Skip to content

Chapter 2: How to Read Research Papers

🎓 Learning Objectives

  • Learn efficient paper reading strategies
  • Understand paper structure and sections
  • Develop critical reading skills
  • Learn to take effective notes
  • Master the three-pass reading approach

Why Reading Papers is Essential

Reading research papers is a core skill for any researcher. It helps you:

  • Stay updated with latest developments
  • Understand state-of-the-art methods
  • Identify research gaps
  • Learn from others' approaches
  • Build your knowledge base

Reading Speed

Experienced researchers can read 10-20 papers per week. Beginners should start with 1-2 papers per week and gradually increase.

Common Mistakes

  • Reading papers linearly from start to finish
  • Trying to understand every detail on first read
  • Not taking notes
  • Reading too many papers without depth

Paper Structure

Most ML papers follow this structure:

1. Abstract

Purpose: Summary of the paper (150-250 words)

Contains: - Problem statement - Proposed method - Key results - Main contributions

Abstract First

Read the abstract to decide if the paper is relevant. If not, skip it.

2. Introduction

Purpose: Motivation and context

Contains: - Problem motivation - Related work overview - Contributions - Paper organization

Purpose: Contextualize the work

Contains: - Previous approaches - Comparison with existing methods - Positioning of the work

4. Methodology

Purpose: Technical details

Contains: - Proposed method/algorithm - Mathematical formulations - Architecture details - Implementation specifics

Most Important Section

This is where you'll spend most of your time. Read carefully and take detailed notes.

5. Experiments

Purpose: Empirical validation

Contains: - Datasets used - Experimental setup - Results and comparisons - Ablation studies

6. Results & Discussion

Purpose: Analysis and interpretation

Contains: - Detailed results - Analysis of findings - Limitations - Future work

7. Conclusion

Purpose: Summary and future directions

Three-Pass Reading Approach

The three-pass approach (by S. Keshav) is the most effective way to read papers:

First Pass: Quick Scan (5-10 minutes)

Goal: Determine if the paper is relevant

Steps: 1. Read title, abstract, introduction 2. Read section headings and subheadings 3. Glance at figures and tables 4. Read conclusion 5. Skim references to see if you recognize any

Questions to Answer: - What is the problem? - What is the proposed solution? - Is this relevant to my work? - Should I read more?

First Pass Decision

After first pass, decide: - Relevant: Continue to second pass - Not relevant: Stop here - Maybe relevant: Mark for later

Second Pass: Careful Reading (30-60 minutes)

Goal: Understand the main contributions and methodology

Steps: 1. Read the entire paper carefully 2. Take notes on key points 3. Mark unclear sections for later 4. Pay attention to figures and tables 5. Understand the experimental setup

Focus On: - Main idea and contributions - Methodology overview - Key results - Experimental design

Don't Get Stuck

If you don't understand something, mark it and move on. You can revisit in the third pass.

Third Pass: Deep Understanding (1-2 hours)

Goal: Understand every detail and be able to reproduce

Steps: 1. Read every section in detail 2. Understand mathematical derivations 3. Trace through algorithms step by step 4. Understand experimental details 5. Critically evaluate the work

Focus On: - Technical details - Assumptions and limitations - Potential issues or flaws - Reproducibility

Third Pass is Selective

Only do third pass for papers that are: - Highly relevant to your work - Papers you need to implement - Papers you're reviewing

Reading Strategies by Purpose

Reading to Understand a Field

Goal: Get overview of a research area

Strategy: 1. Start with survey papers or tutorials 2. Read recent papers in the area 3. Follow citation chains (backward and forward) 4. Build a mental map of the field

Field Overview

  • Read 10-20 papers in the area
  • Focus on introductions and related work
  • Identify key researchers and groups
  • Note common datasets and benchmarks

Reading to Implement

Goal: Reproduce or implement a method

Strategy: 1. Focus on methodology section 2. Understand algorithm details 3. Note hyperparameters and settings 4. Check supplementary material 5. Look for code repositories

