research-paper-workflow

name	research-paper-workflow
description	Use this skill when writing research papers. Covers paper structure, LaTeX best practices, figure generation, experiment result formatting, and academic writing guidelines.

Research Paper Workflow

This skill provides comprehensive guidance for writing high-quality machine learning research papers.

When to Activate

• Writing a new research paper
• Formatting experiment results
• Creating figures and tables
• Preparing camera-ready submissions
• LaTeX troubleshooting

Paper Structure

Standard ML Paper Outline

\begin{document}

\title{Your Paper Title: A Subtitle if Needed}
\author{Author Names}

\maketitle
\begin{abstract}
% 150-250 words summarizing problem, method, results
\end{abstract}

\section{Introduction}
% Problem, motivation, contributions

\section{Related Work}
% Literature review, positioning

\section{Method}
% Technical approach

\section{Experiments}
% Setup, results, analysis

\section{Conclusion}
% Summary, limitations, future work

\end{document}

Section Guidelines

Abstract (150-250 words)

1. Problem: What problem are you solving? (1-2 sentences)
2. Gap: Why is this hard/unsolved? (1 sentence)
3. Method: What's your approach? (2-3 sentences)
4. Results: Key quantitative results (1-2 sentences)
5. Impact: Why does this matter? (1 sentence)

Introduction

1. Establish the problem and its importance
2. Describe limitations of existing approaches
3. Present your key insight/contribution
4. Summarize your approach and results
5. List contributions (typically 3 bullet points)

Our main contributions are:
\begin{itemize}
    \item We propose X, the first method to...
    \item We introduce Y, a novel technique for...
    \item We demonstrate state-of-the-art results on Z benchmark...
\end{itemize}

Method

• Start with overview/intuition
• Use figures to illustrate architecture
• Define notation clearly
• Present key equations with explanation
• Describe training procedure

Experiments

• Dataset descriptions
• Implementation details
• Main results table
• Ablation studies
• Analysis/visualizations

LaTeX Best Practices

Document Setup

\documentclass{article}

% Essential packages
\usepackage{amsmath,amssymb,amsthm}
\usepackage{graphicx}
\usepackage{booktabs}  % Better tables
\usepackage{hyperref}
\usepackage{cleveref}  % Smart references
\usepackage{algorithm,algorithmic}
\usepackage{subcaption}  % Subfigures

% Custom commands
\newcommand{\method}{\textsc{MethodName}}
\newcommand{\eg}{\emph{e.g.}}
\newcommand{\ie}{\emph{i.e.}}
\newcommand{\etal}{\emph{et al.}}

% Math operators
\DeclareMathOperator*{\argmax}{arg\,max}
\DeclareMathOperator*{\argmin}{arg\,min}
\DeclareMathOperator{\softmax}{softmax}

Tables

\begin{table}[t]
\centering
\caption{Comparison with state-of-the-art methods on ImageNet.
Best results are \textbf{bold}, second best are \underline{underlined}.}
\label{tab:main_results}
\begin{tabular}{lccc}
\toprule
Method & Top-1 Acc & Top-5 Acc & Params \\
\midrule
ResNet-50 & 76.1 & 92.9 & 25M \\
ViT-B/16 & 77.9 & 93.9 & 86M \\
\midrule
\textbf{Ours} & \textbf{79.2} & \textbf{94.5} & \underline{45M} \\
\bottomrule
\end{tabular}
\end{table}

Figures

\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{figures/architecture.pdf}
\caption{Overview of our proposed architecture.
(a) The encoder processes input images.
(b) The decoder generates output.}
\label{fig:architecture}
\end{figure}

% Subfigures
\begin{figure}[t]
\centering
\begin{subfigure}[b]{0.48\linewidth}
    \includegraphics[width=\linewidth]{fig_a.pdf}
    \caption{Training loss}
\end{subfigure}
\hfill
\begin{subfigure}[b]{0.48\linewidth}
    \includegraphics[width=\linewidth]{fig_b.pdf}
    \caption{Validation accuracy}
\end{subfigure}
\caption{Training curves for our method.}
\label{fig:training}
\end{figure}

