Research paper in machine learning during my bachelor’s

Publishing a research work can be a lot harder without a proper path

Photo by Riccardo Annandale on Unsplash

"Is publishing a research work that hard? and is it worth it ?" this question indeed comes to mind of any machine learning and data science practitioner now and then. So today, I will be trying to clear some of your confusion and walk you through how I ended up being an author of a research paper in my bachelor's. Along with that, I will be providing a few tips and tutorials about how we should proceed with specific tasks.

If someone asks me how to describe doing research work, it is like a roller coaster ride full of ups and downs. For a few moments, it may seem we have reached a world-changing breakthrough; the next moment, it appears to be the worst idea of the world. It all started with my interest in machine learning in the first year of my B tech, everyone can have a different story, but we all ended falling in love with machine learning. At that time, not being a student of CSE, I started my journey with online MOOCs, tutorials, and our favorite medium; it was then I was introduced to Towards data science, and I wanted to thank TDS that I have learned a lot from it and now is capable of expressing myself to you all.

Until the mid of the year, 2020 covid -19 was at full throttle in India and has also affected our colleges, and all our classes were online; that's when I began my ML journey. Surfing the internet for courses, tutorials and blogs had become a hobby for me. By the end of 2020, I also started applying to a few work-from-home internships. I got a few, it was a great learning experience, and by now, I know what will I be doing my whole life ML ML ML !!!! But somehow, I was not satisfied with the same routine of reading datasets applying the same model; again and again, what I wanted was something new, something unique, something that would require me to use my full brainpower.

The first step towards the final goal

A thought that was long troubling me was is it possible to do some research work in your bachelor's, and it feels like and will it be worthy. And, now I can answer these questions. Believe me working on a research paper, finding something new and novels; this feeling is one of the best in this world, and yes, it is totally possible and worth it. The next question I stumbled upon is how should I proceed ?? who will guide me ?? And let me tell you the best thing is to ask a professor or Ph.D. scholar to guide you. I want to summarize the steps that I have followed to get me one of the best professors in my institute.

1. Firstly, you need to prepare a list of mentors, primarily professors from your institute or other institutes who are interested in the field of machine learning. Try finding those professors who have the same domain of interest as you.
2. Then go through their websites, their research works, their research scholars, and all projects they did, and then fix a domain that you would like to work on like Computer vision, NLP, etc.
3. Try to do some good projects based on the topic chosen, they don't need to be fancy, but simpler ones will also work. But, these projects must show that you are interested in working under the professor or any guide.
4. Now the last and most crucial step, mail the professor clearly stating your ambitions and what topic you want to work on and talk about your earlier achievements and how you read all his/her work and wanted him/her to guide you.

I am attaching a sample mail that I sent to one of the mentors who selected me(he is considered one of the best in my institute):

Email template. Source: Self-made

Just keep going

Remember that writing and even applying for a mentor will be time-consuming, so keep patience and keep going with it. I told you just now that I got selected under one of the best professors, but it is a fact that I took a lot of time, as a bachelor is never preferred over to a Masters's or doctorate person. But I tried to various professors several times and even had to send reminder emails without getting any replies, you must understand these peoples are busy, so have your patience. I was even selected once under a professor, but due to some miscommunication was rejected. But there is one friendly research scholar who took my interview; he guided me and even helped me get a chance to work under the same professor, and this time for real, no miscommunication.

So, I wanted to say that it may take time; you may have to send reminder emails, some may not even reply, but have your patience and try to build and expand your connections; it will help a lot.

A piece of advice, don't try to send mails to multiple professors of the same institute. This may lead to clashes between professors, and believe me, you don't want that.

How should we approach a research paper ???

Once someone is ready to guide you, it becomes a lot easier. Generally, a professor assigns one of his Ph.D. scholars as your mentor to help you out. I, too, was given the same Ph.D.hd scholar who interviewed me and helped me get this research intern. The next thing is to discuss the topic you can work on with your mentor. He may ask you to research to get a list of potential topics and discuss them with him. For this, I will provide a bit of a guide on choosing a project to work on. This will work even if you are not guided by anyone and want to work independently. But remember, if you decide to work independently, do let a professor check your paper; this is one of the factors that can help your writing in getting accepted. The steps one should follow while choosing a topic of research are as follow :

1. Firstly, if you get lucky and get a mentor, ask him about the hot topics in the field you want to work in. Otherwise, do some research on your own. At this time of writing this article in January 2022, some hot topics that you can go for are:

• Federated machine learning: Its application in healthcare, transportation, and cybersecurity.
• Adversarial machine learning: related to fooling ML models like creating fake adversarial attacks
• Natural language processing: Topics like text categorization, text summarization, Text generation, and sentiment analysis.
• Computer vision: Medical image analysis, image and video compression, video summarization, pattern recognition, facial recognition, emotion recognition
• There are other topics that can be applied to any area like Interpretable AI, Multi-modalities-based recognition or analysis, Generative models, etc.

