[ Home | Lab | New Students | Courses | Research | Publications | Activities ]

Things may come to those who wait, but only the things left by those who hustle.
—Abraham Lincoln (often attributed to him)

This page is quite incomplete, especially the parts at the top. Please bring up your comments and concerns when we meet and I will try to improve it gradually.

This page is meant for my students. It enumerates several points that I noticed I repeat for many of my students as I attempt to guide them down a path of research. These are mostly simple things, obvious in retrospect, but worth paying attention to.

PhD students are the primary targets of this advice, although much of it applies to MS students as well. I am of the view that research has not been conducted until it has been written up, hence the emphasis on papers in the first part of this document.

From the point of view of getting to graduation quickly, we are interested primarily in research papers. There are no good research papers that don't introduce useful concepts, make a technical contribution, and evaluate it. But there are several variations as to what contributions and evaluations are acceptable.

1. Theoretical, where the main theme is in refining the formal concepts that underlie the domain of interest in which the target problem occurs. The contribution is the nature of formal results. The evaluation is in showing that the results obtained correspond to important properties of the domain.
2. Systems, where the main theme is in showing how a certain family of systems can be built. The technical contribution lies in the principles underlying the design and analysis of the targeted systems. The fact that you built a system is rarely adequate. It is the science underneath the hard-work that matters most.
3. Experimental, where the main theme is in studying a class of phenomena through experimentation. The experiments should be controlled. Frequently they will be based on simulations, although they can be based on running traces of the executions of physical systems where appropriate.

Every paper tells a story, or should. Think of what story your paper will tell. The top-level story should sound familiar to most people. A good story will include subplots. It is in these subplots where more and more of your creativity and contributions take shape.

Here is a simplified outline for a research paper. This is the way to tell a scientific story. In this sense, even an MS thesis or PhD dissertation should exhibit a similar structure. Every paper will be different and almost every paper will deviate from this outline, but the points below can help you get organized.

1. Introduction
   • Motivation: the problem, why it is a problem, and why it is an important problem.
   • If possible, a short historical perspective. Short is the keyword here.
   • Introduction to the approach and its promise and adequacy (informally). This is where your story begins to take root.
   • Technical challenges we must overcome to make the approach a reality.
   • Contribution: research questions addressed.
   • How the contribution will be (or has been) evaluated: could involve establishing key theoretical results or simulations or measurements from practical deployments (hard).
   • Results: a crisp statement of your takeaways.
2. Approach
   • Conceptual and technical framework.
   • Key concepts and terminology.
   • Formal definitions, if any.
3. Implementation. The purpose of computing research is not so much to build a system that works or works well as to understand how and why it works (well) or fails to work (well). If your implementation has any nontrivial features and especially if the implementation bears some effect on your results or your evaluation, you include sufficient details that a person "skilled in the art" (using the term as the US Patent and Trademark Office uses it) would be able to reproduce your implementation, and thus to verify your results that depend upon the implementation. So be clear about how exactly it was carried out and how it may be replicated.
4. Results.
5. Evaluation. The purpose of the evaluation is to show how we may place credence in your claimed results? You don't need to build a complete working system for it to be useful for your research.
6. In some cases, it helps to evaluate the performance of your system experimentally, especially with regard to its prospective scalability. Here's a simple pattern for doing so:
   • Identify what you mean by scale: number of users, number of artifacts, number of states, number of strategies, number of attributes, or whatever else.
   • Identify a reasonable measure of effort: wall-clock time taken for a decision, number of messages generated, or whatever else.
   • Relate the target measure of performance with the measure of scale. To do so, you would fix some of the variables and evaluate performance along some of the dimensions.
7. Relevant literature: should be classified into the main areas of relevance. Don't just say what other people do; relate it to your contribution.
8. Discussion
   • What your study and its evaluation didn't address
   • Directions of special interest, list a few but don't list too many

1. Know what you are trying to convey.
2. Know your audience.
3. Convey your story with as few distractions as possible to your intended audience.
4. Choose your battles.
   • Don't open up too many kinds of claims in the same paper, especially too many claims that your audience will not readily understand.
   • Don't make unnecessary remarks in passing. Readers often find such remarks manipulative and resist them. You want the reader coming along with you for the main point of the paper.
5. Don't include your solutions in your problem; don't include your answers in your questions. It makes your work sound contrived. At best, this suggests that you haven't thought your problems or questions through. At worst, it suggests that you are manipulating your reader into accepting your problem and question statements because you already know the solutions you are going to find.

