Research Paper Writing

# Ubuntu/Debian
sudo apt install texlive-full

# macOS
brew install --cask mactex

# Windows
# Install MiKTeX from miktex.org

# Verify
pdflatex --version
bibtex --version

Compile

pdflatex paper.tex
bibtex paper
pdflatex paper.tex
pdflatex paper.tex   # run 3x for refs to resolve

# Or with latexmk (auto-detects what to run)
latexmk -pdf paper.tex
latexmk -pdf -pvc paper.tex  # continuous preview

Minimal LaTeX Paper

\documentclass{article}
\usepackage{amsmath, amssymb, graphicx, booktabs, hyperref}
\usepackage[margin=1in]{geometry}
\usepackage[numbers]{natbib}

\title{Your Paper Title}
\author{Author One \and Author Two}
\date{}

\begin{document}
\maketitle

\begin{abstract}
Your abstract here.
\end{abstract}

\section{Introduction}
\label{sec:intro}

\section{Related Work}
Prior work \citep{vaswani2017attention} showed...

\section{Method}
\label{sec:method}

\section{Experiments}

\begin{table}[t]
\centering
\caption{Main results}
\begin{tabular}{lcc}
\toprule
Method & Metric A & Metric B \\
\midrule
Baseline & 70.1 & 65.3 \\
Ours & \textbf{74.8} & \textbf{70.2} \\
\bottomrule
\end{tabular}
\end{table}

\section{Conclusion}

\bibliography{references}
\bibliographystyle{plainnat}
\end{document}

References (.bib file)

@article{vaswani2017attention,
  title={Attention is all you need},
  author={Vaswani, Ashish and others},
  journal={NeurIPS},
  year={2017}
}

@inproceedings{devlin2019bert,
  title={{BERT}: Pre-training of deep bidirectional transformers},
  author={Devlin, Jacob and others},
  booktitle={NAACL},
  year={2019}
}

Get BibTeX from:

• Google Scholar → Cite → BibTeX
• arXiv → Export BibTeX (bottom of abstract page)
• Semantic Scholar

Literature Review via arXiv

# Use /arxiv skill to find papers
# Example searches:
# /arxiv search "GRPO reinforcement learning language model"
# /arxiv search "retrieval augmented generation survey"
# /arxiv get 2401.00001

Figures

import matplotlib.pyplot as plt
import numpy as np

fig, ax = plt.subplots(figsize=(4, 3))
x = np.linspace(0, 10, 100)
ax.plot(x, np.sin(x), label="Our Method")
ax.plot(x, np.cos(x), "--", label="Baseline")
ax.set_xlabel("Epoch")
ax.set_ylabel("Accuracy")
ax.legend()
ax.grid(alpha=0.3)
plt.tight_layout()
plt.savefig("figure1.pdf", bbox_inches="tight", dpi=300)

Include in LaTeX:

\begin{figure}[t]
\centering
\includegraphics[width=0.8\linewidth]{figure1.pdf}
\caption{Comparison of our method vs baseline.}
\label{fig:main}
\end{figure}

Conference Templates

Download official template from the conference website — never use unofficial ones.

Overleaf (Collaborative)

1. Upload all .tex, .bib, .pdf files to Overleaf
2. Share link with co-authors
3. Enable Track Changes for reviews
4. Download final PDF when ready

Common Mistakes

• Vague claims: "significantly better" → give exact numbers
• Missing baselines: always compare against SOTA
• No ablation: show which parts of your method matter
• Figure too small: aim for readable at 100% zoom
• Passive voice overuse: "we propose" not "it is proposed"
• Missing limitations: reviewers will point them out; address proactively