Research Anything AI 101

Acknowledgement: This post is based entirely on Dr. Vinh Dinh Nguyen’s presentation and slides from the course AIO2025 offered by AI VIETNAM.

Relevance: The audience of this lecture were those who are setting foot into the world of AI and pursuing an AI-related career including scientific research. However, skills mentioned here, I think, should be applicable to anyone who desires to explore an arbitrary subject in depth.

In today’s post, we’ll look at some essential skills that a person is encouraged to develop as a starting point for doing research whether in AI or any other field.

How to Find Scientific Documents/Papers

Having the right materials to read up besides textbooks or lecture notes is crucial while you study or research any subject. However, such materials do not just magically appear in front of us. You need to not only dig through the internet for them but make sure their origins are reliable as well.

There are two primary types of resources that scholars use to suppport their projects including published literature and preprints. We will now talk about where to obtain them and what their differences are in terms of reliability.

Published Literature

Published literature refers to peer-reviewed sources which are formally released by recognized publishing companies and made available to the public. Here are several places where such documents/papers in the field of AI can be found using keywords.

Google Scholar is the first go-to webiste for researchers of all areas. This is a fast and straightforward way to learn about articles suitable for your research works. The website provides the sources of most (if not all) academic papers with different versions or publishers and related information. You can narrow down results by using year filters. Say, when you are looking for a literature review/survey since it’s recommneded at papers printed within the last 3-5 years.
IEEE contains scientific and technical documents issued by itself and its partners. Using this website, you can look for related publications based on its content type such as conferences and journals, its affiliation,and many more.
PubMed is a free database for those interested in using AI in biomedical and life sciences. Readers can While full-text archives by PubMed Central (PMC) are free, access to other publishers’ articles may require a fee.
Other Resources: Trusted publishers such as Springer, ScienceDirect, and Elsevier provide access to scientific papers across various field.

Preprints

Even though published literature are validated resources, the peer review and editting process can take about 3-6 months or even a year or more. It means researchers do not have full accesss to up-to-date techniques, which delays the whole research community’s progress. With that in mind, scholars created sites allowing people to post their research works before or while they are submitted to a journal or conference.

Keep in mind that these papers may introduce the lastest methods, but they might not have been evaluated by experts yet. Here are a few popular places where preprints can be checked out.

arXiv offers more than 2 million free scholarly articles in the field of physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics.
bioRxiv specializes in biology and life sciences. Formal version of many articles are updated weeks after their published date.
Papers with Code is a free and open resource specifically for Machine Learning papers. Even though the site does not require the corresponding code implementations upon submission, source code and datasets are suggested being included. This is a great way for readers to understand, reproduce, and improve ideas and experiments presented in these research articles.

How to Read Scientific Documents/Papers

While both scientific papers and essays heavily involve writing, it’s worth noting that they are quite disparate. In few words, scientific papers are factual whereas essays’ tone is biased. The table below comprises key differences between the two.

	Scientific Papers	Essays
Purpose	contribute new knowledge to a specific field	argue a point, explore a topic, or reflect on an issue
Target audience	scholars, researchers, scientists, and professionals	general audience or special readers
Focus	empirical evidence, methodologies, and data analysis	critical thinking, analysis, and personal perspectives
Structure	introduction, literature review, methodology, findings, and conclusion	introduction, body, and conclusion

Reading a research paper can be challenging, especially for first-timers; it can even take them a month to understand one. It is common to struggle with new terms and algorithms since an absolute beginner needs to learn from references. Don’t be discouraged; it gets easier over time. Your reading pace and comprehension will improve if you’re consistent.

If you are totally new to research (like me), you may wonder: Should I read the whole paper? If not, which parts should I skip? Should I try to understand every single detail or aim to get the big picture?

All these boil down to one concern: How do I read a research paper efficiently? We’ll now discuss Dr. Vinh Dinh’s reading technique based on his first-hand experience step by step.

