How narrow should my research question be?

Narrow enough that you can state what result would answer it, and what result would refute it. "How can deep learning improve healthcare" is a topic, not a question. "Does adding attention to this segmentation architecture improve Dice score on this dataset" is a question, because you can tell when you have answered it. The literature review is what gets you from the first to the second.

Do I need novel results to publish?

You need a contribution, which is not the same thing. A careful replication that fails, a negative result that saves other people time, a benchmark or dataset, or a clear survey of a fragmented area are all real contributions. What does not work is a small architectural variation with a fractional benchmark gain and no explanation of why it helps.

How do I find the relevant literature?

Start from recent survey papers in the area, since they map the field and give you a citation list to work backwards through. Then follow citations both directions: what the key papers cite, and what cites them. Academic databases and preprint servers cover the search; the part that takes judgement is deciding when you have read enough to know your question has not already been answered.

What gets papers rejected most often?

Not usually weak results. More often it is a mismatch between what the paper claims and what the experiments actually show, missing baselines that a reviewer expects to see, or an evaluation setup that makes the comparison unfair without the authors noticing. Deciding your methods and baselines before you run anything prevents most of it.

How much time should revision take?

More than first-time authors expect. Getting a coherent draft down is roughly half the job; the other half is cutting, tightening, and fixing the places where you assumed context the reader does not have. Budget real time for it, and get at least one outside read before you submit.

Journal Career December 26, 2022

Career · Guide

How to Write a Research Paper in AI and ML

A step-by-step guide to writing and publishing research papers in artificial intelligence, machine learning, and deep learning — from ideation to submission.

Anurag Verma

3 min read

Research AI Machine Learning Academic Writing Career

How to Write a Research Paper in AI, Machine Learning, and Deep Learning

Key takeaways

The seven steps are sequential for a reason: topic, literature review, research question, methods, data, writing, revision. Skipping the literature review is the most common way to spend months rediscovering a published result.
The literature review is what turns a vague interest into a specific, answerable research question. It is a scoping exercise, not a formality to complete after the fact.
Choose your methods before you collect data, not after. Deciding how you will analyse results once you have seen them is how you end up with a finding that will not replicate.
The paper structure (introduction, literature review, methodology, results, conclusion) exists so reviewers can find what they need quickly. Departing from it costs you goodwill you will want elsewhere.
Get a colleague or mentor to read a draft before submission. An outside reader finds the assumed context and unexplained jumps that you cannot see in your own writing.

Hii!, I and many others want to write a research paper but they are not able to write bcoz of unfair guidance and lack of knowledge and confidence here I have provided seven steps you can follow to write a research paper.

Step 1: Choose a topic

Before you start writing your research paper, it is important to choose a topic that is relevant and interesting to you. Consider what aspect of artificial intelligence, machine learning, or deep learning you want to focus on and try to narrow it down to a specific research question or hypothesis.

Step 2: Conduct a literature review

Once you have chosen your topic, you need to conduct a literature review to familiarize yourself with the existing research in the field. Use academic databases and other sources to find relevant papers and articles that have already been published on your chosen topic.

Step 3: Define your research question or hypothesis

Based on your literature review, you should be able to define your research question or hypothesis. This will be the main focus of your research paper and should be clearly stated in your introduction.

Step 4: Determine your research methods

Next, you need to decide on the research methods you will use to collect and analyze data. This could include experiments, surveys, or other methods that are appropriate for your research question or hypothesis.

Step 5: Collect and analyze data

Using the research methods you have chosen, collect and analyze data that will help you answer your research question or test your hypothesis. This may involve using machine learning algorithms or other statistical techniques to analyze your data.

Step 6: Write your research paper

Once you have collected and analyzed your data, you can start writing your research paper. Follow the standard structure of a research paper, including an introduction, literature review, methodology, results, and conclusion. Be sure to include references to the literature you reviewed and any relevant findings from your own research.

Step 7: Edit and revise

Before you submit your research paper, be sure to edit and revise your work to ensure it is clear, concise, and well-written. You may want to have a colleague or mentor review your work to provide feedback and help you improve your paper.

If anyone has suggestions on this topic you are welcome with an open heart…❤️❤️😍😍

Frequently asked questions

How narrow should my research question be?: Narrow enough that you can state what result would answer it, and what result would refute it. "How can deep learning improve healthcare" is a topic, not a question. "Does adding attention to this segmentation architecture improve Dice score on this dataset" is a question, because you can tell when you have answered it. The literature review is what gets you from the first to the second.
Do I need novel results to publish?: You need a contribution, which is not the same thing. A careful replication that fails, a negative result that saves other people time, a benchmark or dataset, or a clear survey of a fragmented area are all real contributions. What does not work is a small architectural variation with a fractional benchmark gain and no explanation of why it helps.
How do I find the relevant literature?: Start from recent survey papers in the area, since they map the field and give you a citation list to work backwards through. Then follow citations both directions: what the key papers cite, and what cites them. Academic databases and preprint servers cover the search; the part that takes judgement is deciding when you have read enough to know your question has not already been answered.
What gets papers rejected most often?: Not usually weak results. More often it is a mismatch between what the paper claims and what the experiments actually show, missing baselines that a reviewer expects to see, or an evaluation setup that makes the comparison unfair without the authors noticing. Deciding your methods and baselines before you run anything prevents most of it.
How much time should revision take?: More than first-time authors expect. Getting a coherent draft down is roughly half the job; the other half is cutting, tightening, and fixing the places where you assumed context the reader does not have. Budget real time for it, and get at least one outside read before you submit.

Newer dispatch

Gender Determination with Morphometry of Eyes Using Deep Learning

AI Integration

Older dispatch

Time Series Forecasting of Stock Data Using LSTM Deep Learning