Academic Paper Writing

Academic Paper Writing Skill

📋 Metadata

• Skill Name: academic-paper-writing
• Purpose: Generate high-quality research papers by learning structural, rhetorical, and argumentative patterns from top-tier conference papers
• Target Venues: NeurIPS, ICLR, CVPR, ICML, ACL
• Quality Level: Best Paper / Oral / Spotlight standard

🎯 When to Use This Skill

Use this skill when you need to:

• Transform a research idea into a complete paper draft
• Structure your paper following proven patterns from award-winning papers
• Write compelling introductions that set up the problem and contributions
• Craft clear and logical method sections
• Present experimental results with proper argumentation
• Ensure your writing style matches top-tier publication standards

⚠️ CRITICAL RULES

Never Hallucinate Citations

WRONG ❌: Generating fake references like "Smith et al. (2023). Deep Learning Advances. arXiv:2301.12345" CORRECT ✅: Only cite papers that you have verified exist via:

• ArXiv API search
• Semantic Scholar API
• User-provided bibliography

Always Follow the Learned Patterns

The skills extracted from top papers are battle-tested. Don't deviate without good reason.

🔄 Core Workflows

Workflow 1: Generate Paper from Idea

Input: A research idea (1-2 sentences) Output: Complete paper draft with all sections

Steps:

1. Idea Expansion (5 minutes)

   # Use the idea_expander.py script
   python3 paper_agent/idea_expander.py --idea "Your research idea here"
   

   • Identify the core problem
   • List potential contributions
   • Suggest related work areas

2. Outline Generation (10 minutes)

   • Apply structural patterns from data/skills_base.json
   • Create section-by-section outline
   • Define key messages for each section

3. Section-by-Section Drafting (Iterative)

   # Draft each section using learned rhetorical patterns
   python3 paper_agent/generator.py --section abstract --outline outline.json
   python3 paper_agent/generator.py --section introduction --outline outline.json
   python3 paper_agent/generator.py --section method --outline outline.json
   # ... continue for all sections
   

4. Refinement (20 minutes)

   • Check logical flow between sections
   • Ensure contributions are clearly stated
   • Verify experimental claims are supported

Workflow 2: Learn New Skills from Papers

Input: PDF of a top-tier paper Output: Updated skills database

Steps:

1. Convert PDF to Markdown

   python3 paper_agent/processor.py --pdf path/to/paper.pdf
   

2. Extract writing skills

   python3 paper_agent/skills_analyzer.py --md path/to/paper.md
   

3. Update skills base

   # Skills are automatically appended to data/skills_base.json
   

Workflow 3: Refine Existing Draft

Input: Your draft paper (LaTeX or Markdown) Output: Improved draft with suggestions

Steps:

1. Analyze current draft against learned patterns
2. Identify gaps in argumentation
3. Suggest improvements for each section
4. Regenerate weak sections

📚 Learned Writing Patterns

Pattern 1: Introduction Structure (from ICLR Best Papers)

Hook → Gap → Approach → Contributions

1. Opening Hook (1-2 sentences)
   - State the broad problem or exciting opportunity
   - Example: "Large language models have revolutionized NLP, but their computational cost remains prohibitive for many applications."

2. Gap Setting (2-3 sentences)
   - Identify what's missing in current solutions
   - Example: "While quantization reduces model size, existing methods suffer from significant accuracy degradation on complex reasoning tasks."

3. Our Approach (2-3 sentences)
   - Introduce your method at a high level
   - Example: "We propose AdaptQuant, a training-free quantization method that preserves reasoning capabilities by..."

4. Contributions (Bulleted list)
   - List 3-4 concrete contributions
   - Be specific about what's novel

Pattern 2: Method Section Flow

Overview → Components → Algorithm → Analysis

1. Method Overview (1 paragraph)
   - High-level description
   - Key intuition

2. Component Breakdown (2-3 subsections)
   - Each component gets its own subsection
   - Include mathematical formulation
   - Provide intuitive explanation

3. Algorithm (Pseudocode or detailed steps)
   - Make it reproducible
   - Highlight key design choices

4. Theoretical Analysis (Optional but strong)
   - Complexity analysis
   - Theoretical guarantees

Pattern 3: Experimental Argumentation

Setup → Main Results → Ablations → Analysis

1. Experimental Setup (1 subsection)
   - Datasets
   - Baselines
   - Metrics
   - Implementation details

2. Main Results (Tables + Analysis)
   - Lead with your strongest result
   - Compare against all baselines
   - Highlight key takeaways in text

3. Ablation Studies
   - Justify each design choice
   - Show what happens when you remove components

4. Qualitative Analysis
   - Case studies
   - Visualizations
   - Error analysis

🎨 Style Guidelines (from Top Papers)

Vocabulary

• Prefer: "We propose", "Our method", "We demonstrate"
• Avoid: "We believe", "We think", "Obviously"

Sentence Structure

• Average length: 15-20 words
• Mix short (impact) and long (detail) sentences
• Use active voice for your contributions
• Use passive voice for background/related work

Paragraph Flow

• First sentence: Topic sentence (what this paragraph is about)
• Middle sentences: Supporting details
• Last sentence: Transition or conclusion

Common Phrases from Best Papers

• "Building on this insight, we..."
• "To address this limitation, we propose..."
• "Our key observation is that..."
• "This suggests that..."
• "We empirically find that..."

📊 Conference-Specific Requirements

🔗 Deep Dive Resources

For detailed guidance on specific topics, see:

• LaTeX Templates - Conference-specific templates
• Citation Guidelines - How to properly cite and verify references
• Figure Design - Creating publication-quality figures
• Rebuttal Writing - Responding to reviewer comments

🚀 Quick Start Example

# 1. Set your research idea
export IDEA="A new attention mechanism that reduces computational complexity from O(n²) to O(n log n)"

# 2. Generate outline
python3 paper_agent/main.py --mode outline --idea "$IDEA"

# 3. Generate full draft
python3 paper_agent/main.py --mode draft --outline output/outline.json

# 4. Review and refine
python3 paper_agent/main.py --mode refine --draft output/draft.md

📈 Success Metrics

A good paper generated with this skill should:

• Have a clear problem statement in the first paragraph
• List 3-4 concrete contributions
• Include proper related work comparison
• Present reproducible experimental setup
• Show ablation studies for key design choices
• Have no hallucinated citations
• Follow the structural patterns of top papers

🔄 Version History

• v1.0.0 (2026-01-24): Initial release with patterns from ICLR 2024 Best Papers