Chapter 1

I am in hot water. I'm currently working on my Master's thesis in machine learning and things are not going well. The professor that is advising me is a very senior professor at my university and in recent years, some of his Master's students have produced some of the best theses in their graduating classes, and so this professor has a very good track record under them.

In fact, their previous success at producing such great Master's theses is one of the reasons that I chose this professor. When I met the professor several months ago, he proposed two very interesting projects that revolved around tackling a particular theoretical weakness that presents in nearly all machine learning models today - even those that are declared "state-of-the-art".

One of the projects involved developing some broad techniques to tackle this problem and subsequently comparing these techniques to those of other authors on popular, widely-known benchmark datasets.

The other project involved taking existing solutions to the aforementioned problem and applying these solutions to a particular real-world case, and evaluating the results. If time permitted, the plan would be to improve upon these existing solutions with a particular focus on this real-world case.

The professor mentioned that the second project is a "slam-dunk", in the sense that it would be very easy to pass (a requirement for graduation) and still have lots of time left over to produce, test and document novel theoretical ideas which could help one improve their overall grade on the final thesis.

Chapter 2

I ended up choosing the second project and now I am heavily regretting it.

Before I was even able to do anything, the project required me to collect data from some proprietary (i.e. closed-source) simulation software. Because this software is not open-source, there is not a very active community of developers that use it and there are very few open-source plugins that interact with this software.

The company that makes this simulation software provides an SDK that is updated very infrequently; though the lack of access to the main software's source code as well as the lack of publicly available plugins that interface with it, has made this data-gathering process much more difficult than both me any my advisor expected.

Last week, I succeeded in collecting the data, though the models are not performing well at all - they are not much better than random guessing. This might be to do with the quality of data that I collected or my model training procedures and I sincerely hope it is the latter.

Data collection typically takes 3-4 days and my thesis is due in 4.5 weeks. As such, I have to work with what I have and cannot entertain the idea of new data collection strategies if I wan't to meet the thesis submission deadline. Furthermore, because of the aforementioned data collection problems, I am now 5-6 weeks behind schedule and so will likely not have time to follow-through on all the theoretical extensions, algorithm implementations and experiments that I had planned.

Chapter 3

Looking back at some of the recent Master's theses (most of which are very highly-rated) and research papers that have come out of this professor's research group, it appears that none of them have involved development of plugins to extract structured data from closed-source, proprietary software; in fact, it appears that none of them have even involved the collection of entirely novel datasets at all - most of them seem to be focused around developing new theoretical ideas and benchmarking those ideas on well-known publicly available datasets.

Perhaps this professor was simply unaware of the difficulties in extracting data from closed-source (simulation) software - it is almost like trying to squeeze blood out of a stone, to tell you the truth.

Question 1

I know there's not much use in blaming anyone now, so I just need to know, how can I salvage this project and deliver an exceptional thesis? Is there a realistic possibility of delivering an exceptional thesis at the point? (Bear in mind that I attend an Ivy League uni with nearly no slackers in class, so I don't think more man-hours alone will cut it, and I'm looking to see whether you guys have any suggestions for how I can use my remaining 4.5 weeks most efficiently).

Question 2

I'm planning to apply for a PhD in the Fall and this car-crash of a thesis looks like it might just torpedo any chances I have of getting into a good program. A friend of mine who studied at Oxbridge (collective term for Oxford and Cambridge) and left with two good degrees was rejected from 9/10 machine-learning PhD programs she applied to, despite having some significant research experience (albeit, no papers); that is how competitive the current landscape is for PhD admissions in machine-learning (see here: https://twitter.com/leeclemnet/status/1040030107887435776 if you're interested)

Given this, is is possible for me to still be a competitive applicant with a sub-standard Master's thesis? Or have I already been blown out of the water as far as PhD admissions committees are concerned?

Important Note about Question

I know there are a few other similar questions on the site, but I've included lots of detail in this post with intention that the answers below could be more specific to the particularities of my situation as there is typically no one-size-fits-all solution to these Master's theses problems.

Also, feel free to link any other similar question on the site.

  • 3
    The most important thing you’ve left out is communication. How much have you been communicating with your professor during this process? What is the professor’s evaluation of your current progress? Are they aware of your problems and the (hopefully impressive) work you have had to do to overcome them? – Dawn May 14 at 20:54
  • Yes, I've been communicating with the professor on a roughly weekly basis throughout the entire project. I'm much more familiar with all the implementation issues than he is, so hasn't given me much specific advice regarding algorithm implementation issues. The advice he gave me is just to document any significant technical challenges I've encountered throughout the project. My hunch is that he just doesn't want to rock the boat, since I'm a Master's student who'll be out of his hair within a few weeks and not a PhD student who'll be in his lab for several more years. – UnchartedWaters May 14 at 21:25
  • Also, yes. my advisor has said I'm making suitable progress, but because I have not yet run many experiments and evaluations of my machine-learning models due to the aforementioned data collection hurdles, he has not been able to give me much feedback. Other than that, he's left me to my own devices - which I assume is standard for all students at this level. – UnchartedWaters May 14 at 21:30
  • 4
    This all bodes well for you then. A student who has the grit to overcome problems and turn a challenge into something reasonably worthwhile is impressive, if framed well by a letter writer. – Dawn May 14 at 23:48
  • 2
    I can't comment much on your second question, I'm afraid. Regarding how to progress: This is difficult to say, as I imagine most people on this site have little understanding of the content of your thesis. As mentioned by @Dawn, your supervisor's feedback is likely the best indicator at this stage, though I understand your concern about being a few weeks from the end v several more years and how this might affect his comments. Do you communicate with others either in your class or under his supervision? I have found talking to others in my class incredibly insightful in the past. – wgmcgowan May 15 at 20:54

Your Answer

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

Browse other questions tagged or ask your own question.