This repository curates highly cited papers and code related to deep learning interpretability, serving as a valuable resource for researchers and practitioners in AI safety, explainable AI (XAI), and model debugging. It provides a structured list of influential works, categorized by year and publication venue, with direct links to papers and associated code where available.

How It Works

The repository functions as a curated bibliography, meticulously compiling seminal and impactful research papers in the field of deep learning interpretability. It prioritizes works based on citation count, offering a clear indication of their influence and significance within the research community. The inclusion of code links facilitates reproducibility and practical application of the presented interpretability techniques.

Quick Start & Requirements

• Access: The primary method of engagement is by browsing the README.
• Resources: Links to 159 papers are provided, with 2 requiring external retrieval (e.g., Sci-Hub). Some papers include direct links to PyTorch, TensorFlow, Keras, Caffe, Chainer, Torch, or Matlab code repositories.
• Setup: No direct installation or execution is required; it's a reference list.

Highlighted Details

• Comprehensive list of papers sorted by citation count, highlighting foundational works.
• Includes papers from top-tier conferences like NeurIPS, ICML, CVPR, ICLR, and ECCV.
• Covers a broad spectrum of interpretability techniques, from gradient-based methods to concept-based explanations and adversarial attacks.
• Many entries link directly to code implementations, enabling practical experimentation.

Maintenance & Community

• The repository is updated "from time to time" (不定期更新).
• No specific community channels (Discord, Slack) or active maintainer information are provided in the README.

Licensing & Compatibility

• The repository itself does not specify a license.
• Individual papers and code repositories will have their own licenses, which must be checked for compatibility with commercial or closed-source use.

Limitations & Caveats

The repository is a static list and does not offer any interactive tools or integrated environment. The availability and functionality of linked code are dependent on the original authors and may not be actively maintained or compatible with current deep learning frameworks.