A family tree for machine learning, one citation at a time

What it does This repository is a curated index of landmark machine-learning papers, organized by technique rather than by year or venue. Each entry links to the publication the author deems the original or most significant, and a small icon key flags paywalled articles, free author copies, associated code, and historical precursors. The scope runs from CNNs and Transformers to ensemble methods, classic datasets, and even infrastructure papers like MapReduce and TensorFlow.

The interesting bit The real value is the “See also” trail: for nearly every headline technique, the list nods to a forgotten predecessor—the 1980 Neocognitron before LeCun’s CNN, or the GUHA method from 1966 before association-rule mining. That makes it less a bibliography and more a genealogy of ideas, with annotations distinguishing the original breakthrough from the later popularization.

Key highlights

• Covers fifteen categories, from decision trees and boosting to games (AlphaGo, Deep Blue) and recommender systems.
• A compact icon system marks paywalled papers, free author versions, code repositories, precursors (🏛️), and follow-up refinements (🔬).
• Includes not just algorithms but also foundational datasets (ImageNet, Enron) and software systems (Torch, TensorFlow).
• The author openly admits the list is opinionated and invites corrections via issues and pull requests.

Caveats

• The selection is explicitly subjective; the author notes their choices are “by no means the final word.”
• Depth varies sharply by category: deep learning dominates, while other sections amount to only a handful of links.
• It offers no summaries, abstracts, or reading notes; you still have to open the PDFs yourself.

Verdict Worth bookmarking if you are a student or practitioner who wants to trace a technique back to its roots without drowning in citation chains. Skip it if you need code implementations, paper summaries, or a systematic literature review.