REMEMBER: Shoham paper and chapters for FHMV are in a manilla folder in the ATL mailroom. MAKE YOUR OWN PERSONAL COPY AND PUT THE ORIGINALS BACK PROMPTLY FOR THE NEXT PERSON!

NOTE: Readings for the week of 3/29 have been revised!!

Assignment 7 is posted in the assignments directory. It is due 3/31/99.

Office hours have been added Mondays 1-3pm.

Lectures: Monday and Wednesday, 10:30-12:30, 3437 EECS

EECS 492, or equivalent. We assume a solid grounding in search, logic, planning, uncertainty reasoning, and learning. We will take an "agent-based" approach.

This course has two main purposes. One is to provide students who want to become AI researchers and practitioners with a deeper and broader appreciation of the field. The other is to give students experience in reading and understanding cutting-edge research results as presented in recent papers rather than in textbook form. Students will demonstrate mastery of both of these goals by working through exercises, critiquing papers, discussing ideas and approaches in class, and completing a course project. By achieving both of these goals, students should be better prepared in AI breadth (for qualifying exams, for example) and in AI research methods.

Whereas it used to be that AI advanced topics could only be gleaned from current research publications, recently there have begun to emerge textbook treatments on these topics. For this class, we will use a combination of textbook treatments and research publications. The principle textbook will be the one used in the Intro AI course (EECS 492):

Supplemental reading will come from other books, including:

Fagin, Halpern, Moses, and Vardi, Reasoning About Knowledge, MIT Press, 1995.

Weiss, Multiagent Systems. A Modern Approach to Distributed Artificial Intelligence, MIT Press, in press.

Beyond that, we will read recent research papers on a variety of the topics covered in these books. A mechanism for acquiring these papers will be developed shortly after class begins.

During the meetings of this course, we will alternate (irregularly) between studying "textbook-like" treatments of topics, and then delving into research papers dealing with these topics. The textbook-like treatments are intended to give a somewhat balanced view of fundamental issues and approaches, while the papers give a more narrow and deep view of a vein of research within the topic. Thus, the former increases a student's literacy in core AI concepts and techniques, while the latter exposes the student to advanced issues as well as methodologies in AI research and development.

As a participant in this graduate level course, you are expected to not only soak up knowledge but also contribute to the knowledge gained by your classmates (and quite possibly your instructor). To this end, the course requirements include a combination of mastery of the material and going beyond it. Mastery of the material will be demonstrated through homework assignments and contributions in class. Going beyond will be demonstrated through a course project, as well as through some of the assignments and contributions in class.

Homework assignments will typically take one of two forms. One form is the standard "problem set" model, where you will be asked to solve some problems to develop facility with concepts, algorithms, and techniques we have covered. A second form is the "critique" model, where you are to write a short (no more than 1 page) critique of a paper or other piece of work. A critique represents your thoughts on the work (not a simple listing of the contents of the work) conveying:

• What you see as the significant ideas in the work, and why they are significant to you.
• What you see as the strengths and weaknesses of the work, the ideas, and methodology.
• What new opportunities you see for improving, generalizing, or applying the work, and concrete steps that you would take in pursuing these opportunities.
• How the work relates to other work you know about from this course or other courses.

You are expected to complete a course project to get a deeper "hands-on" feel for the material. Generally, what is expected is that you will find, during the course, some idea(s) that interest you, and that you will pursue that/those idea(s). This generally takes the form of duplicating some research results that we learn about in class, and extending them to address some limitations or to examine their strengths and weaknesses on other problems. Reports backed up by implemented programs, empirical testing, or mathematical analyses will be received especially favorably. The project should be substantial but not overwhelming, with a 10-15 page writeup (excluding code, traces, proofs, etc.).

As a graduate student studying advanced AI concepts, you will discover that the ideas, techniques, and methodologies at the frontier of the field are not always as straightforward to grasp and apply as the more established introductory ideas in the field. That is the excitement at the frontier! You are not supposed to simply treat whatever you read as being gospel; rather, you are to analyze and question it. A big part of moving on to advanced AI topics is to develop this inquisitiveness; to engage in discussions and debates about ideas and techniques.

Therefore, we expect and strongly encourage students to raise questions, voice concerns, and (best of all) make suggestions about the tools and techniques we will study. Do not assume that the authors, nor the instructor, have all the answers!! Contribute your answers. To encourage this, you will have a fraction of your grade based on the quantity and quality of your participatory remarks in class. It is thus crucial that you do the readings before class meets, and attend class!

Course grades will be based on performance on (approximately weekly) assignments, on a course project, on a (take-home) final exam, and on class participation. The tentative grade breakdown is:

• Assignments - 50%
• Project - 40%
• Class participation - 10%

Assignments are due at the beginning of class on the day specified by the assignment. This is especially important because some assignments are expected to establish literacy for the lecture/discussion topic; failure to submit the assignment on time will result in being underprepared for class discussion which will harm your grade on a couple of fronts.

We expect adherence to the Engineering Honor Code in all assignments and exams. All problem sets (home work assignments) are to be completed on your own. You are encouraged to discuss ideas and techniques broadly with other class members, but all written work, whether in scrap or final form, are to be generated by you working alone unless otherwise expressly stated in the homework assignment. You are not allowed to sit together and work out the details of the problems with anyone. You are not allowed to discuss the problem set with previous class members, nor anyone else who has significant knowledge of the details of the problem set. Nor should you compare your written solutions, whether in scrap paper form, or your final work product, to other students (and vice versa). You are also not allowed to posess, look at, use, or in anyway derive advantage from the existence of solutions prepared in prior years, whether these solutions were former students' work product or copies of solutions that had been made available by instructors. Violation of this policy is grounds to initiate an action that would be filed with the Dean's office and would come before the College of Engineering's Honor Council. If you find any ambiguity about this policy, it is your responsibility to contact the course staff.

Unless arrangements have been made ahead of time, late assignments will not be accepted. You are permitted to skip one assignment (other than the project!) over the course of the semester.

To be incrementally and interactively elaborated. Readings are: RN (Russell and Norvig); FHMV (Fagin et al); W (Weiss); JAIR (Journal of AI Research); AIJ (Artificial Intelligence Journal)