44

A former PhD student at our lab had this idea for an algorithm for solving an engineering problem that has many well known solutions. The student developed preliminary results but ended up applying the algorithm to some other simpler engineering problems instead and graduating. I 'inherited' the project and it has always been a project my advisor has had high ambitions for. My advisor actually ended up changing my research project to this one after 6 month even though I heavily objected. The project is not funded. I TA to pay the bills.

At this point I have already advanced to candidacy and I'm in what I was planning to be my final 6 month. My research has basically consisted of building other algorithms based on the one the other student came up with. All are for an application in a measurement system. I have derived all of the mathematical solutions that will be at the heart of my algorithm and I am in the process of validating them with new experimental data. The previous data I worked with was all provided by the first student to work on the project.

I have recently discovered there are some issues with the original algorithm that create errors in the solution. The errors are larger than those of the previous (well-established) methods. This does not happen in all conditions, however, the experimental data I was left with all happened to be cases where this error does not show up (I'm not entirely sure this was by coincidence).

How can I salvage this project and graduate without delaying my graduation much. I have family obligations that require me to graduate soon and I'm devastated that I've discovered this so late. This is partly my fault for trusting the research from the previous student too much (it was easy to do since my advisor can't stop praising him).

I personally have always been of the opinion that research projects do not need to have positive outcomes and PhD students should be assured that things will be fine for them even if there research has bad outcomes. I feel that doing research to show that something has a bad outcome is still equally valuable. I also feel that because of pressure to have good outcomes a lot of researchers exaggerate their results out of fear of possible consequences and it ends up doing more bad than good when people try to use their publications.

Do you think I have a good case of being able to graduate given my situation? And how would you approach convincing my advisor?

  • 31
    If I understand correctly, you've found that an algorithm that is in the literature (at least in the previous student's thesis) sometimes fails to work. It seems to me that this information, amplified to show when the algorithm works and when it doesn't, and to explain exactly what goes wrong, would be a reasonable body of research. Whether it's enough for your thesis is, of course, for your adviser to say. It would be even better if you could repair the faulty algorithm. – Andreas Blass Sep 23 '16 at 22:50
  • 1
    Yes. It is only in his thesis (he never published any papers). I would actually like to focus on that a bit in my thesis. I guess I'm a bit worried about how my advisor would react. I haven't told him about my discovery yet and given his lack of command of the material, and the high opinion he has of the other student, I'm afraid he will think I am utilizing his method incorrectly. – somerandomdude Sep 24 '16 at 0:16
  • 1
    Negative results are motivation in research, at least I feel so. – Coder Sep 24 '16 at 13:21
  • 1
    @MikeyMike, I'm in my fourth year. The bulk of my work has this approach in it although some of my work may not be affected by the errors. The more significant parts of my research have definitely been affected though. – somerandomdude Sep 24 '16 at 20:31
  • 1
37

Einstein is attributed as saying (paraphrased) "If I knew what I was doing, it wouldn't be called research". I completely agree that a negative result is still a result, and in fact, a very useful one! The other comments address the technical merits, so I will focus on interacting with your advisor.

You have to provide a good pitch. As an academic, he will likely be swayed with good data, so spend some time verifying your result.

  • How do you know the algorithm is wrong? Find a way to quantify the error. It might be just comparing it to the well-known case you mentioned.
  • Once you have that, run the old experimental data + the new data through your code and make a graph or chart of the error. Show that the original data set results match the original work (so that your code is shown to be correct/consistent implementation of previous work), but that other data sets produce larger error.
  • Propose a modification to the algorithm, then re-run all the data. Show that the values and error are still consistent on the original data set, but with lower error on your new data set. Even if its not a real full solution, it's important to show that you have improved it.

For political reasons, you may want to avoid the term "flawed", so as not to insult the past student, but rather to say that you have extended the algorithm to other situations. That's definitely publishable material, as others said.

I think this is doable quickly since it sounds like you already have most of these pieces together, you just need to arrange them to make the argument to your advisor.

I had an old advisor that loved graphs of everything. You could sway him with a good graph. Hopefully that is true for your advisor too. Good luck.

  • Thank you. I will definitely do this! I have to admit that I usually just describe everything for him and it seems to put him off. I'll come prepared this time :) – somerandomdude Sep 24 '16 at 20:26
  • 2
    +1 for "extended". That's the kind of thing I usually see in papers, and it sounds like the supervisor would immediately react to "error" talk. – Wayne Sep 25 '16 at 15:57
42

Conditions Under-which Fancy Algorithm Has Inflated Error Properties by A. Student. Journal of Science and The Doing Of It. 2017.

This would be a perfectly acceptable paper, and one which should appear in the literature. If you can, correcting said algorithm so it does not do so would be a bonus, but not necessary. Also, it's slightly late for you, but one should always ask when embarking on a thesis/dissertation project "What happens if the answer is "No"?"

  • 1
    I definitely see the merit in this. It will be a difficult discussion with my advisor though. – somerandomdude Sep 24 '16 at 1:06
  • 8
    @Hadi By the sound of it, it looks like a difficult discussion with your advisor is on the cards no matter what. – E.P. Sep 24 '16 at 12:58
  • 3
    @Hadi If your advisor has much experience in academia, they should recognise the merit in your result, if you can frame it properly. What you've done is demonstrate that a well-known algorithm with well-established behaviour in certain scenarios actually deviates from that behaviour to cause significant errors under certain conditions. This is a useful thing to know for anyone in the field. A sizeable portion of science is about learning from other people's mistakes; most of the remainder is about making your own and letting everyone else learn from them too. – anaximander Sep 25 '16 at 10:31
4

I can't comment on PhD theses, but nobody has mentioned how to actually talk to your advisor.

  1. Do not immediately slam your advisor with, "Your favorite students work is wrong, I want to write about how wrong it is."
  2. Start by saying that you think (be non-deterministic, there's still a chance you're wrong) you have found some issues with the previous work.
  3. Present a clearly written memo detailing the edge cases and why they break the existing work. Include an "out" for the previous student and the advisor. Graciously say that the sample dataset did not include these edge cases so couldn't have been proven broken at the time. Do NOT say that you suspect the previous guy fudged the data.
  4. Allow the professor time to digest and draw his own conclusions. (They may have already had suspicions about these edge cases so they may be pretty accepting of this news.)
  5. Listen to the feedback. Make a mental distinction of feedback that is their immediate reaction and feedback after "a while".
  6. Calmly take in the feedback and spend some time going over to see if it is possible you missed something. Give your rebuttal after "a while". (Don't immediately react to your advisor. You'll say something that you'll kick yourself for later. Your absolute main rebuttal point should be the very first one your advisor hears because people tend to block out the next few arguments while they argue the first one in their head.)
  7. After your advisor has accepted that the previous work had flaws, propose the possibility of a paper about why the previous method is incorrect.

This should happen over the course of several days. You both are very good at your technical specialty, but you're both human. This is a social situation and you should approach it as such.

  • I like the point of creating an out in (3), this has helped me deal with some horrible situations and get the things I wanted. Also +1 for its the fault of the data – Repmat Sep 28 '16 at 20:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.