Academic Outline

Academic Outline Skill

Use the available context to plan the requested scope. The templates and sentence counts below are optional starting points, not quotas. Preserve an effective existing structure and omit sections that do not serve the argument.

When useful:

1. Identify the document type from the user's request (conference paper, workshop paper, proposal/grant, thesis chapter). Infer it from context where possible; ask only when the distinction changes the outline.
2. Identify the page limit or venue if mentioned (e.g., "8-page NeurIPS paper", "4-page workshop", "NSF proposal"). This determines scale.
3. Ask the user for a 2-3 sentence summary of their work: what problem, what method, what result. If already provided in conversation, use it directly.
4. Generate the outline following the templates below, adapting sentence counts and section presence to the document type and page limit.

For empirical papers, organize the outline around what the reader will discover, not a sequence of defenses of the design. Let questions motivate comparisons and let the observations carry the story. Reserve main-text space for readable explanations and the findings needed to understand the contribution. Place detailed protocols, edge cases, and extended reviewer-facing qualifications in the appendix, while retaining material limitations in the main text. Proof-oriented structure remains appropriate for actual theoretical results.

Output Format

The outline should be a structured markdown document. For each section and subsection:

• State the purpose of that section in one line
• State the approximate sentence count or paragraph count
• Describe the logical flow as a chain: point A -> point B -> point C
• Where applicable, note what to cite or reference

Do NOT write the actual paper content. Only provide structural and logical guidance.

Template: Conference Paper (8-10 pages)

Title

• Should contain: method name (if any) + task/problem + key differentiator
• 1 sentence, under 15 words preferred

Abstract (~150-200 words, ~8-10 sentences)

• Flow: problem significance (1-2 sentences) -> existing approaches and their limitation (2 sentences) -> our key idea in one line (1 sentence) -> method summary (2-3 sentences) -> key results with numbers (1-2 sentences)
• Must contain at least one concrete quantitative result

1. Introduction (~1-1.2 pages, ~18-25 sentences)

• Para 1 (3-4 sentences): Big picture motivation. Why does this problem matter? Connect to a broad audience.
  • Flow: broad context -> specific problem -> why it matters
• Para 2 (4-5 sentences): Prior work at a high level and their limitations.
  • Flow: existing paradigm -> what they do well -> critical weakness or gap
• Para 3 (3-4 sentences): Key insight or observation that motivates your approach.
  • Flow: observation/intuition -> why this hasn't been explored -> what it enables
• Para 4 (4-5 sentences): Your method summary.
  • Flow: we propose X -> X works by doing Y -> key design choices -> how it addresses the gap
• Para 5 (3-4 sentences): Results highlights and contributions.
  • Flow: experimental setup in one line -> headline result -> secondary results
• Contributions list (3-4 bullet points): Each bullet is one sentence. Cover: (1) problem formulation or insight, (2) method, (3) empirical results, (4) optional: analysis/resource release.

2. Related Work (~0.75-1 page, ~15-20 sentences)

• Organize into 2-4 topical subsections (not a flat list)
• Each subsection: 4-6 sentences
• Flow per subsection: define the line of work -> summarize key papers (2-3) -> state how our work differs
• Final paragraph (2-3 sentences): position your work relative to the closest prior work
• Tip: can also be placed after experiments depending on venue convention

3. Preliminaries / Problem Setup (~0.5-0.75 pages, optional)

• Include if: formal problem definition, background notation, or assumptions needed
• ~8-12 sentences
• Flow: notation setup -> formal problem definition -> key assumptions -> what we aim to learn/optimize

4. Method (~2-2.5 pages, ~35-50 sentences)

• 4.1 Overview (4-6 sentences): High-level pipeline description. A reader should get the full picture from this subsection alone.
  • Flow: input -> key steps -> output
• 4.2-4.4 Method components (one subsection per major component):
  • Each subsection: 8-15 sentences
  • Flow per component: what this component does -> formulation/equation -> design choices and why -> connection to the next component
• 4.X Training / Optimization (5-8 sentences):
  • Flow: loss function -> optimization procedure -> any tricks or schedules
• Include a method figure reference early in Section 4.1
• Equations should be referenced in text, not left floating

5. Experiments (~2.5-3 pages)

This is the most structured section. Follow this organization:

