Papers I read

Notes and open discussion on the papers I read, covering generative models, causality, explainability and medical imaging.

Live site: https://naughtfound.github.io/papers/

Each paper has a short write-up, and anyone can join the discussion by signing in with GitHub. Comments are stored in this repo's Discussions.

How it works

• index.html is the whole site. It is a single static page with no build step.
• papers/index.json lists every paper and its details.
• papers/<slug>.md holds the notes for each paper, written in Markdown.
• Comments use giscus, which maps each paper to one GitHub Discussion thread.
• giscus/light.css and giscus/dark.css style the comment box to match the site's light and dark themes.
• .github/workflows/deploy.yml checks the JSON and publishes the site to GitHub Pages whenever you push to main.

Add a paper

1. Write your notes in papers/<slug>.md. Start at the ## headings, because the page already shows the title, authors, links, summary and takeaways.

2. Add an entry to papers/index.json:

   {
     "slug": "ddpm",
     "title": "Denoising Diffusion Probabilistic Models",
     "authors": "Jonathan Ho, Ajay Jain, Pieter Abbeel",
     "year": 2020,
     "venue": "NeurIPS",
     "arxiv": "2006.11239",
     "code": "https://github.com/hojonathanho/diffusion",
     "tags": ["diffusion", "generative"],
     "added": "2026-09-28",
     "summary": "One or two sentences describing the paper.",
     "takeaways": ["First point.", "Second point."],
     "bodyUrl": "papers/ddpm.md"
   }

3. Commit and push to main. The site updates in a minute or two.

Required fields: slug, title, authors, year, summary. The rest are optional.

Keep each slug unique and don't rename it later, because the discussion thread is tied to it.

Set up your own copy

1. Create a repo named papers and add these files.
2. In Settings, then Pages, set the source to GitHub Actions.
3. In Settings, then General, turn on Discussions.
4. Install the giscus app on the repo, then get your repoId and categoryId from giscus.app.
5. Put the repo name and IDs in the CONFIG block at the top of the script in index.html.

Preview locally

The page loads its data with fetch, so open it through a local server rather than double-clicking the file:

python3 -m http.server 8000

Then visit http://localhost:8000. Comments only appear on the live site, where giscus can reach the repo.