Thoughts on the Structure of CS Dissertations
by
Spencer Rugaber

September 26, 1995



  1. Background

    My good fortune in reviewing four dissertations in the last four weeks has catalyzed my thoughts on what I expect to see when I read one, and I thought that I would convey them to you so you have at least one data point on what a committee member expects when reading a dissertation.

  2. Theses and Dissertations

    First, I think that it is important to differentiate between a thesis and a dissertation. The dictionary defines a thesis to be "a proposition that is maintained by argument" and a dissertation to be "a lengthy, formal treatise, especially one written by a candidate for the doctoral degree at a university." That is, a thesis is a contention or principle of which you would like to convince your reader. A dissertation is a written presentation of it.

  3. Placement of Your Thesis Statement

    Because the purpose of the dissertation is to present a thesis, it makes sense to present it as early as possible. And because you may have readers with dissimilar backgrounds, it makes sense to present the thesis at a high level of abstraction. Combining these two observations leads to the suggestion that you begin your dissertation with your thesis, and that your thesis be initially expressed as a single sentence.

    Whether or not you follow this guideline, it is still a useful exercise to construct such a sentence. This may sometimes be realized as "the elevator speech": You are sharing an elevator with an acquaintance who asks you about your work, but will only be with you for fifteen seconds. What do you say? Alternatively, you can use the abstract thesis sentence as a starter when your Mother asks you what it is that you have been working on for so long.

  4. Your Obligations

    Once you have clearly stated what your thesis is, you have two obligations to fulfill in your dissertation: You must convince your reader that the problem you are solving is a worthwhile problem to attack, and you must unequivocally demonstrate that you have solved it.

    1. Obligation 1: Motivating Your Work

      It is essential that you strongly justify all the effort that it takes to get a PhD. And unless the value of your thesis is self- evident, like your have solved the world hunger problem, then the justification will require you to place your work in relation to a set of larger problems and to the work that others are doing in the field.

      1. Your Problem in the Context of Larger Issues

        It is a useful exercise to periodically justify the time and energy that you spend in terms of your goals in life. If, for example, you are working on solving a particular variant of the P == NP problem, then, presumably, your solution will help computer scientists better understand the power and limitations of computation, which, in turn, helps us make more effective use of computers in solving the world's problems. In your own mind, if not in your dissertation, you should justify each of these steps: are computers a positive tool in solving the world's problems, is understanding the limitations of computation essential to making effective use of computers, and is your own particular work useful in understand P and NP. At least some of this justification should appear in the motivation section of your dissertation.

      2. The Related Work Section

        Of course, an easy way to motivate your work is to point out that other researchers throughout the world have recognized its importance by devoting their energies to solving it. And you can provide evidence of this in the Related Work section of you dissertation.

        This section is frequently misused and often boring. Perhaps it is best to begin with a list of what a Related Work section is not. It is not an inventory of all of the papers in a subject area that you have added to your personal bibliography. And it is not a collection of one sentence dismissals of other researchers. A Related Works section is an opportunity to present the evolution of ideas in your area and how the problem addressed in your dissertation relates to these ideas.

        A proper discussion of an "evolution of ideas" requires both an "ontology" and an "ontogeny". For the purposes of this document, an ontology is a vocabulary of interrelated concepts. They can be formally presented, but more likely are informally, but precisely delineated. Similarly, an ontogeny describes how a collection of related ideas evolves over time, often reflected in the introduction of new terminology or a transformation in problem specifications.

        The culmination of the Related Works section is a description of how your problem naturally fits into the ontogeny and how it relates to other current work in the same problem area. Which raises the question of how exactly to treat the work of other researchers. There are two possibilities: either the other researchers are working on a different problem (which should be clearly differentiate using the ontology, or they have have solved the problem a different way (which places on you the obligation of explaining the nature of the difference and why their solution is inadequate). In practice, of course, your differentiation from your competitors will be a combination of the two.