Equations

% Numbered equation
\begin{equation}
\mathcal{L} = \mathbb{E}_{x \sim p_{\text{data}}} \left[
    -\log p_\theta(x)
\right]
\label{eq:loss}
\end{equation}

% Aligned equations
\begin{align}
q_\phi(z|x) &= \mathcal{N}(z; \mu_\phi(x), \sigma_\phi^2(x)) \\
p_\theta(x|z) &= \mathcal{N}(x; \mu_\theta(z), \sigma^2 I)
\end{align}

% Inline math
The loss function $\mathcal{L}$ is minimized using Adam optimizer.

Algorithms

\begin{algorithm}[t]
\caption{Training procedure}
\label{alg:training}
\begin{algorithmic}[1]
\REQUIRE Dataset $\mathcal{D}$, learning rate $\eta$
\ENSURE Trained model parameters $\theta$
\STATE Initialize parameters $\theta$
\FOR{$t = 1$ to $T$}
    \STATE Sample minibatch $\{x_i\}_{i=1}^B \sim \mathcal{D}$
    \STATE Compute loss $\mathcal{L} = \frac{1}{B}\sum_{i=1}^B \ell(x_i; \theta)$
    \STATE Update $\theta \leftarrow \theta - \eta \nabla_\theta \mathcal{L}$
\ENDFOR
\RETURN $\theta$
\end{algorithmic}
\end{algorithm}

Figure Generation

Matplotlib Style

import matplotlib.pyplot as plt
import matplotlib as mpl

# Publication-quality settings
plt.rcParams.update({
    'font.size': 10,
    'font.family': 'serif',
    'axes.labelsize': 10,
    'axes.titlesize': 10,
    'legend.fontsize': 8,
    'xtick.labelsize': 8,
    'ytick.labelsize': 8,
    'figure.figsize': (3.5, 2.5),  # Single column
    'figure.dpi': 300,
    'savefig.dpi': 300,
    'savefig.bbox': 'tight',
    'savefig.pad_inches': 0.02,
})

# Create figure
fig, ax = plt.subplots()
ax.plot(x, y, label='Ours', color='#1f77b4', linewidth=1.5)
ax.plot(x, y2, label='Baseline', color='#ff7f0e', linestyle='--')
ax.set_xlabel('Epochs')
ax.set_ylabel('Accuracy (%)')
ax.legend(frameon=False)
ax.grid(True, alpha=0.3)

# Save as PDF for LaTeX
plt.savefig('figures/training_curve.pdf')

Comparison Bar Chart

def create_comparison_chart(methods, metrics, values):
    """Create grouped bar chart for method comparison."""
    x = np.arange(len(metrics))
    width = 0.8 / len(methods)

    fig, ax = plt.subplots(figsize=(6, 3))

    for i, (method, vals) in enumerate(zip(methods, values)):
        offset = (i - len(methods)/2 + 0.5) * width
        bars = ax.bar(x + offset, vals, width, label=method)

    ax.set_ylabel('Score')
    ax.set_xticks(x)
    ax.set_xticklabels(metrics)
    ax.legend(loc='upper right', frameon=False)
    ax.set_ylim(0, 100)

    plt.savefig('figures/comparison.pdf')

Architecture Diagram

# Using matplotlib for simple diagrams
from matplotlib.patches import FancyBboxPatch, FancyArrowPatch

fig, ax = plt.subplots(figsize=(8, 3))

# Draw boxes
boxes = [
    ('Input', 0.1, 0.4),
    ('Encoder', 0.3, 0.4),
    ('Transformer', 0.5, 0.4),
    ('Decoder', 0.7, 0.4),
    ('Output', 0.9, 0.4),
]

for name, x, y in boxes:
    box = FancyBboxPatch((x-0.08, y-0.15), 0.16, 0.3,
                         boxstyle="round,pad=0.02",
                         facecolor='lightblue',
                         edgecolor='black')
    ax.add_patch(box)
    ax.text(x, y, name, ha='center', va='center', fontsize=9)

