# Learning-Deep-Learning is a syllabus on top, a reading log underneath > One engineer's paper notes on computer vision, occupancy and speech-to-action models, kept as a single README over fifteen directories. The front matter addresses a beginner, a peer and the author in turn. Underneath sits a dated reading log whose month headings carry a number the page never explains. **patrick-llgc/Learning-Deep-Learning** — Paper reading notes on Deep Learning and Machine Learning - Repository: https://github.com/patrick-llgc/Learning-Deep-Learning - Stars: 1,277 · Forks: 178 - Language: Jupyter Notebook - License: not declared - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/patrick-llgc-learning-deep-learning ## The same file addresses a beginner, a peer and its own author The front matter changes voice three times without saying so. It opens with a syllabus aimed at someone new to computer vision deep learning, who is told to spend a first month on a list of papers at start/first_cnn_papers.md, with the author's own notes on that list alongside it and a fallback page of trustworthy sources at trusty.md. It then becomes a portfolio: twenty-one posts on a Medium column described as regularly updated, spanning planning, occupancy, bird's-eye-view perception, lane detection, mapping and SLAM, and it names the author as Director of AI at Nvidia leading modelling for an end-to-end autonomous driving project. Then it turns inward, with a scratchpad described as quick notes to the author's future self, holding two entries on compute hardware and attention masks. Under all of that sit the paper lists. Four registers in one file, no table of contents, and a reader who starts at the top is not told which voice answers them. ## The number after each month does not match the list beneath it Each dated section of the paper lists carries a bracketed figure. The section headed 2026-04 is marked with a 1 and holds seven entries. The section headed 2026-02 is also marked with a 1 and holds two. The section headed 2026-01 is marked with a 10. The figure is the only quantity stated anywhere in these lists, and it does not reconcile with them under any obvious reading: it is not the number of bullets, and it is not the number of entries carrying a notes link, which is two under 2026-04 and one under 2026-02. Nothing on the page defines it. So the one number a reader could have used to judge how much effort went into a month is the one number the reader cannot interpret, and the gap is most awkward in the newest section, where a 1 sits above seven carefully tagged papers. ## One bracket field carrying topics and authors at once Each entry ends in a free-text square bracket, and the field means two different things depending on the line. Five consecutive entries under 2026-04 are tagged Async infer, which names the subject. The line above them reads Async infer, Song Han, Zhijian Liu, and the next reads Async infer, Physical Intelligence, which name authors and teams. Elsewhere the bracket holds only a person: Dexmal, Hang Zhao, Qwen team, and Yejin Choi with the note latent CoT. There is no key, so the same punctuation marks a topic on one row and attribution on the next, and a reader who filters on it is filtering on two different things at once. The double-bracket notes link beside it is the cleaner signal, since it says whether the paper was read closely enough to earn a written note. Five of the eleven entries on this page carry one. ## Ten of the fifteen directories have no way in from the page The repository root holds fifteen directories alongside the README, a .gitignore and one loose Markdown file called trusty.md. The linked sections reach into five of them: start, topics, podcast, gist and paper_notes. The other ten are unreachable from anything on this page, including all five directories named after subareas of the field, learning_agents, learning_filters, learning_pnc, learning_robotics and learning_slam, and including openai_orgchart, which is not a learning topic at all and shares the naming scheme of the five that are. So the folder layout groups by field, the page organises by content type, and a reader who wants the perception or robotics notes has to guess which of two layouts holds them. The loose trusty.md is the exception that shows the intent: a page-level list is fine when it sits at the root. ## Classified as a notebook repository and indexed entirely in Markdown GitHub reports the primary language of this repository as Jupyter Notebook. Not one path linked from the page is a notebook. Every destination is a Markdown file: two under start, five under topics, two under gist, one under podcast, and the notes files under paper_notes. The only other entry point mentioned is assets, alongside a website described as a minimalistic page generated with GitHub Pages. That combination sets an expectation the content then declines to meet, and it does so in a way that costs a newcomer time rather than saving it, because a notebook-classified repository is usually the place to look for something runnable. There is code in the layout, in a directory named code_notes, and the page gives it no entry point either. Nothing here is wrong; it just means the language badge is describing something a reader will not