Papers

1 Why This Page Matters

Papers are usually the highest-resolution resource in the library.

They are also the easiest place to get lost.

The problem is not only difficulty. It is that papers are optimized for contribution, not for onboarding.

This page is here to help readers use papers as a tool rather than as a chaos stream.

2 How To Use Papers Well

Use papers when you want:

• the cleanest statement of a current contribution
• the exact assumptions, theorem, or experimental setup behind a result
• a direct view of how a field is arguing with itself right now

Do not use papers as your first exposure to a topic if a book, course, or note would explain the same objects more cleanly.

3 Good Entry Shelves

Resource	Best for	Use it when
Journal of Machine Learning Research	mature ML papers with more room for detail and exposition	you want papers that are often easier to study slowly than conference versions
Proceedings of Machine Learning Research	conference proceedings for ICML, AISTATS, COLT, and related venues	you want official conference paper archives grouped by venue and volume
NeurIPS Proceedings	broad ML conference papers across many subfields	you want a direct archive for method-heavy and experiment-heavy ML work
OpenReview	active conference and journal-style paper discussion, especially ICLR and TMLR ecosystems	you want frontier work, open reviews, or discussion around current papers
arXiv	preprints and fast-moving research directions	you want the earliest public version of a paper or you are tracking a fast-moving topic

4 Use Papers By Job

4.1 If You Are Learning A Topic

A good pattern is:

1. use the site page to get the object map
2. use a course, note, or book to get structure
3. use one paper to see how the object appears in research

This is usually better than jumping straight into a random recent paper.

4.2 If You Are Reading For Theory

Prefer papers where the mathematical object is easy to name:

• estimator
• optimization method
• dynamical system
• concentration bound
• information-theoretic limit

Journal-style papers and cleaner proceedings papers are often better here than the flashiest recent preprint.

4.3 If You Are Reading For Frontier Awareness

Use:

• OpenReview
• arXiv recent lists for machine learning
• arXiv recent lists for optimization
• arXiv recent lists for probability

This is useful when you already know the area and want to track movement, not when you still need the definitions.

5 A Good Reading Ladder

If a paper feels too hard, do not force it directly.

Try this ladder instead:

1. survey or roadmap
2. book / course / notes
3. one canonical paper
4. one recent paper
5. paper clusters around the same claim family

This usually creates much faster progress than trying to brute-force novelty first.

6 How This Connects To The Site

• Books, Courses, and Notes help when the paper is too dense relative to your current background.
• Research > Surveys is a better first stop when you need a map of a field before opening individual papers.
• Research > Venues helps interpret what kind of audience and evidence culture a paper is written for.
• Paper Lab is where the site teaches the actual mechanics of reading papers once you have chosen one.