// ⛨ Safety, Evaluation & Security
Membership Inference Attacks
Tells whether a given record was in a model’s training set – a measure of privacy leakage.
成员推断攻击 2016 Intermediate core
IN PLAIN WORDS
Models tend to be more confident on examples they were trained on. The attacker trains shadow models that mimic the target and learns to tell “seen” from “unseen” records by the output confidence. It is the standard way to audit privacy risk, also used to probe whether LLMs memorised copyrighted text or test sets; DP-SGD is the main defence.
Used in: Privacy audits, training-data leakage and test-set contamination checks
01Original paper
Membership Inference Attacks against Machine Learning Models
Reza Shokri, Marco Stronati, Congzheng Song, et al. · 2016
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Learn first
Then you can learn
- Nothing in the atlas builds on it yet
Related:Differentially Private SGD (DP-SGD)LLM Watermarking
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
For AI / data
- Verification: verified, checked 2026-10-22.
- JSON: /api/learn/algorithms.json (id:
membership-inference)