Reproduce “Anatomy of a Machine Learning Ecosystem: 2 Million Models on Hugging Face” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Benjamin Laufer, Hamidah Oderinwale, Jon Kleinberg
Researchers from Cornell University and McGill University empirically analyzed 1.86 million machine learning models on Hugging Face, applying an evolutionary biology framework to understand their development and diffusion. The study identified patterns of 'genetic' similarity, high mutation rates, and directional drifts in model traits such as licenses, languages, and tasks over generations.
Key insights
- ML models exhibit significant "genetic" family resemblance, but sibling models are often more similar to each other than to their direct parents, suggesting non-random, directed mutations.
- Model licenses demonstrate a clear evolutionary drift towards more permissive terms, frequently overriding initial upstream restrictions.
- Model documentation tends to shorten and become more automated over generations, and language support often specializes towards English, while tasks evolve to reflect the typical ML development lifecycle from feature extraction to higher-level applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.