Writing a machine learning paper as an independent researcher

Question

I have completed my undergraduate degree and do not have any universities nearby where I can find research work. I have been working in machine learning and want to write a research paper in this field.

I read quite a few research papers as a part of job description but am unable to come up with a new idea for a research paper. How do I go about finding an idea? Generally, if I find something, it is way too broad. For example, I read about hypergraph decompositions or HTMs and got really psyched about them. I couldn't get to a point though which would be a concentrated topic that could lead to a reproducible research.

How do I go about finding ideas for writing a paper considering I don't have anyone to work with me or guide me in the process?

Answer 1

First point - don't force it. Getting ideas for academic papers comes over time and has to be a natural process. The more you try to get good ideas for papers, the more unnatural the process can become and the less likely you will be to come out with something worthwhile.

The best way to get a good topic is just to carry on reading around in your field as you normally would. At some point you may well come across a topic or a research gap that needs to be filled and one which excites you and stimulates you at the same time. That is the point where you can start planning your research.

Bear in mind that a good research article topic can take time to discover. It can be weeks - or months - before you come across something worthwhile and that you feel you can turn into something research-worthy. You don't necessarily need anyone else to do this with, but you do need natural drive, enthusiasm and acumen in your area. If you have these, you will come across a topic and be able to write about it and conduct research effectively. But don't search too hard: writing academic papers should be a natural process and you'll find that something will present itself soon enough.

Answer 2

There are lots of different types of academic papers, including review articles, novel ideas, etc. While the types of papers varies across a wide range of topics, many academic papers comes from finding something where you can legitimately say one of the following things:

• I figured out something interesting that is not yet known/published;
• I can do things (slightly) better than they are presently done;
• I applied this broad method to this special case and came up with this interesting result;
• I abstracted from this special case to this broader theory/method.

It is great that you are getting excited by reading these research papers, but if you want to publish your own work then you need to be on the lookout for interesting ideas, questions, etc., that are extensions of what are in those papers. For example, if a paper raises an interesting question that is not in the literature, maybe you can ask that question and figure it out. Similarly, if you see some aspect of a method that might be done differently (and you think that might be better) then have a go at it and see if you can do something slightly better than it is done in the literature. You can write an paper where you show how a general theory/method reduces down in a special case, or you can write a paper where you extend a theory/method to greater generality than it is presently applied.

When we read papers we are often trying to teach ourselves an area of practice. An inquisitive researcher usually starts playing around with the problems they are reading about to try to learn about them. Many academic papers come about because the researcher is playing around with existing methods and procedures and they start asking questions that are not answered in the papers they are reading. For every five or ten cases where you play around with a problem you may find that one particular idea/case emerges that you pursue further and it leads to a publishable paper.