MinghuiChen43/awesome-trustworthy-deep-learning

What it solves

Deep learning models are increasingly used in real-world systems, but they can fail in ways that are unsafe, biased, or leak private information. This project gathers research on making these models trustworthy, covering topics like handling new types of data (out-of-distribution generalization), defending against malicious attacks (adversarial examples, backdoor attacks), protecting privacy (model inversion, machine unlearning), ensuring fairness, and making models interpretable.

How it works

This is an awesome list — a curated and organised collection of papers, tools, and resources. The list is updated daily with new arXiv submissions, and each paper includes a short digest summarising its key idea. Papers are categorised by topic (e.g., surveys, out-of-distribution generalisation, evasion attacks, privacy, fairness, interpretability, alignment). The README shows papers from the last year; older papers are available in a separate full list. Beyond papers, the list also links to related surveys, toolboxes, seminars, workshops, tutorials, talks, and blogs.

Who it’s for

Researchers, students, and practitioners who want to stay informed about the latest work on making deep learning models safe, reliable, and ethical. It is especially useful for those looking for a structured starting point to explore trustworthy ML research, or who want to keep up with new developments as they appear.

Highlights

• Daily updates from arXiv so the list stays current.
• Covers a wide range of topics: out-of-distribution generalisation, adversarial attacks, poisoning attacks, privacy, fairness, interpretability, alignment, and more.
• Each paper includes a concise digest and a link to the full paper.
• Provides a full list of older papers beyond the one-year window.
• Links to additional resources: surveys, toolboxes, seminars, workshops, tutorials, talks, and blogs.
• Actively maintained, and contributions (pull requests) are welcome.