/SKILL

IEEE Research Paper Generator

Generate a complete, professional IEEE-format research paper from experiment data and findings.

---

Before You Begin

Collect the following inputs from the user before writing any section. If any are missing, ask explicitly:

1. Paper type: conference (8 pages) or journal (variable; ask for target page count)
2. Title candidate: working title or research topic
3. Authors and affiliations: names, departments, institutions, cities, countries, emails
4. Problem and motivation: what problem is solved and why it matters
5. Proposed method: the core technical contribution (algorithm, architecture, framework, model)
6. Experiment data: datasets used, baselines compared, metrics, tables of results, ablations
7. Key findings: what the results show, performance gains, failure cases
8. Related work references: papers to cite (titles, authors, venues, years); ask for at least 15
9. Target venue (optional): informs tone, focus, and length norms

Do not fabricate results, numbers, citations, or author names. If data is incomplete, flag the gap and prompt the user.

---

Paper Structure

Generate each section in order. Use IEEE Roman-numeral section numbering (I, II, III, …). Every section heading must be in ALL CAPS, centered, using the IEEE style.

---

TITLE BLOCK

[FULL PAPER TITLE IN TITLE CASE — CONCISE, SPECIFIC, ≤12 WORDS PREFERRED]

Author One, Author Two, and Author Three
Department of X, University of Y, City, Country
{email1, email2, email3}@domain.edu

• Titles must name the method or problem specifically. Avoid vague titles like "A Novel Approach."
• For multi-institution papers, use numbered superscripts.

---

I. ABSTRACT

Write a self-contained abstract of 150–250 words following this five-part structure:

1. Context (1 sentence): Situate the problem in the broader field.
2. Problem (1–2 sentences): State the precise challenge addressed.
3. Method (2–3 sentences): Describe the proposed approach at a high level. Name the method.
4. Results (2–3 sentences): Report concrete quantitative results. Include dataset names and metrics.
5. Impact (1 sentence): State what this enables or what door it opens.

Rules:

• Do not cite references in the abstract.
• Spell out all acronyms on first use.
• Use present tense for contributions; past tense for experiments.

Index Terms follow the abstract as a comma-separated list of 4–8 lowercase terms enclosed in \begin{IEEEkeywords}...\end{IEEEkeywords}.

---

II. INTRODUCTION

Structure as five parts:

1. Opening motivation (1–2 paragraphs): Ground the reader in why this problem matters. Use concrete statistics or real-world consequences where possible.
2. Problem statement (1 paragraph): Define the problem precisely. State what makes it hard.
3. Limitations of prior work (1–2 paragraphs): Summarize existing approaches and their shortcomings. Every limitation must be tied to at least one citation.
4. Contributions (1 paragraph + bulleted list): Begin with "In this paper, we…" then list 3–5 specific, verifiable contributions as bullets:
   • Each bullet starts with a verb: "We propose…", "We demonstrate…", "We provide…"
   • Contributions must match what the paper actually delivers.
5. Paper organization (1 paragraph): One sentence per remaining section.

---

III. RELATED WORK

Group prior work into 2–5 thematic subsections. Each subsection heading uses \subsection{}.

For each thematic group:

• Summarize the approach shared by the group (2–3 sentences).
• Note the key limitation that differentiates this group from your work (1–2 sentences).
• Distinguish your work from the closest prior work explicitly.

Rules:

• Do not write a flat list of papers. Synthesize.
• Acknowledge concurrent and closely related work honestly.
• Cite all baselines and datasets used in the experiments.

---

IV. METHODOLOGY

This is the technical core. Organize with subsections matching your method's components.

Required subsections (adapt names to the work):

A. Problem Formulation

• Define all variables and notation in a table or inline.
• State the objective function or task formally.
• Example: "Let $\mathcal{D} = {(x_i, y_i)}_{i=1}^N$ denote the training set…"

B. Proposed Method

• Describe the architecture, algorithm, or framework in full.
• Include a system diagram or algorithm pseudocode block where appropriate.
• For algorithms, use the algorithm + algorithmic environment with numbered lines.
• For models, describe each component and how outputs flow between them.

C. Training Objective / Loss Function

• State each loss term as a numbered equation: $(1), (2), \ldots$
• Explain the role of each term.
• State any hyperparameters introduced and their sensitivity.

D. Theoretical Analysis (include if applicable)

• State and prove or sketch key propositions, lemmas, or complexity bounds.
• Label theorems and proofs with \begin{theorem}...\end{theorem}.

E. Complexity Analysis

• Report time and space complexity.
• Compare against the strongest baseline.

---

V. EXPERIMENTAL SETUP

A. Datasets

For each dataset:

• Name, version/split, and source citation.
• Size: number of training/validation/test samples.
• Preprocessing steps applied.
• Any known limitations or biases relevant to conclusions.

B. Baselines

List all comparison methods. For each:

• Method name and citation.
• Why it was chosen as a comparison point.
• How it was run (official code, reimplementation, results from paper).

C. Evaluation Metrics

• Name each metric and provide its formula or definition.
• Justify metric choice relative to the task objective.
• State how statistical significance is assessed (e.g., mean ± std over N seeds, paired t-test).

D. Implementation Details

Report in a \begin{table} or inline:

• Hardware (GPU model, count, memory)
• Framework and version (e.g., PyTorch 2.1)
• Optimizer, learning rate, scheduler, batch size, number of epochs
• Any data augmentation or regularization

This section must contain enough detail for reproducibility.