1. Participate in regular group meetings. Join our discussion group and see the postings therein. [Aside: would someone please volunteer to set up and maintain a discussion group for our group?]
2. Talk to other students regularly, especially your seniors working on a topic in an area that appeals to you.
3. Start thinking early about your dissertation work. It takes a long time to become sufficiently familiar with everything you will need to know for a dissertation topic.
4. Begin reading the huge literature and standards pertaining to our broad area of interest (e.g., Web services). There are good links from recent work in the lab (papers and presentations). This will easily take several weeks.
5. But don't just read the literature. Try to relate your own ideas to it. Research is an interactive process—between you and the rest of the research community. Ultimately, your goal is to persuade your community (part of the challenge is selecting the right community) that you identified an interesting problem and solved it in a novel manner.
6. Think about additional courses that you will need to complete your skill-set—a relatively easy task but best done well in advance because all the right courses aren't available when you need them.

1. Your initial interest may be along the lines of an application area. However, application areas don't directly and easily lead to the kind of academic research you need to perform in order to obtain a PhD. In particular, if your emphasis is exclusively on an application, you might end up with a substantial implementation effort and with the resulting challenges of evaluating your implemented artifact. Yet, it is clear that you must have some applications in mind when you perform your research. Further, some applications can yield special properties that make a given approach feasible and desirable. For this reason, while you might think in terms of your application area and develop some intimacy with the chosen application, you should attempt to abstract out the technology area underlying the application. Your dissertation topic will be cast in terms of the technology area, but motivated in terms of the application.
2. Identify potential interesting dissertation problems. I generally ask my students to write 1–2 page descriptions of each problem that they are interested in. I can then comment on these descriptions and help identify something that will agree with your tastes.
3. Make some simple throwaway prototypes just to get familiar with interesting ideas.
4. The topic you choose won't be the last thing you work on. There will be occasions to change your mind and do other things, so don't waste time searching for the perfect topic. You will end up changing and refining it anyway.
5. Don't imagine that you are going to come up with The Unified Theory of Computing^TM right at the first go. Most such (purported) theories are bogus anyway. So keep it simple enough that you can do it and explain it, others can follow it when you describe it to them, and—most importantly—that others can build on it in some way. The only way to achieve scientific impact is to find and persuade people who will develop on your work.

1. Make a schedule. An aggressive schedule is preferred, but we should be willing to back down from it in the end if the work isn't (sufficiently) completed as quickly as originally planned. Almost all schedules run over, so any realistically optimistic schedule is a good starting point.
2. In general, for the research questions, it will almost be simpler to do the work (identify and solve the questions) than to plan the necessary steps abstractly :-).
3. Start a dissertation draft right away. This will be a working document that you and I will consult many times—ideally each time we meet. It will give us both a sense of what is needed to complete a dissertation.
4. Stay in my attention range. Come to regular meetings; see me when you come up with something even partially interesting; send email updates frequently. I will forget about the work of students who fail to keep in touch, which will only delay their graduation. Not that I am irresponsible, but I always have enough things jostling for my limited cognitive abilities [:-)] that it is easy to put off something which isn't. Be the squeaky wheel.
5. Keep your promises. I know that there will be exceptions, but these should occur not more than once in a while. If you are running behind, send an email saying you are. You would be surprised how many students don't contact me or send me drafts of papers or theses or presentations that they promised, and fail even to send an email acknowledging that they are late. They are the ones who take the longest to graduate, and it is not because they produce the strongest results.
6. You get most credit (especially from a PhD committee) for answering research questions and very little credit for any implementation you might construct.
7. You can produce research from carrying out implementational work. In fact, some important questions won't emerge until you get down to implementing the architecture and approach. However, we won't know exactly what those questions are until later. The credit you get will be for the questions uncovered and solved, not for the implementation per se.
8. You get a lot of credit for prestigious publications. PhD committees will require additional publications for each of your exams—written prelim (qualifier), oral prelim (proposal), and dissertation defense. An MS thesis or PhD dissertation should include externally published or submitted work and will be the better for it having been peer-reviewed. For PhD students heading for an academic career, the de facto requirements for publications are somewhat higher than for industrially-oriented students.
9. Keep aware of prospective deadlines for publishing research. The best ones are for the major conferences and special issues of journals. Of course, non-special issue publications are also welcome, but because there is no hard deadline for such submissions, they always tend to get put off.

