CSCE421/821, Fall 2002: Foundations of Constraint Processing

Annnouncement: we will try to have at least one researcher from the industry to talk about the market of this technology, techniques implemented in industrial products,  and commercial success stories.

Prereq: CSCE310 (Data structures and algorithms) and CSCE476/876 (Introduction to AI) or permission of instructor.

Course description:   Constraint Satisfaction has emerged as a successful approach to articulate and solve many industrial problems such as design, network diagnosis, vision, scheduling, and resource management in wireless networks.  This technology is now the basis for new programming languages and innovative commercial systems for production scheduling, product configuration, personnel planning and time-tabling, etc.  This course reviews the foundations of constraint satisfaction and the basic mechanisms developed in this area.  It also covers issues on modeling and representation,  special types of constraints such as temporal constraint networks, and theoretical and empirical evaluation  of problem `difficulty.'  The course will examine in particular new methods for decomposition and symmetry identification, designed to overcome the complexity barrier and to support interactions with users.

Time: Tuesday, Thursday, from 9:30 a.m. to 10:45 a.m.

Place: Ferguson 112.

Instructor:   Berthe Y. Choueiry
                      Room 104, Ferguson Hall,
                      email: choueiry@cse.unl.edu, tel: (402)472-5444.
                      Office hours:  Tue/Thu from 10:45 a.m. to 12:00 p.m., or by appointment.

TA:   Mr. Andrew Breiner
         email: abreiner@cse.unl.edu
         Office location: Building 501, Room 6, Desk 14
         Office hours:

• Monday, from 10:00 to 11:00 am,
• Wednesday from 12:00 to 1:00 pm
• by appointment

In an effort to provide you with the best possible support, the following research assistants will be holding office hours:

• Dan Buettner, Tuesday from 2:00 to 3:00 pm, Building 501, Room 3, buettner@cse.unl.edu
• Lin Xu, Thursday from 2:00 to 3:00 pm, Building 501, Room 3, lxu@cse.unl.edu
• Hui Zou, Friday from 11:00 am to 12:00 pm, Building 501, Room 3, hzou@cse.unl.edu

Class schedule: the page will be regularly updated. Check it out regularly for reference to required and recommended reading material, homework texts, and announcements.

Topics include but are not restricted to:

• Properties, computational complexity,  and practical importance.
• Theoretical and empirical classification of backtracking algorithms.
• Theoretical and empirical classification of static and dynamic ordering heuristics.
• Phase transition.
• Global and local consistency: algorithms, properties, and computational complexity.
• Islands of tractability for minimality and global consistency.
• Modeling and reduction methods between representations.
• Decomposition.
• Symmetries and their approximations.
• Reformulation and abstraction.
• Temporal constraint networks.
• Extension: Dynamic/conditional constraint satisfaction,   constraint optimization.
• More, depending on class' interests.

• Edward Tsang, Foundations of Constraint Satisfaction. Available from instructor.
• Rina Dechter, Constraint Processing.  In preparation, check with instructor.
• Technical papers given by instructor or available from the Electronic Reserve at the Love Library.

• Lectures by instructor, twice per week.
• The workload will consist of:
• Required and recommended reading: as  indicated in the Class schedule.   The content of the course will be dynamically adapted to students interests and performance.
• Programming, theoretical, and library-search assignments:   Programming assignments  homework must be turned in using the UNIX handin program on cse.unl.edu and turned in  before class on the due date.  Pen-and-paper assignments must be given to the instructor right before the lecture on the due date. Late homeworks are subject to a 20% deduction per day (including week-ends), any second after the due date counts as an entire day.  Programming assignments can be done in any programming language.  When code and data structures are provided, they will be mostly in Common Lisp (advice: use Allegro Common Lisp on Linux or a sun workstation).  Students are kindly requested to indicate how much time they approximately spend on each exercise; this information will be aggregated and used for planning purposes: it does not affect grading and the evaluation of individuals.
• Surprise quizzes: There will be surprise quizzes throughout the semester  (with a frequency inversely proportional to students'  attendance).  Quizzes will address all material covered during the lectures and/or by the required reading.  No books or personal notes are allowed during the quizzes, unless explicitly specified. Quizzes cannot be made up.
• Tests: There will be a pre-test only.  Test cannot be made up except by instructor's permission.
• Search competition: Students will receive code and data that model a real-world application, all in LISP. Students will choose some search mechanism and implement it.  Results will be compared, winner will receive a book-gift certificate.
• Project: research, report, and  presentation: The project can be either a literature study/survey or the implementation of algorithms.  A presentation to the class is mandatory. The grade will take into account the quality of the research, report, and presentation.
• It is the student's responsibility to ensure an account on  CSE  computers.
• Discussions among students, instructor, and TA are encouraged.  Homeworks however are a strictly individual activity: no sharing is permitted (unless when specified by instructor).  Unethical behavior will be heavily sanctioned (e.g., a null grade on the task, department will be informed). Absolute rule: Always acknowledge any help received from other individuals and always fully reference material used (e.g., encyclopedia, book, paper, journal, web site).
• Absence: maximum 4 sessions, advanced notice required. Attendance is taken.  Students are responsible for the material covered and announcements made during the class.  Also, there will be surprise quizzes.