reach. ## One entry in eleven says where the paper was published Ten of the eleven visible entries link to arXiv and one links to an OpenReview submission. Exactly one of the eleven carries a peer-review outcome: FlashDrive, tagged ICLR 2026, and it is the same row that breaks the arXiv pattern. For a list whose stated purpose is helping a working engineer decide what to read, that is the field worth most and the one filled least, because a preprint from two years ago and a paper accepted at a conference look identical in every other column on the page. The month headings give dates, the brackets give topics and authors, and the arXiv links give you a paper to open, but nowhere does the page record whether anyone reviewed it. The same gap shapes the posts: twenty-one titles, four of them ending in A Review, which is a promise the page makes about its own writing and never keeps about anyone else's. ## The newest heading is two months behind the newest commit The most recent dated section on the page is 2026-04, and the last push to the master branch is 2026-06-04. Those are two months apart, which means the newest thing in the repository has no heading of its own. There is also no section for 2026-03 anywhere between the 2026-04 and 2026-02 headings, so the sequence steps from April to February. Both gaps are consistent with the page having nothing to file in those months and equally consistent with a heading having been skipped, and nothing on the page distinguishes the two cases. There are no releases either, so no tag to point at and nothing to diff against. A reader who wants to know whether a section changed cannot find out, and a reader who wants to recommend a specific version of these notes has nothing to cite. ## No licence, no version, and one open issue The licence field on the repository is empty and there is no licence file at the root. The notes are written to be used: a month of suggested reading, a page of trustworthy sources, twenty-one essays with links back to the underlying papers. All of that is quotable in principle and carries no stated terms, which is an odd position for material with 1277 stars and 178 forks. The default branch is master, there are no GitHub releases, and the project has one open issue. Nothing on the page describes a route for contributing beyond that issue, and nothing describes what should happen to a correction. For a private reading log none of this would matter. For a syllabus other people are told to follow for their first month, the missing licence is the part that costs something. ## Conclusion Use it as a syllabus in the first month and as an index afterwards, because that is the order the page is written in and the order it works in. The strengths are real: a curated starting list, a stated opinion on which sources to trust when the list runs out, and topic notes written by someone who ships end-to-end driving models. The weaknesses are all about state rather than content. There is no licence on the notes, no release to cite, no version to return to, and month headings whose bracketed numbers mean something the page will not say. Before quoting any of it, find out what reuse is permitted, because the repository does not tell you. Before recommending a paper from it to a colleague, check whether it was peer reviewed, because one entry in eleven says. And if you want to add to it, the route is an issue, since nothing else is documented. ## FAQ ### What is the Learning-Deep-Learning repository for? It holds one engineer's paper reading notes on deep learning and machine learning, arranged as a suggested starting path for newcomers, a set of review posts on computer vision for autonomous driving, and dated lists of papers with notes on some of them. ### Where should someone start with Learning-Deep-Learning? The page suggests spending a first month on the paper list at start/first_cnn_papers.md if you are new to deep learning in computer vision. The author's own notes on that same list sit at start/first_cnn_papers_notes.md, and trusty.md holds a further set of sources. ### How are the papers in Learning-Deep-Learning organised? Into dated sections such as 2026-04, 2026-02 and 2026-01, each headed by a bracketed number. Entries carry a title, a link to arXiv or OpenReview, an optional link to a written note, and a free-text tag that names either a topic such as Async infer or an author. ### Does Learning-Deep-Learning record whether a paper was peer reviewed? Almost never. Of the eleven entries on the page, one is marked with a conference, FlashDrive tagged ICLR 2026, and that same entry is the only one not linking to arXiv. The rest carry no venue information at all. ### Can you reuse notes from Learning-Deep-Learning? The page does not say. The repository has no licence file at its root and its licence field is empty, so the terms for copying or redistributing the notes are not stated anywhere on the project. ## Sources - [Issues](https://github.com/patrick-llgc/Learning-Deep-Learning/issues) - [patrick-llgc/Learning-Deep-Learning on GitHub](https://github.com/patrick-llgc/Learning-Deep-Learning) - [README](https://github.com/patrick-llgc/Learning-Deep-Learning/blob/master/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/patrick-llgc-learning-deep-learning