arXiv:2602.18934v2 Announce Type: replace 
Abstract: Membership inference attacks (MIAs) threaten the privacy of machine learning models by revealing whether a specific data point was used during training. Existing MIAs often rely on impractical assumptions, such as access to public datasets, shadow models, confidence scores, or knowledge of the training data distribution, making them vulnerable to defenses like confidence masking and adversarial regularization. Label-only MIAs, even under strict constraints, suffer from high query requirements per sample.
  We propose a cost-effective label-only MIA framework based on transferability and model extraction. By querying the target model $M$ using active sampling, perturbation-based selection, and synthetic data, we extract a functionally similar surrogate model $S$ on which membership inference is performed. This shifts the query overhead to a one-time extraction phase, eliminating repeated queries to $M$. Our method matches the performance of state-of-the-art label-only MIAs while significantly reducing query costs and operating under strict black-box constraints.
  On benchmark tabular datasets, we show that a query budget equivalent to testing the membership of approximately $1%$ of the training samples is sufficient to extract $S$ and achieve membership inference accuracy within $\pm 1%$ of that obtained when attacking $M$ directly. We also evaluate the effectiveness of standard defenses, including DP-SGD and regularization, proposed for label-only MIAs against our attack. Finally, we present preliminary results extending our framework to deep neural networks trained on image datasets, demonstrating promising transferability and membership inference performance under label-only access while highlighting directions for further optimization.

تم اقتراح إطار عمل جديد يسمى LoMime لتحسين هجمات استدلال العضوية (MIA) في التعلم الآلي، مما يسمح بتحديد العضوية بكفاءة باستخدام استخراج النموذج في إعدادات تعتمد فقط على التسمية. تقلل هذه الطريقة من متطلبات الاستعلام العالية المرتبطة عادةً بـ MIAs الموجودة من خلال استخدام مرحلة استخراج لمرة واحدة لإنشاء نموذج بديل للاستدلال.

Se ha propuesto un nuevo marco llamado LoMime para mejorar los ataques de inferencia de pertenencia (MIA) en el aprendizaje automático, permitiendo una determinación eficiente de la pertenencia utilizando extracción de modelos en configuraciones solo de etiquetas. Este enfoque reduce los altos requisitos de consulta típicamente asociados con los MIAs existentes al emplear una fase de extracción única para crear un modelo sustituto para la inferencia.

Un nouveau cadre nommé LoMime a été proposé pour améliorer les attaques d'inférence d'appartenance (MIA) dans l'apprentissage automatique, permettant une détermination efficace de l'appartenance en utilisant l'extraction de modèles dans des contextes uniquement basés sur des étiquettes. Cette approche réduit les exigences élevées en matière de requêtes généralement associées aux MIA existantes en employant une phase d'extraction unique pour créer un modèle substitut pour l'inférence.

A new framework named LoMime has been proposed to enhance membership inference attacks (MIAs) in machine learning, allowing for efficient membership determination using model extraction in label-only settings. This approach reduces the high query requirements typically associated with existing MIAs by employing a one-time extraction phase to create a surrogate model for inference.