Deep learning's greatest hits, abandoned to the flood of 2017

What it does This repository is a curated bibliography of the most-cited deep learning papers published between 2012 and 2016. It locks in a roster of 100 core works across ten categories—optimization, generative models, CNN architectures, and more—each with direct PDF links. The maintainers also provide a Python script to batch-download the papers and a BibTeX file for the complete set.

The interesting bit The list is explicitly frozen: the author stopped maintaining it in 2017 because the daily volume of new papers became unmanageable. That makes it an unintentional time capsule of the field’s formative era, capturing the moment when deep learning went from niche to unavoidable.

Key highlights

• Strict citation thresholds for inclusion: papers from 2012 needed 800+ citations, while 2016 papers needed only 60+.
• One-in, one-out policy: adding a paper to the top 100 requires removing another.
• Covers foundational work across ten categories, from Batch Normalization and ResNet to GANs and YOLO.
• Includes helper scripts (fetch_papers.py, get_authors.py) and a .bib file for the full list.
• Explicitly favors seminal, cross-domain papers over narrow application-specific work.

Caveats

• The list has not been updated since 2017, so anything published after 2016 is relegated to a “New Papers” limbo or ignored entirely.
• Citation thresholds are stated but not automatically enforced; the curation ultimately relies on maintainer discretion.

Verdict Worth bookmarking if you want a no-nonsense reading list for deep learning’s foundational half-decade. Skip it if you need a living bibliography that tracks post-2016 progress.