On the originality of a master thesis: how far does it need to go?

Question

I am in the process of writing a master thesis and I need advice.

I was reading this topic Can master's thesis be simply a description of an original algorithm? and this one On the originality of a master's thesis topic and I kind of facing the same issues. I need the opinion of people who already finished their master thesis and have some knowledge about it to help me.

The context: I am working on a specific problem in Machine Learning. Whatever the problem is, my idea in the beginning was to develop a neural network, call it NN-A, and apply it to a dataset - simulated or not. I have as a reference a paper that implement this neural network but using a different framework, let's call it NN-B. The neural network is exactly the same but written in a different language and applied to simulated data, one with dimension 5.

I have access to real world data from a real lab. This data is of dimension 20, a bigger problem and more difficult to deal with. I gained access to this data recently and I decided to use it in my NN-A.

The problem: NN-B is publicly available on the internet. The code is all there. That is the main reason why I decided to use a different language to build the NN. The idea is the same, just written in a different language (if you know what I'm talking about, NN-B is written using tensorflow and I will rewrite from scratch using pytorch, but the architecture of the network will be exactly the same). Is it plagiarism? I mean, off course everything will be referenced and the credits will be given to the authors of the original network. But using their code in a master thesis applying to a different dataset adapting their code to this new data structure is ok?

Using exactly the same code they published but applying to this higher dimensional problem and working on the specific adaptations is my plan B. My plan A is to rewrite the entire code using this other framework (pytorch) and apply it to this higher dimensional problem.

The problem with plan A is that this framework is very new to me and maybe I won't have time to finish it. Is plan B enough for a master thesis? If so, is it a master thesis at all? I am afraid if this is plagiarism even tough the code is public and I will give the credits. Understanding the architecture of their neural network and the specific problem I am applying it to is a very complex problem in itself and writing about them seems to me to be a very good subject for a dissertation, but I am not sure if it is enough or if it is a good master thesis. And in the case it is how good or bad can be to 'just' adapt an existing code?

Answer 1

For a Masters, originality i.t.o. novelty/contribution to knowledge is not essential or critical. That is not to say that there are not many masters thesis with novelty and contributions to knowledge.

The requirements vary from school to school and from region to region.

Your scenario is kind of too 'specific'.

If you tweak a model and you apply to a different dataset in a different domain and attempt insight into a problem, you'll most likely satisfy the requirements of a master.

So, think of these for plan A and B

• what's the aim
• what's the objective
• if possible, what's the research question
• what's the problem (attempting to 'solve')
• what's the solution being proffered
• is there a methodological approach and methods to achieve the aim/objectives
  In essence, what's the research you are doing. PS: your advisor will play a key role in these.

Inbtw, redeveloping or modifying publicly available libraries would not count as plagiarism. If it is on platforms like GitHub, you do a fork and documents your code.

It appears you're getting fixated on coding more than on research. Most of us did at masters level. Carry your advisor along and get into the stream of researching.
In some countries, masters is practically all research with no course works.

Answer 2

The requirements for a masters thesis are locally determined. What is required is what the university will accept. Usually this means what the advisor will accept.

But, it isn't normally required that a masters thesis be original publishable research. That is because the thesis is normally time constrained and research leading to publishable results is difficult to time constrain.

Some theses are just surveys of a small field's current literature with an appropriate writeup that shows comprehension and some insight. If it is more than that, all the better. But a good lit review is a first step on the road to more serious research.

Talk to your advisor. Find out what they expect. But if the only difference is a change of language and a new implementation of what has been done, then it might not be enough and your advisor might reasonably ask for more. Adaptation to a new problem may well be enough.