# ML-University is four files with no licence and a site config for a site that is not there > A curated list of free machine learning courses across eighteen sections, maintained by one person's pull requests. The list says it is continuously updated, several entries are pinned to specific cohort years, and two of the links go through a URL shortener. **d0r1h/ML-University** — Machine Learning Open Source University - Repository: https://github.com/d0r1h/ML-University - Stars: 955 · Forks: 124 - Language: Unknown - License: not declared - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/d0r1h-ml-university ## Four files, no licence, and a config for a site that is not there The entire repository is four entries: a contribution guide, the readme itself, a site configuration file, and one image file used as a logo. That configuration file is the interesting one. A file with that name and location is how a repository publishes a generated site, and its presence alongside a logo image says the project was set up to be a website. There is no index document, no templates directory and no theme, so nothing in the visible tree renders. What a reader gets is the readme, on the repository page, as markdown. The language of the repository is reported as unknown, which is correct and slightly telling: there is no code, only prose and links. For a project that calls itself a university, the fact that nothing is executable is the point rather than a gap. What is missing is a licence. There is no licence file among those four entries and no licence name recorded for the project. For a personal project that is an omission, and for a curated list that other people may want to fork, mirror or extend, it is the one file that would settle the question. The contribution guide exists, so the intent to receive contributions is clear; what those contributions are licensed under is not stated anywhere in the tree. ## Eighteen sections, and the labels have started to drift The table of contents has eighteen numbered entries, running from getting started through mathematics, machine learning, deep learning, natural language processing, reinforcement learning, large language models, books, production, quantum machine learning, datasets, useful websites, useful repositories, blogs and webinars, must-read papers, company engineering blogs, practice projects, and generative and agentic AI. That is a sensible skeleton. A few labels have drifted from their anchors, which is the predictable decay of a hand-maintained table of this size. The large language model entry is labelled with the abbreviation while its anchor uses the long form, so the two will not match. The entry for useful repositories is labelled with a duplicated letter while its anchor is spelled correctly, which means that one link silently does nothing on a case-sensitive markdown renderer. The natural language processing label is lowercased while every neighbouring label is title cased. None of these is serious on its own. The reason to mention them is the maintenance cost they imply. A list this size needs a generated table of contents or a link checker, and the repository has neither a script directory nor a continuous integration file, so the only mechanism keeping the structure honest is a person rereading it. The same applies to the eighteen anchors themselves, which have to be kept in step with eighteen headings by hand. ## The list is maintained and the links are not The readme states the list is continuously updated, and it asks practitioners who have suggestions or resources to open a pull request. That claim is about additions, and it is probably true. The links tell a different story, because many entries name the specific cohort they were taken from. The Harvard introduction to artificial intelligence points at its 2020 site. One of the machine learning courses points at a spring 2020 syllabus, and an applied machine learning course says 2020 in its title. A systems design course says 2021, as does a widely linked introductory course post. A linear algebra course carries a 2019 term, a matrix methods course carries a spring 2018 term, and two further courses carry 2015 and 2016 terms in their paths. One of the two introductory coding courses points at a cohort numbered 2018. So the curation is current and the destinations are archival. A course link from 2016 usually still works, since university archives are stable, but a video playlist from a 2020 season may have been reorganised since, and a blog post used as the primary path into a course is a single point of failure that nobody is watching. The honest way to read the list is as a set of recommendations that were true when they were made. That is not a criticism of a volunteer list; it is a description of what a list without link checking can promise. ## Two of the course links go through a URL shortener Almost every entry in the list points at a domain you can read: a university course site, a course platform, a channel, a channel playlist. Two entries point at a shortened address instead, one for an introductory machine learning course and one for a healthcare-focused course. A shortened link in a resource list is a specific kind of problem, and it is not the same as an old link. An old link fails visibly. A shortened link succeeds and can succeed indefinitely while pointing somewhere other than where it pointed when it was added. The value of a curated list is entirely in the destinations, so a link whose destination is not visible in the file is a link nobody can audit without following it. The rest of the list makes the standard easy to see: each row names the title and, where relevant, the institution, so a reader can check whether a link still belongs to what the row claims. Two rows out of a list this size is a small number, and the fix is to replace them with the real addresses. It is also the kind of change a contributor would make in one pull request, which is a useful thing to know about a project that asks for pull requests. ## Getting started is two-part courses, one lecture series and an algorithms class The first section is nine entries and it is the one with the most concrete structure, which makes it the most useful part of the list to read as a template. One course is split across two parts and two different domains, with the first part hosted one way and the second part on a subdomain of a separate site, so following it end to end means two sites. The Harvard introduction to artificial intelligence is a single entry with a year in its path. The MIT computational thinking and data science