Friday, April 29, 2005

The Quality Thesis

Too often Ph.D. theses in computer science consist of not much more than a couple of "papers stapled together." A shame as one can use the thesis to truly bring out the importance of one's research.

There is no serious upper page limit on a thesis and you can truly spend the extra time to make your thesis stand out.

  1. Put the results of your earlier papers together in a common framework and add some new results you never bothered writing up. (Harry Buhrman's 1993 thesis has a large collection of results on exponential-time computations that I still often consult.)
  2. Take the time to expand the proof of complicated results to the right amount of intuition and depth. (For many years Madhu Sudan's 1992 thesis had the best write-up of the proof of the PCP theorem.)
  3. The initial chapters of your thesis can serve as an introduction to a relatively new research area, (Michael Kearns's 1989 thesis gave an early broad overview of computational learning theory.)
  4. Or give your own impressions of a more established field (Scott Aaronson's thesis expounds on his views of quantum computing.)
If you are looking for a job you'll be too stressed to do research anyway so why not take the time to write a quality thesis which will get your thesis widely cited and possibly even widely read.


  1. I would advice students against spending too much time in their thesis. Most of them are inconsequential and would rarely be refered to. The examples you list are easily among the top 100 CS thesis of all time, so no surprise you refer to them. But most thesis contain a few middle of the road papers which others are unlikely to care about in great deal.

    We agree in that the thesis is the right place to expand complicated proofs.

    Also theses are a good place to discuss failed avenues of attack, such as "notice that a straightforward simulation technique would not work as the result can be relativized in either direction" or "note that a simple probabilistic analysis doesn't work as the Chernoff bound for this case is polynomial, while for this to work we need an exponential bound".

  2. I think both Lance and the commenter are right, in that there are really two modes in doing a thesis. The "low-work, low-payoff" mode is to as much as possible staple things together and do little else. This approach will work best for most people. The "high-work, high-payoff" mode is to put time into your thesis and make it something people read and cite.

    In many other fields (like history), where publications often correspond to books, the tradition is that your thesis becomes your first book. We don't have that tradition in CS, but it would be a nice model to adopt: people who have done groundbreaking work should go the extra mile and publish their thesis as a book.

  3. What is the current best write-up of the PCP theorem? (Not counting Irit's recent paper...)

  4. Offtopic, but I found it sad that the google ads on this page are offering "Free College Term Papers". Am tempted to say more, but that would be rude.

  5. Winners of the ACM Doctoral Dissertation Award do publish their thesis as a book. I don't know how much work for the resulting book is done over and above what would normally be done for a thesis, but the dissertations in this series that I've read have all been excellent.

    -David Molnar

  6. Winners of the ACM dissertation award get over a year to prepare the final book version.

    If you're research is well-accepted by the community (which you will know by how well your papers are read/appreciated), and you consider yourself a strong candidate for the ACM award, then might as well follow Lance's advice and do a good job.

    In most other cases, I feel it is better not to waste time writing the thesis. Better to spend the time doing research, and submit a staple job of your papers as your "thesis". After all, only the quality and quantity of your papers matter for your tenure case.

  7. Beyond being eligible for the ACM dissertation award there is another value in writing things up well for one's dissertation: The odds are that the write-ups you did for conferences were
    (a) done in a rush,
    (b) don't have all the needed details, and
    (c) don't cover all the work you've done.

    Writing them up well enough for your dissertation probably is
    (1) something you'd have to do for journal versions of your conference
    papers anyway (and that DOES count).
    (2) something that will lead you to think about related research problems and get better results on the problems you already are tackling. (I got an extra paper out of mine.)
    (3) a stepping stone for writing your first grant proposal since it forces you to think about the meta-issues surrounding your research.

    This shouldn't drag out a long time (a few months at the outside) but it can be well worth it if the timing works out.

  8. I believe that attention should be given to the writing of the thesis regardless of "work/payoff" considerations. It is just that I view the writing of a quasi-book as part of what consists a PhD, and so it has to be done to the best of the student's abilities just like the other duties (passing courses, etc).

    I would also like to add fifth advice: Regroup your material by method, instead of the grouping by result that happens when the research papers are stapled together.

  9. >> If you are looking for a job you will be
    >> too stressed to do research anyway...

    Don't think this is a valid argument. Stress
    can be a very fruitful condition for research. Who knows how many papers were engendered by deadlines?

  10. I would advise anyone writing up their thesis to make sure they don't over do it. I collapsed after submitting mine and it really is not worth it. I have been very ill for the last five months. I am only slowly recovering. My medical Doctors are convinced I put too much effort into writing up. I normally worked 14 hours 7 days a week for two years. Anyway as a nurse said to me while I was in hospital: "It does not matter if Doctor or Mister is on you tombstone if you are under it". The morale is a good work life balance is essential

  11. Many institutions limit access to their online information. Making this information available will be an asset to all.