Implementation Reading

  • Read methodology multiple times
  • Draw diagrams to understand flow
  • List all hyperparameters
  • Note any missing details

Reading to Review

Goal: Critically evaluate a paper

Strategy: 1. Read carefully and critically 2. Check experimental validity 3. Verify claims against results 4. Identify limitations and issues 5. Assess novelty and significance

Critical Reading

  • Question assumptions
  • Check if experiments support claims
  • Look for missing comparisons
  • Verify statistical significance

Taking Effective Notes

Note-Taking Template

# Paper Title
**Authors**: [Author names]
**Venue**: [Conference/Journal, Year]
**Link**: [URL]

## Summary
[2-3 sentence summary]

## Problem
[What problem does this solve?]

## Method
[Key idea and approach]

## Key Contributions
1. [Contribution 1]
2. [Contribution 2]
3. [Contribution 3]

## Results
[Main experimental results]

## Strengths
- [Strength 1]
- [Strength 2]

## Weaknesses
- [Weakness 1]
- [Weakness 2]

## Questions/Unclear Points
- [Question 1]
- [Question 2]

## Related Papers
- [Paper 1]
- [Paper 2]

## Code/Resources
- [Link to code if available]

Note-Taking Tools

  • Zotero: Reference management
  • Mendeley: PDF management
  • Notion/Obsidian: Note-taking
  • Paper notebooks: Traditional approach

What to Note

Essential Information: - Problem statement - Proposed solution - Key contributions - Main results - Datasets used - Limitations

For Implementation: - Algorithm details - Hyperparameters - Architecture specifics - Training details - Code availability

For Literature Review: - Positioning in the field - Comparison with other methods - Citations to follow - Related work

Critical Reading Skills

Questions to Ask

Critical Questions

  1. What is the problem? Is it well-defined?
  2. Is the solution novel? What's new?
  3. Are experiments fair? Proper baselines?
  4. Do results support claims? Check figures/tables
  5. What are limitations? What doesn't work?
  6. Is it reproducible? Enough details?
  7. What's missing? What should be included?

Red Flags

Warning Signs

  • No baselines: Comparing only to weak methods
  • Small datasets: Results may not generalize
  • Missing details: Can't reproduce
  • Overstated claims: Results don't match claims
  • Poor experimental design: Unfair comparisons
  • No code: Hard to verify

Reading Workflow

Daily Reading Routine

graph LR
    A[Morning: Scan New Papers] --> B[Select 2-3 Relevant]
    B --> C[First Pass: 10 min each]
    C --> D{Relevant?}
    D -->|Yes| E[Second Pass: 1 hour]
    D -->|No| F[Archive]
    E --> G[Take Notes]
    G --> H[Add to Reading List]

Reading Schedule

  • Morning: Scan new papers (arXiv, Twitter)
  • Afternoon: Deep reading of selected papers
  • Evening: Review notes and organize
  • Weekly: Review reading list and plan next week

Building a Reading List

Categories: - Must read: Highly relevant, read soon - Should read: Relevant, read when time - Maybe read: Possibly relevant, review later - Read later: Archive for future reference

Reading List Management

  • Use reference managers (Zotero, Mendeley)
  • Tag papers by topic
  • Keep notes organized
  • Review regularly

Resources

📚 Reading Guides
  1. How to Read a Paper - S. Keshav
  2. Efficient Reading of Papers in Science and Technology - Kevin Murphy
  3. Reading Research Papers - Columbia Guide
🛠️ Tools
  1. Zotero - Reference management
  2. Mendeley - PDF management
  3. Papers With Code - Papers with code
  4. Connected Papers - Paper graphs
📖 Practice Papers
  1. Start with survey papers in your area
  2. Read classic papers (cited frequently)
  3. Follow recent papers from top venues
  4. Read tutorial papers for new areas

Next Steps

Key Takeaways: - Use three-pass approach: Quick scan → Careful read → Deep understanding - Understand paper structure: Abstract → Introduction → Method → Experiments → Results - Take systematic notes using templates - Ask critical questions while reading - Build a reading routine and manage your reading list