Hints for Writing a PhD Proposal
By Angelos Keromytis (April 2010):
- a description of the problem in enough detail to clearly state the
thesis proposition (next item)
- a proper, concise thesis proposition; this is not an abstract
statement like "we're going to investigate the insider problem", but
something along the lines of "our hypothesis is that the use of XYZ
technology in environment Z under constraints Q can identify insider
attackers with probability Z" --- obviously, the fewer qualifiers the
better, but you also need to be accurate; since this is a thesis
proposal, we will cut you some slack -- but it's in your best interest
to think hard about this, since it is the anchor point of your whole
thesis (and the next few years' worth of work for you)
- a description of the related work, how it does not solve the problem,
and how your hypothesis has not been tested before
- preliminary results (if any) that indicate that you have reason to
believe that the hypothesis holds
- additional experiments that you will run to prove that the hypothesis holds
- what you'll need to build to run said experiments (and what you've
- what happens if you can't run some of these experiments, or if they
give you "bad" results --- what's your failover?
- what are the expected contributions, keeping in mind that each major
contribution must demonstrate novelty, non-triviality, and usefulness
(so, "first", "best", "only" are good adjectives here)
- how long you expect all this to take
Hints for PhD Proposal Defenses
- PhD proposal defenses in Computer Science allow student audience;
this is a good opportunity to find out what works and doesn't from your
more senior colleagues.
- Proposal defenses consist of four parts: first, the candidate
introduces themselves, then presents a summary of their work,
interrupted and followed by questions from the committee. Finally, the
committee meets in private to discuss the presentation and the plan.
- While most of the committee will have read most of your proposal, you
cannot assume that everyone has read every page in detail.
- Avoid high-level talks: "... they usually fail to convey the
intellectual substance, creativity, ingenuity of the speakers'
accomplishments - what takes the work out of the routine. Naturally,
these comments apply to all of our speakers who want to impress people
with their ability as opposed to the breadth of their knowledge or the
size of their project." (Ed Coffman)
- When presenting experimental work, be prepared to defend your
methodology. What was your sample size? Confidence intervals?
- Standard presentation guidelines apply:
- Talk to your audience, not to your slides.
- Project; speaking softly conveys the impression that you are unsure
of what you are saying.
- Make sure that all your graphs are readable. Check this in the
actual presentation environment (using a video projector), not just on
your laptop screen. A common problem is that the lines are too thin.
- Avoid flashy or cheesy animations, such as animated GIFs, or
PowerPoint word art. This is not a sales talk and these gimmicks distract
from the message and make you look unprofessional.
- Keep to the allotted time of no more than 45 minutes.
- Your presentation needs to address the following:
- What is the problem you are studying?
- Why is it important?
- What results have you achieved so far and why to they matter?
- How is this substantially different from prior work?
- What do you need to do to complete your work?
- Your workplan should be sufficiently detailed so that the committee
can judge whether it is realistic or not. You don't have to account for
every day between the proposal and your thesis defense, but a roughly
monthly or quarterly granularity is to be expected, depending on how far
away your anticipated graduation date is. Specify the experiments you
need to run, the software you need to write and the algorithms you want
to try out. This should not just be one page that says "I will do
- The committee should be handed a copy of your slides.
- No more than 25 slides, plus "back up" slides with additional
material in case of questions. The committee will get anxious once the
presentation lasts longer than 35-40 minutes.
- List your contributions early and explicitly. You don't want to
create the impression that related work is yours, and vice versa.
- One of the most important concerns during the proposal is to
convince the audience that you are aware of all related work. Since
some of your work may date back a few years, it is not sufficient to
just copy the reference list from your first paper. Check common recent
conferences to see whether any recent work applies to your thesis. If
applicable, point out your work predates work presented by somebody else
done more recently. (Given the duration of most theses, it is not
uncommon that others pursue a direction after you have stopped working
- When presenting your contributions, be sure to use "I" and not "we"
so that the committee will know what aspects of the work where yours,
and which were group projects.
- You must convey a clear plan how you are going to evaluate your work
systematically - by measurement, simulation, user experiments. This is a
core part what makes computer science science and not just
- Be prepared to back up any comparative statement with facts, in
particular statements like "works better", "faster", "scalable" or
"optimal". If you are presenting a protocol, how do you know that it
works correctly? If your algorithm is optimal, can you prove that it is?
(If not, avoid the term.)
By Yechiam Yemini (March 2010):
A thesis contribution is a technical result that is both substantially
novel and creates significant new knowledge.
A technical result is a solution of a technical problem. There are four
types of technical results:
- A theory consisting of a body of theorems and their proofs from
- An algorithm that computes certain output from a given input.
- A performance analysis describing quantifiable behaviors of a large
class of mechanisms, or characterizing optimal selection of their
- A design for a hardware, software or protocol mechanism capable of
resolving a broad class of problems.
A result is substantially novel if it cannot be derived as a simple
application or extension of known results. A result creates significant
new knowledge if it is (a) not obvious; and (b) if it is sufficiently
abstract to be applicable to a large class of problems.
[Hints for PhD defenses] [Writing style] [Writing bugs]
by Henning Schulzrinne