Title: The very first thing you should do when picking up a paper is to read its title. The title consists of keywords that helps determine whether a paper matches your research topic or not. If the title has or is related to the name of ideas or techniques you are interested in, you can move to the next step. Otherwise, you can pass.
Abstract: Once you’re good to move forward, read its abstract. The abstract is considered the essense of an article, summarizing every remaining section within 250-300 words. It enables you quickly assess its relevance and value to your work so you can decide if the paper is worth your effort.
Introduction: If the abstract is intriguing, start reading the introduction section. The introduction shows you the big piture of what this research is about. Here you’ll learn about the research context and questions which all help layout the paper’s structure and main arguments. Basically, this is where scholars sell their papers.
Experimental Result: It is imperative that you read this section right after the introduction because how they advertise their “product” and how good their ideas are may not be what you are looking for. You’d like to know if what the authors propose is actually practicable in your case and provides the result meets your expectations. This section may contain data visualizations and statistics helping you make a decision. The idea is for you to skip some long parts of the paper that may be unnecessary to you.
Proposed Method: If you’re satisfied with the result, you can go back to the proposed method to learn how they produce, brainstorm, and implement their innovative ideas to resolve their research problem. Note that it may take up to a couple of weeks to grasp the picture. Also, you won’t, most of the times, be able to comprehend their method immediately. That’s when you read the related works or look up references.
Related Works: This section is a survey of previously related works and their drawbacks. Readers can refer to this part and the Reference section to obtain relevant knowledge, techniques, and algorithms.

NotebookLM and Prompting Skills

We’re talking about doing research in AI; why not make good use of the product of the field? NotebookLM allows you to upload a scientific paper and assists you with summarizing the paper or explaining a concept. Incorporating this site into your study can speed up the process quite a bit.

To utilize this tool effeciently, you need to have prompoting skills. Without a SMART question, an AI assistant may yield results that are too broad/general or contain incorrect attached resources. These skills will be discussed later in the course.

How to Make a Research Plan

Similar to reading a scientific paper, you can use AI to learn AI when it comes to making a research plan. One recommended AI assisting you to complete this task is Gemini Deep Research. It can generate a detailed document with just a simple prompt: “I want to conduct research on object detection using deep learning models; help me write a research plan.” You can try and see what it can do.

Where to Implement Your Code

You have ideas; you want to put it into action. Here are a couple of places where you can do so.

Jupyter Notebook: You can use it via either Python Compiler or Anaconda. The latter one has various tools in addition to Jupyter Notebook and is easier to install. Since this is run locally and AI training requires GPUs with high VRAM, you computer’s system resources might not be sufficient. Google Colab, which is mentioned below, can resolve this issue to a large extent.
Google Colab: Google Colab provides free limited (but enough to start with) online resources. You just need to connect with your gmail account and begin using it without any installation. This site allows you to mount your notebooks to Google Drive or Github and download/upload them to/from your computer and has built-in Gemini AI. In Colab, your code runs on CPU by default; however, you can change runtime type to GPU or TPU.

AI Coding Assistants

In addition to Google Gemini, consider using ChatGPT and Microsoft Copilot, and what they can do for you. Of course, this again requires your prompting skills.

Generate a sample code: “Generate a Python example that uses Matplotlib and plots a square function from -10 to 10.”
Explain an existing code step by step.
Add comments to an existing code.
Debug an existing code and suggest a solution: “This program has anunexpected result. Find it out.”
Design a solution for a program based on a prompt or paper: “I have an ID array of numbers. I want to get themaximum number. Design an algorithm and give apseudo-code using Pythonic style.”
Explain theory/concepts: “What is theLambda function in Python?”

I encourage all of us try prompts out with AI and see how it performs and how you can make it work better for us.

While AI assistants can help you learn coding faster, overreliance on AI can reduce your critical thinking and programming skills if you first learn coding.

You should verify anything AI assistants provide. They are just your assistants! They can be wrong.
-Dr. Vinh Dinh Nguyen

How to Document Your Results

While you can use Microsoft Word to make a document, it’s probably universal that scholars document their research works using LaTeX, which is more convenient for professionals to create scientific and technical content.

You can download softwares like MikTex and TeXworks (included in MikTex’s package) to use and preview offline.
You can also write LaTex online using Overleaf. The perk of using this is that it allows multiple collaborators to edit on the same file simultaneously.

FAQ

Here are some good questions asked during class. The answers to these questions came from either some admin, TA, STA, or members of AIO2025.

Questions	Answers
A project may have a large file of codes and repositories. Where do I start to read? How can I find its codebase?	Repomix packs the source code into one .xml file which you can ask AI to identify the codebase.
How can I use Markdown?	Markdown Cheat Sheet
what do I start with when using Overleaf?	AIO Template by TA Nha and LaTex Cheat Sheet
Where do I start when it comes to coding, specifically Python?	Python Tutorials w3schools, Python Tutorials YouTube, and Python 35 Days by TA Nha
I want to review math for machine learning. Which resources can I use?	Mathematics for Machine Learning
How do I install Python on VSCode?	https://www.youtube.com/watch?v=20XpOtMNWs4 (in Vietnamese)
Where can I check ranking of a journal?	Scimagojr