C266 - Harvard University

SCHOOL OF ENGINEERING AND APPLIED SCIENCES
HARVARD UNIVERSITY

CS 266: Biologically-inspired Distributed and Multi-agent Systems

Prof. Radhika Nagpal
Fall 2007

Course Description Assignments and Guidelines Schedule and Reading List Paper Reviews Project Milestones and Suggestions Fall06 Staff

Assignments

This course will involve reading research papers, leading and participating in class discussions, and a final project.

Paper reading and in-class discussion: For each class, there will be 1-2 research papers assigned which will be discussed in class. These readings and in-class activities are the key learning tool in this class and so it is very important to keep up with the reading so that you can participate fully in the class. [read guidelines below]

Paper reviews: Before each class, you will write a short review for the starred paper assigned for that class. The review should consist of: 1 summary paragraph, 1 major strength, 1 major weakness, and 1 question (more guidelines). Reviews are due MIDNIGHT the day before class; post your review using the link at the top of the page and read other people's reviews. [read guidelines below]

Class presentation: You will be expected to present one of the paper topics and lead the class discussion in at least one lecture. As part of this presentation you will be expected to do extra reading on the topic beyond the assigned papers. Before your presentation you should meet with the TF during office hours to discuss your presentation. [IMP!! read guidelines]

Final project: Finally, everyone will undertake a final project in groups of 2. Proposals will be due at the end of the first month. There will be several suggested projects, however you are also encouraged to pick a general area that interests you and formulate a project within that area. You can also take a look at projects from previous years (e.g. fall 2004). During the course there will be several milestones to help you along the way. At the end of the term, everyone will present projects in class and prepare a written project report. This report will be due by the last day of classes (i.e. projects may not extend into the reading period). More detail on all aspects of the final project

More details on each of these aspects will be presented at the first class. The final grade will take into account each of these aspects.

Guidelines for Paper Reviews, In-class Discussions and Lecture Presentations

This course will involve reading research papers, leading and participating in class discussions. This is a key learning (and grading) tool in this course. Here are some guidelines on what we expect and how to make it productive.
Paper Reading, Reviews, and Participation
For each class, there will be 2 assigned research papers. Often one might be more important to read in depth than the other -- I will mark that paper with an asterix. That is also the paper that you need to review. In class we will not review the paper, so if you have not read it then it will put you at a significant disadvantage. Also under that topic there are several additional papers that you may read if you want to follow up on more threads.
In reading the paper there are several aspects to pay attention to --- what was the key idea, what would it require to be implemented in the real-world, what other problems could it be applied to. Also how well was it evaluated, what would be an optimal solution to compare it to, and can it be mathematically analyzed. We will also read several papers on models in biology with the goal of identifying abstract principles that lead to robustness and adaptability. A key thing to keep in mind while reading these papers is that many of them are "firsts" on a completely new idea and therefore less fully evaluated than one might expect for a more developed topic.
For the paper review, you should submit a review of the asterixed paper. Reviews are due MIDNIGHT the day before class and are posted on a discussio forum that all of us can read. Your review should consist of the following:

Short para on any of the following points: What do you feel the main contribution of this paper is? What did you find interesting aout this work? What's the essential principle that the paper exploits?
One major strength of the paper (one sentence)
One weakness of this paper (one sentence)
One question or future work direction you think should be followed OR a reply to a previous question posted on the discussion forum.

The point of the reviews is not to regurgitate the paper but to identify what *you think* is the key concept to learn from the paper and what your opinion is of the strength/weakness of the idea and or paper. It is also to identify some of the issues to think about before coming into class.
In-class participation is key. The lead presenter will give a short presentation on other literature - e.g. work that was done afterwards, or additional areas. They will also have a list of specific discussion topics and problems for participation. It is essential that you speak up and contribute meaningfully to the discussion both by raising questions and proposing answers. We will do everything possible to encourage a comfortable discussion environment.
Lecture Presentations
The goal of the lecture presenter is to educate the rest of the class through a meaningful discussion of the paper topic (as opposed to just the paper). This means that the lecture presenter has to do a more indepth study of the paper topic and previous/future work as well as come up with some interesting problems to discuss in class. Here are some guidelines on how to go about this.
Preparation

(Must) Meet with the class TF the week before to discuss your presentation. He will suggest some additional literature to read and also questions on the particular topic. It helps if you do some reading of the paper itself before this meeting.
Read addition background material on what's been done before and what's been done since . For each topic I have a number of additional papers on the website that you can read. But you can also follow up on references to previous work (or read a textbook chapter on the topic, or look for a review paper on that topics) and look up what the authors (or others) have done since with this idea. You don't need to become an expert in the area but instead you should be able to bring to the class new information that will help them understand the topic better.
Make a list of structured questions that the class can discuss in depth . The point is to think critically (not the same as criticize) about the work and the approach taken. Examples might be -- how do you extend this work to other types of networks or settings, what is a centralized approach and how badly would that perform, if the system is not mathematically tractable how do you construct an evaluation strategy, can the approach be modified to work in a completely new setting, etc. You should have thought somewhat about the answers to these questions before class, and you should expect that only 2 or 3 will get covered in class.

Lecture Format

Give a short discussion of the paper (~5 mins), with your opinion on the main contributions of this paper. The goal is not to review the paper or restate the major results of the paper, most everyone should have read the paper well enough for that. Instead the goal is to provide a more high-level treatment: What (in your opinion) is the real contribution this paper? What's the main principle exploited or what's the take-away lesson that you think you could apply elsewhere, What did you really like about the paper and where did you think it fell short.
Occasionally we will read a long paper or a paper with a significant amount of technical material/depth (e.g. the AntNet paper). In those cases it is ok to violate the above. You can spend time to go through the mathematics, or experimental data, or some interesting part of the paper in more detail. The goal is for everyone to understand the technique reasonably well - if that is hard, then we should spend time on it in class.
Give a short presentation on related work (~15 minutes). This is very valuable to the class, since many objections and ideas and questions relate to what has been done before and what has been done since. Also many times there are related areas that are interesting. And there are now lots of movies online and those can add a very visual understanding for some of the topics. Finally, you can (and should) use your personal interest to guide which part of the future/past work you want to focus on -- a long survey is much harder to retain than a good story about one direction.
Pick a short list of questions you want to discuss, and a format to discuss them in . For example, you can choose one topic to do as a big group, and one question/problem to do by breaking into smaller groups. This is especially good for discussing issues that the whole class might have wondered about, or essembling a list of properties of a system. The smaller groups format works very well for fairly open-ended problems -- problem solving, alternative algorithms, extensions, type of questions . Basically you break the class into groups, and allocate some time. Then we could get back together after some fixed amount of time and each group could summarize their "findings". I will also bring some questions to the class, and so I may derail the class direction during this section. Therefore front load the topics you want to discuss most and dont get too upset that some may never be reached.
Finally, don't worry too much about the paper reviews, its your presentation. You pick a few questions for discussion (<=4 total) - I will bring a few too. The more "deep/general" the questions are the better. For example, if there are two papers, then one can spend a good amount of the discussion time constrasting them and learning alot just from that.