• 5.1 Experimental Setup (8-12 sentences)
  • Datasets: name, size, why chosen (3-4 sentences)
  • Baselines: list with 1-sentence description each (4-6 sentences)
  • Metrics: which ones and why (2-3 sentences)
  • Implementation details: defer to appendix but state key hyperparameters (2-3 sentences)
• 5.2 Main Results (10-15 sentences)
  • One main comparison table or figure
  • Flow: overall trend (1-2 sentences) -> best result highlighted (1-2 sentences) -> comparison to strongest baseline (2-3 sentences) -> discussion of when/where method wins or loses (3-4 sentences) -> takeaway (1-2 sentences)
• 5.3 Ablation Study (8-12 sentences)
  • One ablation table
  • Test each major design choice by removing/replacing it
  • Flow: what we ablate and why -> results -> which component matters most -> interpretation
• 5.4 Analysis (8-15 sentences, 1-3 subsections)
  • Pick 2-3 of the following based on what strengthens the paper:
    • Sensitivity analysis (hyperparameters, data size)
    • Qualitative examples / case studies / visualizations
    • Computational cost comparison
    • Failure case analysis
    • Scaling behavior
  • Each analysis block: question we ask -> setup -> result -> insight

6. Conclusion (~0.5 page, ~6-8 sentences)

• Flow: restate problem and why it matters (1-2 sentences) -> summarize method and key idea (2 sentences) -> main empirical takeaway (1-2 sentences) -> limitations and future work (2-3 sentences)
• Never introduce new results here

Appendix (no page limit usually)

• Implementation details, full hyperparameter tables
• Additional experiments, full ablations
• Proofs (if any)
• Dataset details, licenses, broader impact

Template: Workshop Paper (4 pages)

Scale down the conference template:

• Abstract: 100-120 words, 6-7 sentences
• Introduction: 12-16 sentences, merge motivation and related work; skip separate related work section
• Method: 1-1.5 pages, 20-30 sentences, fewer subsections (2-3)
• Experiments: 1-1.5 pages
  • Setup: 5-7 sentences
  • Main results: 8-10 sentences (1 table)
  • One ablation or analysis: 5-8 sentences
• Conclusion: 4-5 sentences
• No appendix (usually)

Template: Research Proposal / Grant

Structure varies by venue (NSF, internal, PhD proposal). General skeleton:

• Project Summary / Abstract (~250 words, 12-15 sentences)
  • Flow: problem and significance -> proposed approach -> expected outcomes and impact
• Introduction and Motivation (1-2 pages)
  • Heavier on "why this matters" than a paper; emphasize impact, timeliness, and gap
  • ~25-35 sentences
  • Flow: societal/scientific significance -> state of the art -> gap -> our vision
• Background and Related Work (1-1.5 pages)
  • More thorough than a paper; demonstrate expertise
  • ~20-25 sentences
• Proposed Research (2-4 pages, core section)
  • Organize by aims/thrusts (typically 2-4)
  • Each aim: ~15-20 sentences
    • Flow: research question -> approach -> preliminary evidence (if any) -> expected outcome -> risk mitigation
  • Include a timeline or milestones subsection
• Preliminary Results (0.5-1 page, if available)
  • Show feasibility, not completeness
  • ~10-15 sentences
• Broader Impact / Intellectual Merit (0.5 page)
  • ~8-10 sentences
• References: generous, show breadth

Template: Thesis Chapter

A thesis chapter is typically a paper expanded with more context:

• Add a longer literature review (2-4 pages, organized thematically)
• Expand method section with more intuition, derivations, proofs
• Expand experiments with all results (nothing in appendix, put everything inline)
• Add a "Discussion" section (0.5-1 page) between Experiments and Conclusion
  • Flow: what did we learn -> connections to broader theory -> limitations in depth -> open questions
• Sentence counts scale ~1.5-2x from the conference paper template

Adapting the Outline

After generating the base outline, tailor it:

• If the user provides their specific contributions, map each contribution to where it appears in the outline
• If the user mentions specific baselines or datasets, slot them into Section 5.1
• If the paper is more theoretical (proofs, theorems), expand Section 3 (Preliminaries) and add a "Theoretical Analysis" section before experiments
• If the paper is empirical-heavy, expand Section 5 and keep Section 4 concise
• If the paper has no experiments (pure theory, position paper), replace Section 5 with "Theoretical Results" or "Discussion"

Optional review questions

Use these to check the outline. Include them in the response only when they help the user:

• Does the introduction clearly state the gap?
• Is each contribution backed by an experiment or analysis?
• Does Section 5.2 directly support the claims in the abstract?
• Are ablations testing meaningful design choices?
• Is the method figure referenced and informative?
• Does related work position (not just list) prior work?