The Seminar: Content and Form You are responsible for more than you may think! True or false: Guide has given you the topic Guide only gives you a rough direction. You need to formulate the precise topic. Report should contain a summary of every paper you have read. Completely false. Take only what YOU consider relevant. You have the privilege of determining this; but like most privileges this is also a responsibility. ------------------------------------------------------------ seminar = story Elements of a good story: conflict: Hero wants to marry the heroine, but must battle the villain. conflict resolution: Hero cleverly/with strength/... overcomes the villain. conflict must be universal, deep. shy hero vs. flashy villain. poor hero vs. rich villain. Man vs. nature. Man vs. fate. ------------------------------------------------------------- Elements of a good CS seminar: interesting problem. "conflict". We would like to have XXX but cannot have it because of YYY (villain). conflict resolution. Through a series of (clever) ideas "hero" the problem gets solved. universality/depth of the problem: Is the problem really difficult from a CS viewpoint? Are there analogues of the problem in many subjects? e.g. shortest path problems may arise in many disguises. The more universal the problem is, the more interesting the story. universality/depth of the solution. Can the solution be applied to many problems? "Cacheing" is a powerful general idea. ---------------------------------------------------------------- Good Story != list of dates and battles. Some battles are more important. It is important to find common themes between the battles or the different papers that you read. Good story : reader identifies with the hero. Good CS seminar != summary of different papers on the topic. Must say what is important and what is not. May say "I didnt study this paper because it did not have enough of xxx;" or "Even without this paper there were enough interesting ideas". Good seminar: the basic problem and its nuances must be formulated in enough detail so that the reader is fascinated by it. ---------------------------------------------------------------- What makes for a good CS seminar topic: Algorithm to solve problem X Design of software/hardware for problem X (issues: how to balance different conflicting constraints) Impossibility Results Modelling of real life problems e.g. how to model train movement and scheduling, how to model "electronic teaching" Unification of previously known problems or previously known solutions. Surveys are interesting. But these must be organized by ideas and not by individuals or by chronology. Above are only examples. There could be other topics as well. Just as it is difficult to say what makes a movie plot good, it is really hard to say what makes a seminar topic good. ---------------------------------------------------------------- Telling a story properly 1. Everything is well with the world. 2. Conflict appears. 3. Simple attempts to solve the conflict. Dont work. 4. Battle royal. Hero wins. 5. Hero reflects on the situation. Draws lessons for the future. ------- 1. Nice pleasant introduction. We all know that "Internet has been hailed the next best thing since sliced bread". 2. But internet causes xxx (intrusion into homes, people rely on something which is not tested, subject to attacks...) 3. Why simple fixes dont work. 4. The important ideas in solving the problem. How they solve the problem. 5. Is the problem completely solved? What more is needed? In theory? In practice? Items 1-3 should comprise the introductory chapter. This is very important. The crisis/conflict must be properly established. The problem must be explained in detail. Maybe give examples. Explain why simple ideas dont work, unless this is obvious. Item 4 is the core of the seminar. This could be several chapters. Item 5 comprises the last chapter. It should reflect on what has been done, what more is needed. --------- Form vs. content Form and content are intertwined. Sorted list may contain the same items as unsorted, but has different properties. Form must match content important ideas --- chapter (use appropriate title) less important --- section (use appropriate title) ... --------- Add value: simpler proofs. worked out examples. alternative interpretations. e.g. geometric interpretation of definitions. e.g. philosophical implication. original work is not expected. However, write simple programs to construct/check out examples. -------- Consistency and Integrity Even if you read 5 papers, each with a different notation, have only a single notation in your report. If the same name is used but the precise definitions are different, then change the name consistently. e.g. Paper 1 defines Laplacian = certain matrix L. Paper 2 defines Laplacian = L normalized somehow. Then you define the term "normalized Laplacian" to distinguish the new definition from the old. Same thing about argument style, description style, examples, figures. -------- A good story is hard to define precisely. A good seminar also. Both are works of art! -------