Read an arXiv Paper with Markdown and a Mind Map
Find the research question, then follow the sections, formulas, and figures. Keep your questions and observations so the next reading session has a clear starting point.
A paper rarely settles into place after one read. The abstract leads you to the method; a figure sends you back to the experimental conditions. A conclusion may reveal an assumption you want to revisit.
DeMinds brings supported arXiv HTML papers into Markdown and Universal Mind Map. The structure helps you navigate, the text carries the detail, and the document gives you somewhere to develop your own reading notes.
Bring the paper into your reading workflow
Start with arXiv HTML
Check whether the paper's arXiv page offers an HTML version, then bring that page into DeMinds through the article-import workflow.
Open the arXiv HTML page in DeMinds, then choose Import HTML Paper.
You can also replace arxiv.org with xiv.deminds.app in a supported paper link to hand the address to the app. For example:
If your browser stays on an instruction page, follow its app-opening guidance or return to DeMinds and import the paper's HTML URL. HTML availability and structure vary between papers. This workflow handles supported HTML, rather than arbitrary PDF or TeX source files.
The shortcut hands the paper to DeMinds; the imported structure becomes a navigable map.
Check that the important material arrived
Review the title, authors, abstract, and section order. Inspect a formula, an important figure, and a table. Supported imports can preserve recoverable formulas, figures with captions, and recognizable internal links. Recognized complex tables can retain relationships such as merged cells and multi-level headings.
If something is missing or looks wrong, compare it with the original page. A clear map helps you read; the paper remains the reference for checking the content.
Follow a question between the structure and the text
Locate the part you need to understand
Open Universal Mind Map, look at the main sections, and unfold a branch relevant to your question. To judge whether a method fits your work, look for its assumptions and evaluation conditions. To understand a result, follow it back into the method and experiments.
Expand long branches as needed and read the detail in the document or available notes. Finding what your current question needs is often more useful than opening every node at once.
Practice with a real paper
In version 7 of Attention Is All You Need, locate Figure 1, then read the attention formula in section 3.2.1, and finally visit the training and results sections.
Use one question to guide your notes: which passages explain how the model works, and which explain the conditions under which it was evaluated? These locations give you a reading path and checkpoints for comparing the imported result with its source.
Switch to Markdown Preview for the argument itself. Follow available section, figure, table, and reference links, then return to the map when you need to get your bearings.
Keep the question and its evidence on screen together
On a Mac or iPad with enough document space, choose Open Side-by-Side from Markdown Preview. The text stays on the left and Universal Mind Map appears on the right. This works well when reading methods or results while keeping your place in the paper's structure. If you begin from the map instead, the map stays on the left and the text opens on the right.
Side-by-side reading keeps the formula, the text, and the paper's structure in view together.
Suppose you want to know whether a reported improvement applies to your own task. Locate the results, read the experimental setup, then follow valid internal links to the model, table, or references. If the map has no separate node for a target but the body retains that location, navigation can open it in Markdown Preview. Back and Forward let you retrace your reading path without repeatedly hunting for the previous passage.
Not every detail needs its own map node. Use the map to locate a question and the text or notes to read the argument. Write your judgment in Markdown, save, then check the structure of the added notes. Saving updates the map to include the structure of your notes.
Check relationships in formulas and complex tables
A formula can look complete while a subscript, symbol definition, or surrounding explanation has become disconnected. Tables also carry more than numbers. A spanning header may identify the task associated with a metric; a merged row label may establish a condition shared by several results. Without those relationships, even a correct number can be misread.
DeMinds aims to retain supported formulas and complex table structures, but paper HTML varies, as does the content other Markdown tools can display. For a table with grouped headers, compare the column groups in Preview with the original paper, then inspect the export. Do not simplify the table into an unexplained list of numbers at the cost of its meaning.
Choose between reading help and a translation to keep
Understand the current content temporarily
Where the system supports it, Preview Translation can help you read Markdown Preview and rendered map content. It does not change the saved Markdown or automatically translate your exports.
On iPhone and iPad, this requires iOS or iPadOS 18 or later and supported system translation languages.
Keep a translation you can develop
Use Continue with AI… and choose Translate as well when you want a lasting translated document. Retrieve the file, compare it, and decide whether to apply it. You can also save it as a separate document to maintain alongside the original.
Pay particular attention to terminology, formulas, experimental conditions, and qualifications such as “may” or “under these conditions.” If you ask an AI for a plain-language explanation, keep it clearly separate from the authors' text.
Leave a record of your own understanding
Give your notes a clear place
Add a section called “My reading notes” at the end of the document, separate from the authors' work. Four things are enough to begin:
- Reading question: what am I trying to understand?
- Evidence to revisit: which section, figure, or table matters?
- Limits and open questions: what was not covered, and what still needs checking?
- Next step: read further, compare another paper, or try a specific method?
You do not need a finished review after one session. A source location and a useful question are often enough to make the next session easier.
Turn “I read this” into a question you can pursue
Instead of writing only “This method works well,” record something like: “The paper reports an improvement on these tasks under the stated experimental setup. I still need to check whether my data meets comparable conditions.” Add the relevant section or table location. The note preserves both the result and the clues needed to assess its scope.
If you add an AI-assisted introduction, give it a clearly labeled section, identify it as reading assistance, and retain questions that need verification. Readers should be able to distinguish the authors' findings, an AI's explanation, and your own inference. A fluent overview should not overwrite the paper's original caution.
Save, return, and take it further
Save the document and return through Continue Working. Choose Download Standard Bundle when you want the Markdown and available local resources together, then check formulas and complex tables in the destination tool.
For public discussion, you can link to the original paper and share clearly labeled notes of your own. Before redistributing an Interactive View containing paper content, confirm the necessary permission.
What you keep is more than a paper you once opened. It records where you got to, what you understood, and the question you want to pursue next.