Robby955

Robby955 / FormalSLT

Public

Machine-verified statistical learning theory in Lean 4 — 45 modules, 19,521 lines, 0 sorry

lean4 machine-learning mathematics pac-learning statistical-learning
23
1
100% credibility
Found May 10, 2026 at 23 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Lean
AI Summary

A collection of computer-verified mathematical proofs for key results in statistical learning theory, designed for reproducibility and education.

How It Works

1
🔍 Discover FormalSLT

You come across this project while exploring the math that powers machine learning predictions.

2
📖 Read the guides

You dive into simple explanations, intuitions, and diagrams that map out the key proof paths.

3
🗺️ See the proof chain

You marvel at the visual roadmap connecting basic risk ideas to powerful generalization guarantees, all verified correct.

4
🛠️ Prepare your checker

You get the free proof-checking tool set up on your computer to explore these math treasures.

5
🚀 Run the proofs

You start the checker, and it confirms every single proof works perfectly with no gaps.

6
🔬 Explore theorems

You click through exact statements of famous bounds, seeing all assumptions clearly laid out.

✅ Trust the math

Now you understand and rely on solid, computer-vetted foundations for machine learning theory.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 23 to 23 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is FormalSLT?

FormalSLT is a Lean 4 library that machine-verifies core statistical learning theory, spanning 45 modules and 19,521 lines with zero sorry. It proves finite-sample bounds from empirical risk minimization to VC generalization, contraction mapping, sub-Gaussian chaining, algorithmic stability, and PAC-Bayes confidence intervals. Users get exact, kernel-checked theorems with all constants and assumptions in plain sight—no prose ambiguities.

Why is it gaining traction?

It stands out with total reproducibility: every factor like the 8B² exponent or Sauer-Shelah form is a Lean term, verified against Mathlib. The compact theorem spine offers clickable proofs for chalkboard classics, plus scaffolds for frontier topics like finite Dudley bridges. Developers hook on the zero-sorry audit and minimal axioms, making it a trustable base for extensions.

Who should use this?

ML theorists formalizing generalization bounds, grad students dissecting SLT proofs, or researchers prototyping PAC-Bayes/stability analyses in Lean. Ideal for anyone needing finite-class ERM tails or high-probability Rademacher controls without re-proving from scratch.

Verdict

Solid early foundation at 1.0% credibility score and 23 stars—docs are thorough, builds clean, but it's niche and maturing. Grab it if you're in formal learning theory; otherwise, watch for continuous Dudley and sharper constants.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.