I have been working on improving a loss function to account for hierarchies in the data. I seem to be getting promising results so far, and I would like to publish a paper. I do not come from an academic or research background, and have no idea how the whole process works. A few questions I have are:
Where can I search to see if someone else has done what I am doing or similar before (both for research and proper citation as well as to ensure I am not just re-inventing a previously published paper)? I am only familiar with ArXiv, but are there other key places I should be checking? Is it important to show what I am doing in math notation, or can python code and explanations suffice? How do I know that this is a good general idea vs something that happens to work in the small datasets I am testing it on? Should I focus on 1 dataset for a first paper, then if it works well across many datasets write a follow up - or should I start with a wide scope if possible? Are there any resources that can help guide me through the paper-writing process?
Also - For someone who has gone through this process I am open to any level of engagement you desire ranging from solely answering this forum post, to answering questions on a call, to advising, to co-authoring (assuming this turns out to be paper-worthy).
I’ve never written a research paper before, but here are some thoughts:
You can’t just upload your paper to arXiv: you need to be endorsed by someone. This bit of gatekeeping is to make sure people don’t just spam arXiv with crap. See also https://arxiv.org/help/endorsement
I would just start writing the paper and see where you end up. Becoming proficient at writing is mostly a matter of writing a lot. The first draft of your very first paper is probably going to suck. It’s part of the process.
I think it’s a good idea to share your first draft here on the forums to solicit feedback. You may find someone who is willing to improve the paper / co-author it with you. (But do provide some kind of first draft already. It’s much easier to find collaborators if you’ve already made a start.)
You can find the typical format of a paper online. Typically it’s something like this: You start with a description of the problem and a short summary of your solution. Usually there is a section on related work. Then you explain your method in detail. Next, you show the results of experiments. Make sure to also include experiments that show it’s really your new method that creates the improvements and not some random thing (ablation study). Summary and conclusion. References. Appendix or supplementary material (if any).
Use LaTeX or Word (but preferably LaTeX).
It might be a good idea to start with a blog post rather than a paper. You need less rigor for a blog post and it will help to find a co-author.
What is needed in the actual contents of the paper (i.e. math) depends on where you’re going to submit the paper (like a particular conference or journal). If there’s not enough math, the reviewers for that conference / journal will probably tell you.
Also haven’t written a paper before, but you could consider writing a shorter paper for your first attempt. This FixRes note/paper is a nice example of a slightly shorter paper format which was designed to just update a previous paper https://arxiv.org/pdf/2003.08237v4.pdf
The process took three years from submission to publication. The publishers will do an initial assessment as to whether it is original work. Your paper then goes out for peer review to experts who will critique and may suggest further work or revisions. I had a couple of revisions along the way.
I suggest identifying the journals that cover your area and to which you want to submit. Their website will have submission guidelines and format specification. You can look at their previous editions to see if they have published your topic previously to see the state of the art.
I’d also recommend a general search on scholar.google.com to see what has been published before. And you’d also need this to cite previous work anyway in your paper. Actually this would be the place to start as you need to know what has been done before to know whether it is even worth starting on writing something.
I submitted my first paper to the IEEE few days ago. I also came from a non-academic background and tried to answer your questions with limited experience.
Based on your ArXiv paper titles, you can use ArXiv Sanity to search related/similar papers. Then, use I use Mendeley to organize.
Depending on where you publish to put math notations within a paper or as an appendix. Most of the time, pseudo-codes are more sufficient for a paper but you can link your GitHub for codes. You can use Snip Notes to generate math notation in LaTeX. Life is much easier.
Again, depending on where you publish. Just a general rule of thumb, the higher the impact factor is; the more work and data/proof/evidence is required. There is a lot of work involved even a short paper.
Do your literature review to ensure you want to publish;
Identify the journals/conferences you want to publish. Their instructions (and restrictions) can guide you through the submission process.
Most of the publishers accept Words documents, but eventually, I learnt LaTeX. I use Overleaf as recommended by the IEEE. (Note: It is free and you can collaborate remotely.)
I will strongly recommend NOT to share your finding publicly at this stage. During my submission, I need to declare the work is original (fair enough) and NOT published elsewhere. It seems the copyright will transfer to the publisher once the paper is accepted. I am not 100% sure at this stage.
I submitted to a special issue of IEEE. So, there is a hard deadline. But, I will find out an outcome by a given date. The possible outcomes can be:
b) accepted with minor changes;
c) accepted with major changes; or
From my understanding, the outcome will come with reviewers’ feedback. So, you can improve your manuscript further.