• Grades will be allocated as follows
• Pretest: 3%.
• Quizzes:  32%
• Assignments: 30%
• Individual project
• Search competition: Solving a real-world optimization problem (in Lisp): 35% (Code: 25%, Report: 10%)
• Project: 35%.  (Research and report: 25%. Presentation: 10%.)

• Make a presentation of a research paper in class: 10% per presentation.
• Make a critical summary of research papers/topics discussed in class: 5% per summary.
• Write a chapter of a textbook: 20% per chapter.
• Do the weekly and final glossaries: 5% (total). Increased to 10% by popular demand.
• Bonuses will be awarded to students who attend all lectures, interact lively, and participate in discussion in class.

Reminder from the College of Arts & Sciences:  a C (2.0) is the minimum passing grade in a PASS/NO PASS course (NOT a C-) as well as the lowest grade one can receive and still count the class toward a major.

Books on reserve at the Love Library (LL):

• Foundations of Constraint Satisfaction by Edward Tsang.

• Artificial Intelligence, A Modern Approach (AIMA), by Russell & Norvig. CALL number Q335 .R86 1995.

• The MIT Encyclopedia of the Cognitive Sciences. CALL  number BF311 .M556 1999, LIB USE ONLY.
• Encyclopedia of artificial intelligence, 1992, SECOND EDITION, call number Q335 .E53, LIB USE ONLY.
• AI Topics of the American Association for Artificial Intelligence
• Web search engines (Google, Altavista, etc.)

Some industrial companies and start-ups implementing advanced constraint-based systems: (listed in random order)

• COSYTEC,  I2 Technologies,  PeopleSoft/Red Pepper, J.D. Edwards/Numetrix (production scheduling),  Blue Pumpkin, Ilog, Trilogy, Parc Technologies Ltd, Carmen Systems, Firepond, On Time Systems Inc., etc.

• Constraint Networks by Rina Dechter
• Constraint Programming Approach to AI Applications, by Michela Milano
• Foundations of Constraint Satisfaction, ESSLI 2002, by Roman Barak
• Constraint Programming: online textbook by Roman Barták.
• Reasoning Methods in AI (CS329A, CS227): courses by Pandu Nayak.
• Constraint Satisfaction Problems (CSC2512F): course by Fahiem Bacchus
• Tutorial on Constraint Programming, by Barbara Smith.
• Lecture notes by David McAllester.
• Bacchus' Constraint Programming Bibliography

• LIA@EPFL
• APES Group in the UK.
• Constraint Computation Center of Freuder in Univ. of New Hampshire.
• Cork Constraint Computation Center (4C) of Freuder in University College Cork, Ireland.

• Random generator in Java by Zheying Yang (zyang@cse.unl.edu)
• Links to Constratint Solvers, via the CSPLIB
• The ECLiPSe Constraint Logic Programming System
• Constraints archives at CCC.
• CSPLib: Problem Library on the Web.
• Random generator in C at LIRMM
• Random generator in Common Lisp by Patrick Prosser
• Random generator in C courtesy of Fahiem Bacchus (and P. van Run).
• JACK: a library providing constraint programming and search for Java (high-level language, generic search engine, and a visualization tool).

• Constraint Programming (CP), AAAI, IJCAI, ECAI, FLAIRS, etc.

• Lisp resources
• Eric Moss: personal help page with Lisp
• Eric Moss:  There is a new Common Lisp implementation on the market (beta copy at this point).  It's supposedly good for multiprocessor systems that need multi-threaded apps using real OS threads. <http://www.scieneer.com/>
• Latex resources
• Joel Gompert: Getting memory-efficient, pretty-looking, robust graphics with very fancy robust labels into LaTeX
• Ryan Lim: The not so short introduction to LaTex2x