1. Collaboration between students of equal vintage is helpful to the collaborators as they can learn about more topics than through their individual efforts.
2. If you are a senior PhD student, you can benefit by having someone else do some of your implementation. You get the same amount of credit from the committee as if you did the implementation.
3. If you are a junior PhD student, you can benefit by having someone else lead you along and answer your basic questions about the area and about studies in general. You can find the senior student's unfinished work as a source for hypotheses and research ideas.
4. For collaboration, especially where one of the students is senior or farther along than the other, it is best to keep the initial efforts independent or well-modularized so that you can work separately. Since each of you will have a different agenda, there is a risk of either party inadvertently slowing the other down.
5. Ultimately, students must make sufficient individual contributions to their dissertations. However, for large problems, it is OK to collaborate with other students on some publications.

Some general points about reviewing.

1. Write the kind of review you would like to receive. Remember there is a person on the other end too, so be as nice as possible.
2. Say what you think the paper attempts. Say what is good with it; then say what its limitations are. If you love the paper you will still notice some shortcomings; if you hate the paper you will still notice some desirable features.
3. Spell check and proof read so you don't convey the impression of being careless.

This section is motivated by a conversation I had with a former student. The student had done quite well but not as well as he himself thought he was capable of. (I agree with his assessment.) He brought up this topic and mentioned that I hadn't bugged him enough to work and that he will try to "work harder" to improve his record [doesn't help me directly now :-)]. I thought I had to remind him about some work on several occasions, but he didn't listen well. Maybe I let my managerial duties slip and when I did so he lost focus. Consequently, there were long periods on low or no productivity, especially when I had some other activity drawing my attention.

I have been mulling this over lately. It is clear to me that I didn't help by being extra flexible and not enforcing regular deliverables. So in the spirit of better management and in helping improve everyone's scientific productivity, I propose the following.

Importance and Urgency

We all wish to do important tasks, but we are often drawn into spending our time on urgent activities. All too often, the important and the urgent do not line up. Quite often, too, the urgent activities are mandated by others. You must catch this game or this movie or you simply need to turn in your course assignment tonight. For some, especially the latter, it is sensible to put your research on hold a little. However, if you find yourself putting the research on hold quite a bit, you know there is a problem.

Since as humans we are inevitably drawn to the urgent, the opportunity that will surely never come again :-), the only way to keep on the important tasks is to make them urgent. As students, somehow you must acquire a sense of urgency about your research. In fact, I can attest that you will need this sense of urgency throughout your career if you want to perform even slightly above average.

Public commitments are a good way to go about doing so. There are of course the major deadlines, but commitments you make to me and to your other colleagues would also yield the desired effect. But only if you try extremely not to have to cancel them. If you end up canceling such commitments a lot, then you know you are not assigning enough resources to the corresponding tasks. So you must make such commitments more urgent.

Expectations of Effort

What matters ultimately is just the results, best measured for doctoral research purposes in terms of prestigious publications. This is what I emphasize instead of rules and regulations. However, if we are careless about the rules, the main goal is hurt as well.

