Tips for writing papers

How to start (proposal)

1. Consider what the main selling point of your paper is (why should people read it?) and consider this in drafting the general structure and abstract.
2. Decide on general structure (see below), and write it down as headlines
3. Under each section, write down what the purpose of the section is (what should the reader take away)
4. Iterate on the general structure
5. Decide which figures are needed, add (early versions) of the figures in the respective sections.
   • Tip: Generate all figures from a single script to make it simple to update figures. in LaTeX, figures are referenced by file name and update automatically if the figure file changes. In Word, this is also possible. Click on the little arrow in the “Select file” dialogue and select “Link and Embed”).
6. Decide which references are needed and add them to the respective sections.
7. Once you are happy with the general structure, add the content in the style of bullet points for each section
8. Review the paper carefully
9. Translate the bullet points into continuous text, section by section.

Templates

• Scientific papers: REVTex (Journal: PRB)
• More technical/engineering papers: IEEE

General structure

Most papers follow the structure below. See also the sections More details on the general structure and Prompt ideas later in this document.

1. Abstract
2. Introduction
   • Problem (including general setting)
   • Challenges
   • Key Idea
   • Contributions
3. Related Work
4. Problem Formulation
5. Method
6. Analysis / Theory (optional)
7. Experiments / Ablation Studies
8. Discussion
9. Conclusion

You can have a look at the following papers to see this in practice: - https://arxiv.org/abs/1404.1712 - https://www.nature.com/articles/s41467-020-17865-3 - https://openreview.net/forum?id=GjQ5JXpRQF - https://arxiv.org/abs/1906.00950

Figures

• Figures should have a descriptive title
• Clear axis labels including units (also for colorscale)
• Use an appropriate colorscale (see https://matplotlib.org/stable/users/explain/colors/colormaps.html)
• Figure caption
  • What is shown in the figure
  • What are the conclusions from this (refer to text if that is too long)
• Reference figures in text, figure order should reflect referencing in text (i.e., Figure 1 should be mentioned first)

Prompt ideas

Idea and positioning - What is the single sentence that explains why someone should read this paper?
- If a reader only remembers one thing from this paper, what should it be?
- What concrete problem does the paper solve, and who cares about it?

After drafting the structure - Does every section have a clear purpose?
- Does the structure logically guide the reader from problem to solution to evidence?
- Could a reader understand the story of the paper just from the section headings?

After adding bullet point content - Does each section answer a clear question?
- Is every claim supported by either reasoning, references, or experiments?
- Are there sections that only repeat information instead of advancing the argument?

Before writing continuous text - Can the entire paper be summarized by reading only the bullet points?
- Does each section clearly lead to the next one?
- Are the main contributions clearly visible?

After adding figures - Does each figure answer a specific question?
- Could a reader understand the figure without reading the entire paper?
- Is every figure explicitly referenced and explained in the text?

Before submission or final review - Is the main contribution clearly stated in the introduction?
- Are the assumptions, limitations, and scope clearly described?
- Would someone outside the immediate research area understand the motivation?

More details on the general structure

Abstract

Typical content

• Very short summary of the entire paper (problem, approach, key results, implications).
• One or two sentences describing the context and importance of the problem.
• A brief description of the proposed method or idea.
• The most important quantitative or qualitative results.
• A short statement about the broader significance or applicability.

Questions

• What is the problem addressed in one or two sentences?
• Why does this problem matter?
• What is the core idea or approach?
• What are the most important results?
• What should the reader remember after reading only the abstract?

Introduction

Problem (including general setting)

Typical content

• Description of the application domain and context.
• Explanation of the concrete problem addressed in the paper.
• Motivation for why solving this problem is important.

Questions

• What exactly is the problem I am addressing?
• In which context or application does this problem occur?
• Why should the reader care about solving this problem?

Challenges

Typical content

• Explanation of why the problem is difficult.
• Limitations of existing approaches.
• Key technical or conceptual obstacles.

Questions

• Why has this problem not been solved satisfactorily before?
• What makes the problem technically or conceptually difficult?
• Which aspects cause existing approaches to fail or perform poorly?

Key Idea

Typical content

• High level description of the main idea behind the proposed approach.
• Intuition behind why the idea works.
• Conceptual explanation without technical details.

Questions

• What is the central idea that enables the solution?
• Why should this idea work?
• How does this idea address the previously mentioned challenges?

Contributions

Typical content

• Clear list of the main contributions of the paper.
• Typically three to five concise bullet points.
• Each contribution describes a concrete novelty (method, analysis, dataset, result, framework).

Questions

• What exactly is new in this work?
• What are the most important outcomes of the paper?
• Which contributions could future work build upon?

Related Work

Typical content

• Overview of existing research related to the problem.
• Grouping of prior work into categories or approaches.
• Discussion of similarities and differences to the current work.

Questions

• Which previous approaches address similar problems?
• How do these approaches differ from each other?
• In what way does my work differ from or improve upon them?

Problem Formulation

Typical content

• Precise and formal definition of the problem.
• Introduction of notation, variables, and assumptions.
• Mathematical description of the task or objective.

Questions

• What exactly is the input and output of the problem?
• Which variables, symbols, or quantities are needed?
• What assumptions are made about the system or data?

Method

Typical content

• Detailed description of the proposed approach.
• Explanation of algorithmic steps, model components, or pipeline stages.
• Diagrams or figures illustrating the method.

Questions

• How does the proposed method work step by step?
• What are the main components of the method?
• Why is each component necessary?

Analysis / Theory (optional)

Typical content

• Theoretical analysis of the method.
• Proofs, guarantees, complexity analysis, or formal properties.
• Explanation of theoretical insights into the behavior of the method.

Questions

• What theoretical properties does the method have?
• Can the performance or correctness be formally justified?
• What insights does the theory provide about the method?

Experiments / Ablation Studies

Typical content

• Description of experimental setup (datasets, metrics, baselines).
• Quantitative evaluation of the method.
• Ablation studies analyzing the contribution of individual components.

Questions

• How is the method evaluated?
• Which baselines are used for comparison?
• Which parts of the method are most important for performance?

Discussion

Typical content

• Interpretation of the results.
• Discussion of strengths, limitations, and possible failure cases.
• Broader implications or potential applications.

Questions

• What do the results mean in practice?
• Where does the method work well and where does it fail?
• What are the limitations of the approach?

Conclusion

Typical content

• Short summary of the main findings.
• Restatement of the main contributions.
• Possible directions for future work.

Questions

• What are the key takeaways of the paper?
• What should the reader remember most?
• What are promising next steps for future research?