course carries a term in its path. There is a practical data ethics course hosted by a well known teaching site, an introductory course from a commercial course blog, and a university algorithms course that predates most of this by a decade. Two of the nine are video: a university principles and techniques course as a channel playlist, and a series of lectures on private AI hosted by a privacy-focused project. That last one is the only entry in the whole visible list about privacy-preserving machine learning, which is a reminder that a list of eighteen sections is a set of editorial choices as much as a set of links. The pattern to copy, if you are building your own path, is to take one of these as a primary course and one of the video entries as a second, rather than working through all nine. ## The mathematics section is mostly video playlists Ten entries, and the format split is informative. Four are video playlists: a statistics-in-machine-learning series by an individual instructor, a linear algebra series from a well known visualisation channel, a mathematics-for-machine-learning playlist, and a statistics-for-applications playlist from a university. Four are websites: a university linear algebra course, a university matrix methods course in data analysis and machine learning, a statistics video site, and a visual probability site from a university. One is a fast.ai course repository on computational linear algebra, which is the only entry in this section that is code you can read rather than a page you watch. The last is a mathematics teaching programme rather than a course, which is a useful reminder that the section is not exclusively about machine learning prerequisites. Two of these entries carry a term in the course path, and both are the kind of archive link that keeps working, which is the good case. The two visual ones are the kind that tend to survive better than a course page, since a demonstration of probability is not tied to a cohort. What the section does not include is anything on the calculus or optimisation side, so a reader starting from the fundamentals will find a gap between the linear algebra material and what a first machine learning course asks for. ## Eighteen sections, several institutions, and one curator The list draws on named institutions throughout: two universities in the mathematics section, a separate pair in the starting section, and in the machine learning section alone a course from one large university on introductory machine learning, one on systems design, one on graphs, and another hosted by a different institution in the same field, plus a reinforcement learning course from a large public university and a probabilistic machine learning course from a European university. So the sourcing is strong on universities and on the community course platforms that grew out of the fast.ai ecosystem, with a commercial crash course, a free code school course and a video channel library filling out the rest. The governance is the other half. The readme describes the project as an idea for free learning from one enthusiast to other enthusiasts, and it asks practitioners to contribute by pull request. There is a contribution guide, so the process is not entirely undocumented. But there is no review policy, no syllabus, no assessment, no tracking of which entries have been opened, and no indication that anyone checks whether a recommended course is still taught. That is a legitimate shape for a link list, and it is a shape that should be read with the same expectations as a bookmark file: useful for finding things, not a substitute for deciding what to learn. ## Conclusion Use this list as a map rather than a curriculum, because that is what it is: one person's index of other people's courses, with a table of contents you can scan in two minutes and a set of links you will have to verify. Three things to know before you plan a study path from it. The maintenance claim is about additions rather than link health, since several entries name the cohort they were taken from and two route through a URL shortener whose destination you cannot see. The courses are mostly well chosen and mostly old, so treat a listed course as evidence that a topic is worth learning rather than as the current way to learn it. And with no licence file present, ask before you fork or republish it, since a resource list is exactly the kind of thing people want to mirror without asking. ## FAQ ### What is the ML-University repository? A curated list of free machine learning courses organised into eighteen numbered sections, from getting started and mathematics through machine learning, deep learning, natural language processing, reinforcement learning, large language models, books, production, quantum machine learning, datasets, websites, repositories, blogs, papers, company engineering blogs, practice projects, and generative and agentic AI. ### Does the ML-University list get updated? The readme states that the list is continuously updated and asks practitioners to open a pull request with suggestions or resources to share. The last push to the repository was 2026-05-23 and there are no published releases, so additions happen at commit level rather than at version level. ### Which universities' courses appear in the ML-University list? Several are named across the sections, including Harvard, MIT, Stanford, Cornell Tech, Carnegie Mellon, UC Berkeley, Brown and the University of Tübingen, alongside community courses from the fast.ai ecosystem, a commercial course blog, a crash course from a large platform and a free code school. Many entries name the cohort year they were taken from. ### Is the ML-University list licensed? Nothing states a licence. The repository has four top-level entries, none of which is a licence file, and no licence name is recorded for the project. A contribution guide is present, so contributions are invited, but the terms for forking or republishing the list are not stated. ### How current are the course links in ML-University? The list is current but many destinations are archival, since entries name specific cohorts and terms including 2015, 2016, 2018, 2019, 2020 and 2021 seasons. Two entries are routed through a URL shortener rather than a direct address, and nothing in the repository indicates that links are checked. ## Sources - [d0r1h/ML-University on GitHub](https://github.com/d0r1h/ML-University) - [Issues](https://github.com/d0r1h/ML-University/issues) - [README](https://github.com/d0r1h/ML-University/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/d0r1h-ml-university