Surveys

A curated map of surveys, course-style overviews, and broad entry points that help readers move from stable math into active literature without drowning in papers.
Modified

April 26, 2026

Keywords

surveys, literature map, tutorials, course notes, entry points

1 Why This Page

Surveys are useful, but they are easy to misuse.

People often grab a survey hoping it will remove all difficulty. Instead, a good survey usually does something narrower and more valuable:

  • gives vocabulary
  • organizes a literature patch
  • shows how results fit together
  • tells you what is stable and what is still moving

This page is here to help you pick the right survey for the right job.

2 Survey Snapshot

  • Type: top-level research map
  • Setting: readers who want an entry point into a literature area without starting from random recent papers
  • Main claim: the best survey depends on whether you need backbone, bridge, or frontier context
  • Why it matters: surveys save time only when they are chosen with a purpose

3 Three Kinds Of Surveys

3.1 1. Textbook-scale survey

This is what you read when the field still feels mathematically unstable to you.

Use it for:

  • definitions
  • standard objects
  • proof backbones
  • historical shape of the area

Examples:

  • a full book
  • lecture notes that function like a book
  • a mature graduate course

3.2 2. Course-scale survey

This is often the best starting point for research readers.

Use it for:

  • a structured reading path
  • topic selection with some curation already done
  • a bridge from math to modern applications

Examples:

  • a course homepage with reading list
  • course notes
  • official lecture schedule with curated references

3.3 3. Frontier survey

This is what you read when you already know the basics and want a snapshot of the active literature.

Use it for:

  • recent variants
  • terminology changes
  • open problems
  • where the area has moved in the last few years

Examples:

  • recent survey papers
  • living tutorials
  • current course pages tied to active research groups

4 How To Choose A Survey

Ask one question first:

What is my current bottleneck?

If the bottleneck is:

  • missing math backbone -> start with textbook-scale
  • missing big picture -> start with course-scale
  • missing frontier map -> start with frontier survey

Do not ask a frontier survey to do the job of a textbook.

5 How To Read A Survey Without Drowning

Use this rule:

  1. read the abstract, introduction, and conclusion
  2. locate the section that matches your direction
  3. extract a small reading trail
  4. stop and read one anchor paper before continuing

A survey is not a moral obligation to read from page 1 to page 80.

6 Survey Entry Point 1: High-Dimensional Probability

6.1 Best use

Choose this when concentration, random matrices, and non-asymptotic bounds keep appearing in papers you care about.

6.2 Good starting sources

6.3 Why this works

This direction benefits from having:

  • one course-scale entry point
  • one book-scale backbone
  • one shorter bridge text

6.4 Internal trail

  1. Concentration and Common Inequalities
  2. Theorem Families
  3. one of the sources above

7 Survey Entry Point 2: Convex Optimization And Solver Structure

7.1 Best use

Choose this when your papers keep using:

  • convexity
  • certificates
  • duality
  • constrained optimization
  • solver layers

7.2 Good starting sources

7.3 Why this works

Optimization is a field where the best survey is often a course rather than a standalone paper.

7.4 Internal trail

  1. Optimization
  2. Duality and Certificates
  3. Optimization for Machine Learning
  4. one of the sources above

8 Survey Entry Point 3: Graph Learning And Geometric Deep Learning

8.1 Best use

Choose this when you want a broad map of:

  • graph neural networks
  • representation learning on graphs
  • homophily / heterophily
  • spectral and geometric viewpoints

8.2 Good starting sources

8.3 Why this works

This area moves fast, so pairing one current course hub with one broad survey is often better than reading many isolated papers first.

8.4 Internal trail

  1. Eigenvalues and Diagonalization
  2. Graph Rewiring, Homophily, and Heterophily
  3. Long-Range Dependence and Oversquashing in Graphs
  4. one of the sources above

9 Survey Entry Point 4: Modern Learning Theory

9.1 Best use

Choose this when your questions revolve around:

  • generalization
  • implicit bias
  • overparameterization
  • distribution shift
  • theory for large nonlinear models

9.2 Good starting sources

9.3 Why this works

Learning theory is broad enough that a course-scale map is often more useful than one giant survey paper.

9.4 Internal trail

  1. Generalization, Overfitting, and Validation
  2. Regularization, Implicit Bias, and Model Complexity
  3. Directions
  4. one of the sources above

10 Survey Entry Point 5: Diffusion, Score, And Flow Models

10.1 Best use

Choose this when you want a wider map around:

  • score-based generation
  • diffusion models
  • flow matching
  • transport views of generation

10.2 Good starting sources

10.3 Why this works

This area changes quickly, so a short conceptual overview plus a current guide is usually a better opening than a stack of recent papers.

10.4 Internal trail

  1. Diffusion Models and Denoising
  2. Score Matching and the SDE View of Diffusion
  3. Flow Matching and Transport Views of Generation
  4. one of the sources above

11 Common Survey Mistakes

11.1 Reading surveys too early

If the survey uses notation you still cannot translate, go back to topic pages or Paper Lab.

11.2 Reading surveys too late

If you already understand the area, the survey may no longer be the best use of your time. Jump to a paper lab or current paper instead.

11.3 Treating surveys as proof sources

Many surveys explain ideas well but compress proofs aggressively. They are often for structure, not full derivation.

11.4 Turning one survey into a new unread backlog

The point is to extract a trail, not to collect a dozen references you never return to.

12 What To Learn Next

  • Venues, if you want to understand where the literature you just entered tends to live
  • Paper Lab, if you want paper-reading workflow instead of literature entry
  • Directions, if you want to choose a frontier before choosing a survey

13 Sources And Further Reading

Back to top