Surveys
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 mapSetting: readers who want an entry point into a literature area without starting from random recent papersMain claim: the best survey depends on whether you needbackbone,bridge, orfrontier contextWhy 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:
- read the abstract, introduction, and conclusion
- locate the section that matches your direction
- extract a
small reading trail - 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
- Concentration and Common Inequalities
- Theorem Families
- 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
- Optimization
- Duality and Certificates
- Optimization for Machine Learning
- 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
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
- Generalization, Overfitting, and Validation
- Regularization, Implicit Bias, and Model Complexity
- Directions
- 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
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
- High-Dimensional Probability course -
Second pass- one of the cleanest current course-scale entry points into modern concentration and random-matrix ideas. Checked2026-04-25. - High-Dimensional Probability, 2nd edition -
First pass- durable book-scale backbone for the probability side of current theory work. Checked2026-04-25. - EE364a: Convex Optimization I -
Second pass- current official course-scale map for convexity, duality, and certificates. Checked2026-04-25. - CS224W 2024 -
Paper bridge- current official hub for graph-learning readings and topic sequencing. Checked2026-04-25. - Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges -
Paper bridge- broad survey bridge into graph and geometric learning. Checked2026-04-25. - CS229T Course Description -
Paper bridge- concise current map of modern learning-theory topics. Checked2026-04-25. - Flow Matching Guide and Code -
Paper bridge- useful current guide for the score/flow side of generative modeling. Checked2026-04-25.