There are many more active fields of research, do your research properly to choose an unsaturated area. It will be easier to write your first paper in such an area.

2. The next thing to be done after choosing your topic of interest is selecting a suitable dataset. Selecting the right dataset is one of the most critical steps in writing a successful paper. If you choose the wrong dataset, you can wander around that dataset for days and achieve nothing. So, collect a list of a few datasets related to the topic, filter out the ones used in other good papers, and choose papers from some good publications. Try selecting some cleaner and more extensive datasets having a large size; this impacts your paper acceptance rate.

3. Then, collect a list of conferences with their submission date. The conferences must be according to the topic chosen as most of the conferences are field-specific such as audio or visual etc. You can get a list of conferences and their deadlines from this AI Conference Deadline. ( But note that this contains only the best conferences in the field of machine learning, there are other good conferences also, so do some research about them).

4. After choosing the topics and suitable datasets and conference, we are all set to go. Now make a timeline so that the work can be done within the required deadline. Then you have to start experimenting. Make some hypothesis, experiment this out, compare the results with the existing state-of-the-art model, experiment again and keep this cycle going until you get some result. A piece of advice always try to include at least two novel ideas or approach

The most important and the final step

I assume you must have a mentor guiding you or must surely have chosen a topic on your own to work independently and have started doing experiments. This step generally takes most of the time. If I have to divide, then 30% time goes for deciding the topic and data collection, and 50% of the time must be allotted for experimentation rest must be assigned to the most important and deciding step, the paper writing and the presentation of the paper. A good presentation can get a simpler paper selected easily in good conferences, even if there is less novelty. So, you must know what must be included in the paper and how to present it to be concise and easily understandable. I will summarize the points that I think one should remember while writing a paper in the next few steps.

1. Firstly, remember to do in-depth research to find and include any and all state-of-the-art papers and their results in your paper, and try to choose those datasets in those papers. It is essential to show comparison. Not including any such paper can be a problem, and reviewers can question you.
2. Next, it is essential to write your Introduction section to clearly state what problem you will address and your novel contribution. Try including at least two contributions.
3. As mentioned earlier, an in-depth study is essential to include all the important papers in the field; this must all be mentioned in the Literature survey section of your paper( remember, some pieces don’t require a literature survey, so read the conference instructions carefully).
4. Suppose you propose some architecture or any new algorithm; try to include an exemplary diagram representing your work. In that case, this helps the reviewer to understand your paper in a better way. Also, you must consider including mathematical equations and pseudo codes if possible, this must all be done in the Proposed work section.
5. Remember to do a proper Ablation study, showing how you got to the proposed approach. If a table is possible, try to include that and report all the findings, whether accuracy metric or some time complexities, etc.
6. Now, in the Result section, try to include any confusion matrix, any images of the result, and give a brief description of what they represent. Also, include tables to compare your proposed approach with the state-of-the-art approach.
7. At last the Conclusion section, it is very important to conclude your paper properly. Summarize all your findings, your results or tables, or how much better your results are from the state-of-the-art.

Note: Remember to include all the references used in the pape; leavingg any can put you in some serious trouble.

Now, you are all done from your side, submit the paper within the deadline and pray to God for the good news. Most of the conference paper has a rebuttal round, in this reviewer ask some questions that are not clear to them about your work. Answer them concisely and adequately, which can lead to selecting your paper.

Conclusion

In this blog, we have discussed how we can start writing a research work from a beginner. We discussed the steps to get a mentor and how to approach one, next, we discussed how to choose a topic and discussed mistakes that mustn’t be done while writing a paper. At last, the most essential thing in this field is being patient, we will get many ups and downs, and will sometimes feel that you won’t be able to produce some work but be patient and you will figure some way out. May the spirit of “Andrew Ng” be with you.

Thank you,

If you like the content do follow me for more amazing content and subscribe to my mailing list, to avoid missing anything.

Get an email whenever Sarthak Malik publishes.