# Draw arrows
for i in range(len(boxes)-1):
    ax.annotate('', xy=(boxes[i+1][1]-0.08, boxes[i+1][2]),
                xytext=(boxes[i][1]+0.08, boxes[i][2]),
                arrowprops=dict(arrowstyle='->', color='black'))

ax.set_xlim(0, 1)
ax.set_ylim(0, 1)
ax.axis('off')
plt.savefig('figures/architecture.pdf')

Results Formatting

Main Results Table Generator

def generate_latex_table(results: dict, baseline_method: str = None):
    """Generate LaTeX table from results dictionary."""
    methods = list(results.keys())
    metrics = list(results[methods[0]].keys())

    # Find best values for each metric
    best_vals = {}
    second_best = {}
    for metric in metrics:
        vals = [(m, results[m][metric]) for m in methods]
        sorted_vals = sorted(vals, key=lambda x: x[1], reverse=True)
        best_vals[metric] = sorted_vals[0][0]
        second_best[metric] = sorted_vals[1][0] if len(sorted_vals) > 1 else None

    # Generate table
    lines = [
        "\\begin{table}[t]",
        "\\centering",
        "\\caption{Main results. Best in \\textbf{bold}, second best \\underline{underlined}.}",
        "\\label{tab:main}",
        "\\begin{tabular}{l" + "c" * len(metrics) + "}",
        "\\toprule",
        "Method & " + " & ".join(metrics) + " \\\\",
        "\\midrule",
    ]

    for method in methods:
        row = [method]
        for metric in metrics:
            val = results[method][metric]
            if method == best_vals[metric]:
                row.append(f"\\textbf{{{val:.1f}}}")
            elif method == second_best[metric]:
                row.append(f"\\underline{{{val:.1f}}}")
            else:
                row.append(f"{val:.1f}")
        lines.append(" & ".join(row) + " \\\\")

    lines.extend([
        "\\bottomrule",
        "\\end{tabular}",
        "\\end{table}",
    ])

    return "\n".join(lines)

Ablation Table

\begin{table}[t]
\centering
\caption{Ablation study on model components.}
\label{tab:ablation}
\begin{tabular}{ccc|cc}
\toprule
Component A & Component B & Component C & Metric 1 & Metric 2 \\
\midrule
\checkmark & & & 75.2 & 82.1 \\
\checkmark & \checkmark & & 78.5 & 85.3 \\
\checkmark & \checkmark & \checkmark & \textbf{81.2} & \textbf{88.7} \\
\bottomrule
\end{tabular}
\end{table}

Writing Guidelines

Clear Writing Principles

1. Active voice: "We propose X" not "X is proposed"
2. Be specific: "improves by 5%" not "significantly improves"
3. One idea per sentence: Break complex sentences
4. Define acronyms: First use should be spelled out
5. Consistent terminology: Pick one term and stick with it

Common Phrases

Problem statements:
- "A fundamental challenge in X is..."
- "Despite recent progress, existing methods struggle to..."
- "However, this approach has limitations..."

Method introduction:
- "To address this, we propose..."
- "Our key insight is that..."
- "The main idea behind our approach is..."

Results:
- "Our method achieves state-of-the-art performance on..."
- "We observe that X outperforms Y by Z%..."
- "These results demonstrate that..."

Ablations:
- "To understand the contribution of each component..."
- "We ablate X and find that..."
- "Removing X leads to a Y% drop in performance..."

Submission Checklist

Before Submission

• Abstract within word limit
• All figures are vector (PDF) or high-res
• Tables fit within margins
• No overfull hboxes
• All references complete
• Supplementary materials organized
• Anonymous for blind review (if required)

Camera Ready

• Author information added
• Acknowledgments added
• Copyright form signed
• Page limit respected
• Correct format/template used

File Organization

paper/
├── main.tex           # Main document
├── sections/
│   ├── intro.tex
│   ├── related.tex
│   ├── method.tex
│   ├── experiments.tex
│   └── conclusion.tex
├── figures/
│   ├── architecture.pdf
│   ├── results.pdf
│   └── ...
├── tables/
│   └── results.tex
├── supplementary.tex
├── references.bib
└── Makefile