire input, it is not surprising that for many non-trivial properties P, there do not exist such algorithms. However, in property testing, the algorithm is allowed to give the following relaxed answer: accept the input if it satisfies P and reject the input if it is "far" from satisfying P (for other cases, the algorithm can behave arbitrarily).

The aim of this seminar is to present the entire proof of Dinur (and other implications of the PCP theorem, esp. in hardness of approximation) along with a thorough treatment of the background needed to present the proof. However, to maintain a reasonable pace, we will in this seminar only cover expanders and property testing (and their applications). We will talk about the PCP theorem and Dinur's proof in Spring 2009. Students are encouraged to take the seminar in both the semesters. However, the material presented in Fall 2008 will be self-contained and the material is very important in its own right.

The first few lectures will be presented by the instructors followed by student presentations. Few of the meetings might also play host to some external speakers talks as part of the UB theory seminar which got re-started from Spring 08.

Registering for the seminar: Atri Rudra and Hung Ngo are the instructors in charge of the seminar. However, the "official" seminar for Fall 2008 is Atri Rudra's CSE 704, so students should sign up for CSE 704.
Work load: Students will be evaluated via class participation, scribe note taking, and presentation. We will provide a suggestive list of papers for students to choose from. However, students can choose their own papers related to the seminar to present with consent from one of the instructors.

PREREQUISITES: Familiarity with Algorithms and NP-completeness. Mathematical maturity. If you have questions, contact the instructors.
WEB PAGE: Eventually, it will be up here. The course blog will eventually be up here.

NEWCSE 719: Seminar on Security and Economic Incentives

TOPIC: Security, Privacy Issues, and Economic Incentives in Wireless Networks, Data Mining, etc.
INSTRUCTOR: Sheng Zhong
DAY & TIME: Thu 6:30-9:30
DESCRIPTION: With rapid advances of Internet technology, security concerns and economic incentives have become crucial in the world of computing. In this seminar, we discuss recent research results and problems in the areas of computer security and economic incentives. Topics to be covered include, but are not limited to: privacy-preserving data mining, privacy in databases, incentive-compatible mobile computing, wireless network security, computational game theory, and applied cryptography. We stress on a balance of practical usefulness and theoretical depth; most papers covered in this class attempt to solve practically interesting problems using mathematically rigorous methods. This is a research seminar with the target of exposing students to frontier research. There are no homeworks or exams. A number of lectures are given by the instructor and other classes are student presentations and discussions. Each student is also required to study a topic assigned by the instructor and submit a written report.
PREREQUISITES: None
WEB PAGE: http://www.cse.buffalo.edu/~szhong/courses/719