---

VI. RESULTS AND DISCUSSION

A. Main Results

• Present results in a \begin{table*} spanning both columns.
• Table format: rows = methods, columns = datasets × metrics. Bold the best result per column.
• Caption begins with "TABLE I" and includes a concise description.
• Narrate the table: explain what the numbers mean, not just that "our method is best."
• Quantify improvements: "Our method outperforms the strongest baseline by 3.2 points in F1 on…"

B. Ablation Study

• Present a table isolating each component's contribution.
• Each row removes or replaces exactly one component.
• Discuss what each ablation reveals about the method's design.

C. Analysis and Discussion

Include at least two of the following, as appropriate:

• Qualitative examples: figures with captions comparing outputs of your method vs. baselines.
• Error analysis: categorize failure cases; explain why they occur.
• Sensitivity analysis: plot performance vs. a key hyperparameter.
• Generalization: results on an out-of-distribution or held-out test condition.
• Efficiency: training time, inference time, or memory comparison.

D. Limitations

Dedicate 1–2 paragraphs to honest limitations:

• Conditions under which the method fails or degrades.
• Assumptions that may not hold in deployment.
• Do not bury limitations or frame them as future work only.

---

VII. CONCLUSION

Write 2–3 paragraphs:

1. Summary: Restate the problem, the approach, and the main results in fresh language (do not copy the abstract verbatim).
2. Takeaways: What should a reader conclude about when and why to use this method?
3. Future work: 2–3 concrete directions, each one sentence.

Do not introduce new results or claims in the conclusion.

---

ACKNOWLEDGMENTS (optional)

Standard single paragraph. Do not include in anonymous submissions. Acknowledge funding sources (grant numbers), compute credits, and individuals who contributed but are not authors.

---

REFERENCES

Format all references in IEEE citation style — numbered in order of appearance [1], [2], etc.

Templates by source type:

Journal article:

[N] A. Author, B. Author, and C. Author, "Title of article," *Journal Name*, vol. X, no. Y, pp. ZZZ–ZZZ, Mon. YYYY. doi: 10.XXXX/XXXXX.

Conference paper:

[N] A. Author and B. Author, "Title of paper," in *Proc. Conf. Name (ABBR)*, City, Country, YYYY, pp. ZZZ–ZZZ.

arXiv preprint:

[N] A. Author, "Title," *arXiv preprint arXiv:XXXX.XXXXX*, YYYY.

Book:

[N] A. Author, *Title of Book*, Xth ed. City, Country: Publisher, YYYY.

Rules:

• List references in citation order, not alphabetical.
• Do not cite sources not referenced in the body.
• Every baseline, dataset, and related-work paper must be cited.
• Aim for 20–40 references for a conference paper; 30–60 for a journal article.

---

Figures and Tables

Figures

• Every figure must be referenced in the text before it appears: "As shown in Fig. 1, …"
• Caption format: Fig. 1. Description of figure content in one or two sentences.
• Captions go below figures.
• Figures must have sufficient resolution (300 DPI minimum for print).

Tables

• Caption format: TABLE I \\ Description of table. (caption goes above the table)
• Use \begin{table} for single-column and \begin{table*} for two-column spanning tables.
• Align numerical columns on the decimal point.
• Bold the best result per column using \textbf{}.

Equations

• Number all equations that are referenced in the text: \begin{equation}...\end{equation}
• Reference as "Eq. (1)" or simply "(1)."
• Define all variables immediately after introduction.

---

Writing Style Rules

• Tense: Present tense for the proposed method and general claims; past tense for experimental observations.
• Voice: Prefer active voice. Use "We propose…" not "A method is proposed…"
• Precision: Every claim must be (a) proven in the paper or (b) supported by a citation.
• Hedging: Replace vague hedges ("somewhat," "rather," "quite") with quantified statements.
• Acronyms: Define on first use in the abstract and again on first use in the body.
• Sentence length: Flag and restructure sentences exceeding 40 words.
• Person: Use "we" even for single-author papers, following IEEE convention.
• Numbers: Spell out numbers one through nine; use numerals for 10 and above, and always for measurements (e.g., "3 layers," "5 epochs").

---

Quality Checklist

Before finalizing the paper, verify every item:

Completeness

• All nine sections (Abstract through References) are present
• Author block includes affiliations and emails
• Index terms are listed
• Every table and figure is referenced in the body text
• All equations are numbered and defined

Technical Rigor

• Hypothesis is falsifiable and tested by the experiments
• Every baseline has a citation and a justification for inclusion
• Metrics are defined and appropriate to the task
• Ablation isolates each component individually
• Limitations are stated honestly and directly

IEEE Style Compliance

• Section headings use Roman numerals and ALL CAPS
• References follow IEEE numbered format
• Table captions are above tables; figure captions are below figures
• Equations are numbered in order of appearance
• Acronyms are defined on first use in abstract and body

Writing Quality

• Abstract is 150–250 words with all five parts
• Introduction contributions list uses verifiable, specific claims
• Related work is organized thematically, not as a flat list
• Main results table has bold best-results and a descriptive caption
• Conclusion does not introduce new claims

---

Output Format

Produce the full paper as a single, continuous Markdown document using LaTeX notation for math and structure cues. Delimit each major section with a horizontal rule and labeled heading. After the full draft:

1. Provide a Word count estimate per section.
2. Flag any data gaps — sections where user input is needed to complete the content.
3. List recommended next steps: additional experiments, missing citations, figures to create.