Assessing Per-Sample Membership Inference Vulnerability without Retraining

📰 ArXiv cs.AI

Learn to assess per-sample membership inference vulnerability without retraining models, crucial for protecting sensitive data

advanced Published 27 May 2026

Action Steps

Read the paper to understand the concept of per-sample membership inference vulnerability
Apply the proposed method to assess vulnerability without retraining shadow models
Use the assessment to identify and protect sensitive data points
Implement data-dependent measures to reduce exposure to membership inference attacks
Evaluate the effectiveness of the proposed method using metrics such as precision and recall

Who Needs to Know This

Data scientists and AI engineers working on privacy-preserving machine learning models can benefit from this technique to identify vulnerable data points

Key Insight

💡 Per-sample exposure to membership inference attacks is governed by both loss and data-dependent factors

Key Takeaways

Learn to assess per-sample membership inference vulnerability without retraining models, crucial for protecting sensitive data

Full Article

Title: Assessing Per-Sample Membership Inference Vulnerability without Retraining

Abstract:
arXiv:2602.15919v2 Announce Type: replace-cross Abstract: Recent work in the privacy literature shows that sample-targeted membership inference attacks (MIAs) significantly outperform untargeted approaches by a wide margin. Motivated by this observation, we address the following question: can the privacy vulnerability of individual training points be assessed without training shadow models? We show that per-sample exposure to MIA is governed not only by a point's loss, but also by a data-depende

Read full paper → ← Back to Reads