    2. Obligation 2: Validating Your Thesis

      The second obligation--to prove your case--is much more problematic and interesting, than the first. The bulk of your dissertation is an argument that step by step demonstrates that you have in fact reached a solution. My second suggestion is that you present these steps early and in a bulleted list. Doing so will have the added benefit of giving you an opportunity to describe the structure of you dissertation without the boring litany of "Chapter 1 contains ..., Chapter 2 contains...".

      There are three basic paradigms in Computer Science for presenting a convincing argument: the theoretical, the empirical, and, for lack of a better term, the aesthetic.

      1. Theoretical Theses

        A theoretical argument is a proof. That is, your problem space is a theory, your thesis a proposition, and your proof demonstrates that your thesis follows from the axioms of your theory. Such arguments are typical of the more mathematical and formal aspects of Computer Science.

      2. Empirical Theses

        Empirical arguments are also possible, particularly in areas like human factors and psychology, where it is possible to actually design and execute experiments. There are well-established rules in these areas for stating hypotheses and for demonstrating significant results. A variant of this kind of experiment is sometimes seen when a simulation is constructed and executed on some suitably random distribution of input data. Another variant is the systematic case study, sometimes seen in situations where it is impractical to isolate the independent variables being considered.

      3. Design Theses

        By far the most problematic situation arises with a "design" dissertation; that is, where a problem solution is demonstrated by actually constructing an artifact and arguing that somehow it exhibits certain difficult-to-measure properties, such as "elegance" or "simplicity". For this reason, I have called it an "aesthetic" argument, and warn you that it opens the door to all kinds of criticism.

        One way to forestall or mitigate the criticism is by way of testimonials; that is, an independent user of the artifact who is willing to state that he/she has actually used the artifact for some external purpose. Of course, supporting an external customer is additional work, and spanning the gulf between a "proof of concept" prototype and a production quality tool can be a formidable challenge. Nevertheless, research funding agencies are now looking more and more for the latter than the former. And actually putting in the extra effort while still a graduate student can give an aspiring assistant professor a head start in his/her research program development.

  5. The Future Work Section

    A section of the dissertation which is often written in a boilerplate fashion is the "Future Work" section. In particular, your reader has very little interest in what you intend to do in the future. In many cases, the author has little interest as well and is merely staking out some turf in the hope that either the reader will be persuaded not to pursue some promising topic because somehow the author has already branded it, or that his/her committee will not persevere in requiring an answer to some tough question because of a promise made in the Future Works section.

    Despite these difficulties, there is still a need for a speculative section in a dissertation; that is, some discussion of questions that the dissertation raises but will not attempt to answer. In particular, any good research insight will raise interesting questions, and it will often be the case that one or more answers or answer directions will appear promising. It is entirely appropriate to raise and discuss these issues. After all, you are probably in a better position to do so than anyone else.

    Another spin on this section is to discuss the implications of your now validated thesis. That is, feel free to predict the future in a well reasoned argument, the hypothesis of which you have just proven.

  6. Conclusion

    A dissertation has two purposes, roughly corresponding to viewing it as a product or as a process. On the product side, the dissertation is the description of a successful piece of research and, as such, should contain a properly motivated hypothesis and a well-grounded argument that validates it. To this extent, the comments made in this document can be readily specialized to any research paper. Viewed as a process, the dissertation serves as a final exam of an intensive course of research training. As such, it needs to give evidence of an understanding of the current state of a research field and the research procedures of that field (whether they be theorem proving, experiment design, or prototype implementation) can be correctly applied. It is the intent of this document to convince you not only that both views can be presented simultaneously, but that, in fact, your dissertation will be better for having done so. Oh yes, and don't forget the other important rule of CS dissertations construction: always include a quote from Lewis Carroll.


Contact Information:

Spencer Rugaber <spencer@cc.gatech.edu>

HTML by John M. Gravley <jmg@cacs.usl.edu>

Last modified: Jan 14 1997