Reading Advice

Advice for reading AI research papers and academic articles

Authors

Yuriel Ryan

Updated

June 5, 2026

Introduction

Reading research papers is a skill — it takes deliberate practice and good habits. This guide compiles advice from graduate students and professors on how to approach AI research papers and academic articles efficiently and critically.

Note

This isn’t a page that tells you “how to read a research paper”. It’s more of a “you can try these tips because it has worked for us” kind of page. Learning when to skim and when to read deeply is part of the skill which you should apply to this page as well!

Before You Read

Everybody does research a little bit differently, here are some pointers to take note of before we even start

  • Know why you are reading. Are you looking for related work, a specific method, or a broad survey of a field / problem (e.g., AI hallucinations)? Your goal determines how broad or deep you go.
  • Check the venue. NeurIPS, ICML, ICLR, CVPR, EMNLP — knowing the conference gives you a sense of the paper’s quality and focus area.
  • Read the abstract twice. Once to understand the claim; once to identify what you should be looking for in the body.

A great practical tip is to read to understand how to position your work/proposal! Researchers are in many ways content creators amongst a sea of other content creators. The main thing to consider is to find out where your work distinguishes itself among all the content (research) that is available out there.

This means trying to work on something surprising. For example, a method that is unusually simple but effective; a novel approach that no one has tried before; or, a solution that resolve a significant/difficult problem in the existing body of work. There are many more ways to position your work (e.g., proposing a new problem) but hopefully, you get the idea!

The Three-Pass Method

To start you off, the “three-pass method” is a well-known technique to become more efficient in your reading. “Efficiency” here means quickly identifying if a paper is even relevant to your research. There are many new and exciting stuff happening in the AI space everyday — just look at the the number of papers submitted to HuggingFace. Your goal here is to learn how to filter out the noise and grab one or two key works to inform your hypotheses/experiments!

Pass 1 — The Bird’s Eye View

Read: title, abstract, section headings, figures and captions, conclusion.

Goal: Decide if the paper is worth a deeper read and form a broad impression about it’s contributions.

One practical tip to build an impression is to identify the authors and/or their affiliations. Some institutions, or even specific labs within those institutions, prefer a more empirical approach as opposed to a theoretical one. Others focus on industry impacts over social ones — for instance, improving token efficiency vs identifying demographic biases. Depending on the focus of your work, the title and the authors could sometimes provide enough information to proceed onto the next pass.

Note

Overtime, researchers naturally build an internal “bank” of other researchers/labs that they follow. You should build one too to quickly identify which work is worth your (time) investment.

Caution

While a case could be made to filter the quality of the work through authors/affiliations, it’s important to remember that good work can come from anywhere! Yes, maybe I’m a bit of an idealist, but let’s not blind ourselves or discount the work of others simply because they are not affiliated with a prestigious institution.

Pass 2 — Understanding the Contribution

Read the body carefully, skipping proofs and implementation details. Pay attention to figures, tables, and the experimental section.

Goal: Understand what the paper claims and how it was evaluated.

Pass 3 — Investment

Read the remaining sections, including appendices and referenced papers where needed.

Repeat the experiments on their GitHub (if applicable)! This is the best way to understand what the authors are trying to do. More importantly, you can decide for yourself if this work is truly impactful/useful to your project.

Goal: Deeply understand and critically evaluate the work.

Critical Questions to Ask

  • What is the core problem being solved?
  • What is the key insight or novelty?
  • What assumptions does the method rely on?
  • How strong is the experimental validation?
  • What are the limitations the authors acknowledge — and those they don’t?

Starting Points

Part of my research journey was figuring out which conferences are considered “reputable” and/or “impactful”. While this list is non-exhaustive, I listed some of them here to help you (hopefully an aspiring AI researcher) find a starting point.

Conference Full Name Primary Focus & Specializations
NeurIPS Neural Information Processing Systems Machine learning: deep learning architectures; reinforcement learning; theoretical foundations.
ICML International Conference on Machine Learning Algorithmic machine learning: optimization techniques; statistical learning theory; generative modeling.
ICLR International Conference on Learning Representations Representation learning: deep neural network optimization; self-supervised learning; geometric structures.
CVPR Conference on Computer Vision and Pattern Recognition Computer vision: image recognition and synthesis; video analysis; multimodal vision-language models.
ACL Association for Computational Linguistics Natural language processing: computational linguistics; large language model alignment; text generation.
KDD Knowledge Discovery and Data Mining Data mining: large-scale knowledge discovery; graph neural networks; industrial applications.
SIGIR Special Interest Group on Information Retrieval Information retrieval: search mechanics; recommendation systems; web text mining.
AAAI Association for the Advancement of Artificial Intelligence General artificial intelligence: knowledge representation; multi-agent systems; heuristic search.
Note

AI is often cross disciplinary! If you are working on such projects, please do your homework to find out which other venues — both conferences and journals — are considered strong/impactful for your field.

Resources