Line-by-line video dissections of deep learning’s greatest hits
Because skimming the abstract is not the same as understanding the architecture decisions.

What it does This repository catalogs video lectures that walk through deep learning papers—both canonical and recent—paragraph by paragraph. The README is essentially a broadcast schedule: each row links to Bilibili and YouTube recordings, lists duration, and shows a cover image, covering everything from ResNet to OpenAI Sora. A side series on research communication from The Craft of Research rounds out the syllabus.
The interesting bit Most paper summaries are speed runs; these are slow runs. The 33k stars suggest developers will happily outsource the grunt work of close reading to a patient narrator. It turns the lonely slog through a 40-page PDF into communal, appointment-viewing homework.
Key highlights
- Spanning history and hype: entries range from ResNet and the original Transformer to Llama 3.1 and Sora, with multi-part deep dives for meatier works.
- Dual-platform distribution: every session links to both Bilibili and YouTube, complete with live view-count badges.
- Not just models: includes a multi-episode series on academic writing and argumentation drawn from The Craft of Research.
- Serious runtime: most videos clock in at 45–90 minutes, indicating the author is not racing to the conclusion.
Verdict Bookmark it if your “read later” pile is mostly wishful thinking; look elsewhere if you need code, checkpoints, or quick API recipes.
Frequently asked
- What is mli/paper-reading?
- Because skimming the abstract is not the same as understanding the architecture decisions.
- Is paper-reading open source?
- Yes — mli/paper-reading is open source, released under the Apache-2.0 license.
- How popular is paper-reading?
- mli/paper-reading has 33.6k stars on GitHub.
- Where can I find paper-reading?
- mli/paper-reading is on GitHub at https://github.com/mli/paper-reading.