Ten Commandments for Doctoral Students

Based on my recent experiences as supervisor, I feel the time has come to formulate my Ten Commandments for Doctoral Students in Computer Science and related fields. With just a tiny bit of exaggeration.

1. Be the fiercest opponent of your own claims. Never disregard findings that shed any negative light on your results. On the contrary, invest most of your effort not into your idea itself, but into exhaustive research and clarification of anything that may affect its value negatively. Do not omit any such finding in your publication. Do not avoid any unpleasant questions. Instead, invest maximal effort into attacking your own idea from all imaginable points of view. Only then will you be able to present the idea in a solid defensible way.
2. Murphy’s Law of prior art existence. With every idea or solution of yours do assume that it already has been published in some form somewhere, even if you were not able to find it yet. Almost certainly you just did not dig deep enough. You need to search twice as hard than you thought necessary, even if you had taken this advice into account to begin with. (For a sliver of hope see remark below.)
3. Cite, cite, cite, ad nauseam. Cite exhaustively all prior art that may be interpreted as (even remotely) overlapping with, or contributing to, or enabling in any way, or even resembling your original result. None of your un-original claims shall remain uncited. None of your original claims shall remain unambiguously unexplained or unproven. When writing, act as if the sole goal of the thesis review committee (and then the wider scientific community) was to identify the single citation you had missed and then to kill your defence.
4. Unambiguous, verifiable statements only, please. Throughout the text use only unambiguous and verifiable statements. Quantifiable information is infinitely more valuable than vague literary description. All imprecise information is suspect of being unprovable. Claims of the type “in most of real problems our method works well” are misleading at best. Have most of the real problems really been explicitly evaluated? How is the set of all real problems defined? How big is it? What is the meaning of "well"?
5. There is always a reviewer more knowledgeable than you. Never assume that you know more than the reviewers you are about to face. Never make claims showing off such an assumption. Even if the rare thing happened and you indeed knew better than the reviewers, you still need to refrain from exaggerated claims. If for nothing else, alienating fellow scientists does not help anything. And there is nothing more embarrassing than a claim “my method is the best” based on small-scale experimental comparison against a randomly or lazily selected prior art that just happened to be readily available within your department.?
6. Your work must have a clear story. Build a storyline that connects all your results into a cohesive and internally consistent study, where all partial results contribute together to address a bigger goal. Give the thesis a clear and meaningful structure. Once you clarify for yourself all the key messages you intend to convey, the storyline and structure become easier to construct.?
7. Do not drown in unnecessary detail. Do not drown the key message in the sea of detail. A good high level structure and balanced focus should help. Avoid the mistake of mentioning key points just briefly and then flooding the text with extensive side-detail of no particular importance. Avoid detailed descriptions of trivial and well known concepts. The main thesis text should first of all contain all that is necessary to convey the key original results and findings. Supporting technical detail that is not necessary for understanding the key concepts should better be described concisely, or moved to appendices.
8. Cover all that is necessary to understand your key results. Do not forget to cover each important aspect that is necessary to understand the main message and key principles covered in the thesis. A reader from a related scientific field should not be forced to dig in external sources to understand the thesis.?
9. Be a pedant about notation and consistency. Maintain maximal discipline in notation and consistency. The same symbol must never appear at various places with different meanings without well justified reason. Whenever a symbol has an expected meaning within the respective scientific field, it should be used with the same meaning in the thesis. Do not introduce non-standard notation unless absolutely necessary. Do not repeat the same claims and formulations at various places in text. Do not use forward references; symbols and technical terms must be introduced and explained before their first use in text, figures and formulas shall not appear after the page of their first reference. Readers’ attention must not be strained.
10. Your thesis is not a novel. Although your thesis needs structure and a well formulated story, avoid excessive literary formulations and low density discussions. Never prolong text for the sake of length. Consider the time of your readers as infinitely more valuable than the time you spend writing the thesis.

Remark on Murphy’s law of prior art existence - a bit of hope

• Does the law say that there is no chance to come up with a sufficiently new and original idea? No. Although it is virtually always possible to identify close prior art, the closeness may vary, and quite often it is possible to identify relevant differences and advantages of your idea against the prior art. Needless to say you must be absolutely fair in evaluating and describing the delta. If it is not possible or the delta is too small, then the idea is likely not worth following. But even if groundbreaking ideas are very rare, there is always enough space for meaningful advances, be it through modest steps in some fields, or through applying an unorthodox perspective to re-formulate known ideas in unknown context, or even through connecting known isolated bits into a meaningful system that gives previously unknown higher-level picture.
• Do not despair even after a very long series of disappointments, when all your ideas turned out to be flawed, unoriginal, or inferior to what others had published before. We stand on the shoulders of giants, and we have to accept that our plight is hard - many brilliant people before us had already scaled almost all meaningful research paths, as it feels. But there is hidden value in the long seemingly fruitless study. The best scientists say that the more they have learned, the wider horizon of the unknown they see. The more you study prior art, the more failures you experienced and learned from, the better you will recognize the horizons and your true opportunity.
• When comparing your work to prior art, or building on top of it, be cautious about what the prior art says. Do not automatically trust all that is published, even in respectable channels. Published claims quite often suffer from various biases and contextual limits of validity, not always well presented. Claims are often too bold, excited authors often generalise too far based on too limited evidence. It is not rare to find out that in reality the seemingly superior prior art would perform worse than your idea. The only way for you to show real value of your idea and to provide reliable comparison, however, is to recompute all experiments from relevant prior art in the same controlled environment in which you tested your idea.
• The search for prior art is a craft that needs to be learned by practice. Although the better known channels give better chances of finding what you need, there is no guarantee that a brilliant result does not lie hidden somewhere in lesser known conference proceedings. Unfortunately, it takes quite an effort to minimise the risk of missing something important.

Remark on Murphy’s paradox of tough reviewers. The chance of meeting someone considerably more knowledgeable than you are should be treated as being many times higher than you think, and gets dangerously high especially in the community of potential reviewers of your thesis. This is the case even if you are an extremely hard working genius. If you are not, this paradox needs particular attention. There is a paradoxical corollary to this paradox: the chance does not decrease even if your field is narrowing down and its expert community size converges to one.