1. For a full-time student, 40 hours of studying per week is not a lot—most times in life you will need to work a lot more unless you vegetate in your profession. As a student, you have far fewer responsibilities than you will have later in your career. It is in your own interest to graduate quickly and get ahead professionally while you are still a student.
2. For those on an RAship, the RAship is only half-time, but you are a full-time student. The RAship is meant merely to support you. Sponsors who really want just 20 hours of work from a half-time RAship usually offer work that isn't academic research.
3. For those on a fellowship, there are no formal job duties, but you are still a full-time student. I would make no distinction between an RAship and a fellowship for these purposes.
4. For those on a TAship or on other kind of support, you do indeed have other responsibilities. Please keep me posted if there will be conflicts with your teaching work. When my students need a TAship, I try to get them to be my TAs, so I can help manage the conflicts between their TA work and research.
5. In research, you won't always (usually) get results at a uniform rate, but you should contribute your effort at as uniform a rate as possible. Determine to have something to say about what you did on a given day, week, or month. Mostly you should just think these over and correct your direction, but we can discuss this too.
6. If you are not making progress on a particular line, let me know early. Experienced researchers manage their risk by not going down a deep unproductive path—as a computer scientist, you know the risk of depth-first search.
7. Beginning students will usually need to juggle just a few constraints—maybe courses and your research paper. As your work stabilizes, there will be fewer courses, but the number of urgent constraints will increase. For example, you will often need to prepare multiple submissions concurrently. In such cases, it will help immensely if the early work on a paper is well-organized and you take on a challenge that you are more likely to succeed with. Most submission deadlines are announced well in advance. So why not be thinking of what papers you imagine yourself submitting? Continually tracking your own progress is crucial to success.
8. Please don't cancel your meetings with me or scheduled presentations except if it is a true emergency. We should have a regular time—if we don't yet, make sure we do. Even a brief meeting can yield some advantages. First, I might give you some quick suggestions. Second, you might find that since I "consumed" your results, you are under pressure to think up some additional results in time for the next meeting. This is the Le Chatelier Principle at work. In consideration of the distractions points below, don't look for this on the Web :-).
9. I don't at all mind you having downtime and going on personal trips, but a couple of considerations apply. One, discuss absences with me early enough so we know if there will be any conflicts and how to resolve them—most cases will cause no problems. Two, choose your trips wisely. I know when I go on a trip, there is a considerable overhead in preparation and recovery. There is a lot of work to catch up on, especially when you return from a long trip. In this case, perhaps you shouldn't take further fun trips till the work is back under control after the first such trip :-).
10. A strategy is to take charge of the scheduling of such ventures. Just because a bunch of students is dropping everything to go catch a concert, you don't have to. If it were of high importance to you, you would have known and would have been able to plan. Btw, this continues even after you graduate—people will try to draw you into various activities, which aren't—or shouldn't be—top priority for you.

Distractions

In Greek mythology, the sirens are creatures (part women, part birds) who distract sailors. They sing enchanting songs causing the sailors to run their boats into the rocks. In modern times, the Web must be most powerful siren for knowledge workers!

1. There is a classical distinction between what is urgent and what is important. Some tasks are both; most are neither; and several are urgent but not important. Those of the last category have a tendency to distract us from our important goals. If you can learn to avoid the merely urgent and focus on the truly important, you can be a lot more effective. This is not easy. I can't say I have learned to do so myself :-).
2. The Web can be extremely distracting. I too often fall into a search that leads nowhere but consumes a lot of time. One thing I have taken to doing is that I set my browsers to open on a blank page. It used to be that whenever I opened a new browser window, it would show me a page (in my case, usually a news page), and I would end up reading it. Now I get distracted a lot less on this account.
3. Email too can be distracting, especially if it beeps every minute or so. Set the polling period high enough for your mail reader that you can get some work done between beeps. I disable the beeps, tray icons, and popups for my email reader.
4. When you need a break, take a break. Go take a walk or get coffee. It is usually better than trudging along at low efficiency or finding oneself distracted repeatedly.
5. When you quit work for the day (or night), leave some easy task with which to begin work the next time. Some people find this helps them get back in the flow more easily than would having to open with a daunting task.
6. If you are finding yourself unable to concentrate on a paper (and you have reason to believe the paper is important to your goals), try to write something about it. As you force yourself to find something to say, you might find that your thoughts begin to become better organized.

I don't want to be overly rigid. Because it might sound unfriendly if you asked for some waiver and I refused, it is best if you ask for only those that are somehow exceptional. The number of exceptions should be small (0–5%). You earn increasing flexibility as you establish your credentials through publishing papers.

Attending a conference is expensive. The time you spend physically traveling to a conference and the time you spend there are significant. Add to that the financial expense and the considerable effort of preparing a submission and you would realize that conference trips are not to be taken lightly. In particular, you are not there to sightsee, though you might add a day or two to your trip to sightsee after the conference.

Conferences are also great opportunities to learn about your field. There only a few major conferences and they come around once a year at best, so you should be careful not to lose the opportunity when it does show up.

1. Prepare your presentations well in advance. Practice!
2. Don't be shy. Ask questions where appropriate (make sure to phrase them politely). Talk to speakers after their presentations. Talk to people during coffee breaks.
3. Don't stick in a clump with your university gang—you can see enough of them when you are back.
4. Do look up other people that you know, especially those who graduated from our department.
5. Ask me to introduce you to any people that I know whose research interests might overlap with yours. I will make several introductions on my own as well. Feel free to join me when I am talking to someone. It is rarely anything secret.
6. You can expect the above from my former students (and, after you graduate, be equally helpful to those who come later).