csinva.io: ML Research Notes, Slides, and Cheat Sheets from a Microsoft Researcher
Slides, paper notes, class notes, blog posts, and research on ML 📉, statistics 📊, and AI 🤖.
At a glance
- What is it?
- csinva.github.io is the source repository for csinva.io, a personal research website maintained by Chandan Singh, a Senior Researcher at Microsoft Research, covering machine learning, statistics, and interpretability through notes, slides, research overviews, and blog posts.
- Who is it for?
- csinva.io is worth bookmarking for researchers and students who work in interpretable ML, causal inference, or adjacent areas and want a curated second opinion alongside primary papers. The content is most useful as a supplement to courses and papers, not as a standalone reference.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 3 days ago.
- What is it written in?
- Mainly HTML, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 29, 2026, and from our analysis. They are not legal advice.
Editorial analysis
What This Repository Contains and Who Maintains It
csinva.github.io is the source for a personal research website built by Chandan Singh, described in the README as a Senior Researcher at Microsoft Research who has worked on interpretable machine learning since his PhD at UC Berkeley. The README frames the site as a product of compulsive note-taking that started during his PhD and has continued since.
The repository contains the full source for csinva.io, which serves five categories of content: presentation slides, research overviews, cheat sheets, class and topic notes, and blog posts. The content covers machine learning, statistics, and neuroscience, with a concentration on interpretability.
This is not an open-source software project. There is no package to install, no API to call, and no library to import. The repository is a Jekyll-based static site with markdown source files. Its value is the content: written explanations, annotated summaries of research areas, and slides from courses taught at Berkeley.
How the Repository Is Organized
The top-level structure separates content by type. The `pres/` directory contains slide decks, including ML slides from Berkeley CS 189 and AI slides from Berkeley CS 188. Slides are built with reveal-md from markdown source and can be edited or exported.
The `_notes/` directory is the core of the site. It contains:
- `_notes/research_ovws/`: overviews and paper summaries for research areas including interpretability, causal inference, transfer learning, uncertainty, deep learning theory, complexity, scattering transforms, and deep learning for neuroscience. - `_notes/cheat_sheets/`: condensed reference sheets covering topics such as interpretability and classification. - Topic notes on computational neuroscience, causal inference, linear algebra, information theory, and computer vision.
The `_blog/` directory holds blog posts on ML and statistics topics. Selected posts listed in the README include paper writing tips, a 2022 post on imodels on the BAIR Blog, and a GPT paper title forecasting tool.
The site is built with Jekyll, hosted on GitHub Pages, and uses the timeline theme with particles.js for the landing page.
Research Overviews: What They Cover and How to Use Them
The `_notes/research_ovws/` files are the most distinctive part of the repository. Each is a long markdown document covering a research area: `ovw_interp.md` covers interpretability, `ovw_causal_inference.html` covers causal inference, and others cover transfer learning, uncertainty, deep learning theory, complexity, scattering transforms, and deep learning in neuroscience.
These files are summaries and annotations of papers in each area, not original research. Their purpose is to give a reader a structured entry point into an area, with links to primary sources. This is most useful when entering an unfamiliar sub-area adjacent to one you know well, or when reviewing what was covered in a research area before a specific date.
The interpretability overview (`ovw_interp.md`) is the most directly connected to the author's own research at Microsoft Research. Teams working on interpretable ML who want to see how one active researcher in the area has organized and annotated the literature will find it a useful secondary source. It is not a survey paper and has not been peer-reviewed.
Slides from Berkeley CS 189 and CS 188
The `pres/` directory contains two complete course slide decks. CS 189 is Berkeley's machine learning course; CS 188 covers AI. The slides are built from markdown using reveal-md, which means the source is editable and the output can be exported to HTML or PDF.
The CS 189 slides at `https://csinva.io/pres/189/#/` cover standard machine learning topics at course depth. The CS 188 slides at `https://csinva.io/pres/188/#/` cover AI topics. Both are publicly accessible through the site.
For educators or students looking for a Berkeley-style ML course structure in slide format, these are a concrete resource. The reveal-md build process is documented in a blog post at `https://csinva.io/blog/misc/reveal_md_enhanced/readme`. The slides can be forked and adapted since the repository is MIT-licensed.
The Interpretability Cheat Sheet and Topic Notes
The `_notes/cheat_sheets/interp.svg` is an interpretability cheat sheet in SVG format. SVG at this URL renders directly in a browser, which makes it usable as a quick visual reference without any additional tools.
Other cheat sheets and notes cover linear algebra, information theory, classification, computer vision, causal inference, and computational neuroscience. The README notes that there are many more notes beyond the selected list. The full set is browsable at csinva.io.
These notes are self-contained per topic but vary in depth. Some are dense with equations and citations; others are briefer orientation documents. The most developed notes appear to be in the interpretability and causal inference areas, consistent with the author's research focus.
Limitations and What This Repository Is Not
The repository is a personal website, not a collaborative knowledge base. There is no contribution process described in the README, no issue tracker for content requests, and no review process for notes. Content accuracy depends on one person's judgment and time.
Note files can fall behind current research in fast-moving areas. The repository was last pushed on 2026-09-26, and the overall repository is active, but individual files may not have been updated in months or years. A note on deep learning theory from 2022 may not reflect developments since then.
The imodels package mentioned in a 2022 BAIR Blog post is a separate repository, not included in this one. The blog post link appears as a reference in the README, but the code is elsewhere. Readers who find the blog post interesting should look up the imodels repository separately.
The site uses particles.js and a timeline theme for its visual presentation. Readers accessing content through the GitHub repository rather than the rendered website will see raw markdown, which is readable but loses the navigation structure.
Using the Repository Without Visiting the Website
The repository is MIT-licensed, which means its content can be forked, adapted, and used with attribution. Educators who want to adapt the CS 189 slides for their own courses can clone the repository and modify the reveal-md source. Researchers who want to use the research overview files as a starting point for their own literature notes can copy individual `_notes/` files.
Cloning the full repository for local browsing is practical since it is a static site. Running `jekyll serve` locally with the correct Ruby and Jekyll version would render the full site, but the README does not document a local development setup. The raw markdown files in `_notes/` and `_blog/` are readable directly without building the site.
For updates, the README suggests starring the repository or following the author on Twitter at `@csinva`. The last push to this repository was on 2026-09-26. The project is MIT-licensed.
Editorial conclusion
csinva.io is worth bookmarking for researchers and students who work in interpretable ML, causal inference, or adjacent areas and want a curated second opinion alongside primary papers. The content is most useful as a supplement to courses and papers, not as a standalone reference. Before relying on any specific note, check the last commit date for that file: the repository as a whole is active, but individual note files may not have been updated to reflect recent developments in fast-moving areas like large model interpretability.
Frequently asked questions
Who maintains csinva.io and what is their background?
The site is maintained by Chandan Singh, described in the README as a Senior Researcher at Microsoft Research who has worked on interpretable machine learning since his PhD at UC Berkeley. The notes have been compiled since his PhD.
What machine learning topics does csinva.io cover?
The site covers interpretable ML, causal inference, transfer learning, uncertainty, deep learning theory, complexity, scattering transforms, deep learning in neuroscience, linear algebra, information theory, computer vision, and computational neuroscience, primarily through research overviews and class notes.
Can the Berkeley CS 189 slides from csinva.io be reused?
The repository is MIT-licensed, so the slides can be forked and adapted with attribution. The slides are built from markdown using reveal-md and can be exported. The README links to a blog post explaining the reveal-md build process.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/csinva-csinva-github-io)