technique

Membership Inference Attack

techniqueactiveprovisionalmembership-inference-attack-25577641·1 events·first seen 3d ago

Aliases: Membership Inference Attack

Co-occurring entities

Privacy Vulnerabilities of Attention Layers in Tabular Foundation Models and Protection of High-Risk Queries AMIA

More like this (12)

Privacy Inference Attack LLM inference Conditional Inference Inference Endpoints Research Gap Inference Natural Language Inference Text Generation Inference mistral-inference Tracing Target Answers in Poisoned Retrieval Corpora via Token Influence Attribution black-box adversarial attacks OEIS Conjectures Your Privacy My Cloak: Backdoor Attacks on Differentially Private Federated Learning

Recent events (1)

5arXiv · cs.AI·3d ago·source ↗

AMIA: Attention-based membership inference attacks on tabular foundation models with k-anonymity defense

Researchers demonstrate that tabular foundation models using in-context learning are vulnerable to membership inference attacks (MIAs) via attention mechanism leakage, even when pre-trained on synthetic data. They introduce AMIA, a shadow-model-free attack exploiting transformer attention concentration patterns, achieving a 7.7% average gain over confidence-based attacks. A k-anonymity-inspired inference-time defense reduces membership leakage by 50% against AMIA and 25% against confidence-based attacks with only 3.9% performance degradation. The paper also shows fine-tuning amplifies memorization risk through confidence shifts.

Evaluation and Benchmarking AI Safety Research Membership Inference Attack Privacy Vulnerabilities of Attention Layers in Tabular Foundation Models and Protection of High-Risk Queries AMIA