This repository curates a list of the most cited deep learning papers published between 2012 and 2016, serving as a foundational resource for researchers and practitioners. It aims to provide a manageable starting point for understanding seminal works in the field, categorized by research domain.

How It Works

The list is curated based on citation counts, with specific thresholds for papers published between 2012 and 2016. Papers are prioritized for their impact and applicability across various research areas, rather than being purely application-specific. The list is maintained by community contributions, with a system for adding new papers and removing older ones to maintain a "top 100" focus.

Quick Start & Requirements

No installation or execution is required. The repository is a curated list of academic papers. Links to PDFs and other resources are provided within the README.

Highlighted Details

• Comprehensive categorization of papers across key deep learning subfields: Understanding/Generalization, Optimization/Training, Generative Models, CNNs, Object Detection, NLP, Speech, and Reinforcement Learning.
• Includes sections for "More Papers from 2016," "New Papers" (published within 6 months of the last update), "Old Papers" (pre-2012 classics), and resources like datasets, frameworks, books, and video lectures.
• Citation criteria are clearly defined to guide paper inclusion and removal.
• Provides links to download all top-100 papers, collect author names, and access bib files.

Maintenance & Community

The project explicitly states it is no longer actively maintained due to the rapid pace of deep learning research since 2017. Community contributions are encouraged for suggestions and corrections.

Licensing & Compatibility

The work is licensed under a Creative Commons CC0 1.0 Universal Public Domain Dedication, allowing for unrestricted use and modification.

Limitations & Caveats

The list's primary limitation is its age, with the last significant update in 2017. It does not cover advancements in deep learning beyond that period, making it a historical snapshot rather than a current overview.