Guardado en:
Detalles Bibliográficos
Autores principales: Khan, Mohammad Wahiduzzaman, Chen, Sheng, Mironov, Ilya, Zhang, Leizhen, Noor, Rabib
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2503.12801
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912278379495424
author Khan, Mohammad Wahiduzzaman
Chen, Sheng
Mironov, Ilya
Zhang, Leizhen
Noor, Rabib
author_facet Khan, Mohammad Wahiduzzaman
Chen, Sheng
Mironov, Ilya
Zhang, Leizhen
Noor, Rabib
contents Model memorization has implications for both the generalization capacity of machine learning models and the privacy of their training data. This paper investigates label memorization in binary classification models through two novel passive label inference attacks (BLIA). These attacks operate passively, relying solely on the outputs of pre-trained models, such as confidence scores and log-loss values, without interacting with or modifying the training process. By intentionally flipping 50% of the labels in controlled subsets, termed "canaries," we evaluate the extent of label memorization under two conditions: models trained without label differential privacy (Label-DP) and those trained with randomized response-based Label-DP. Despite the application of varying degrees of Label-DP, the proposed attacks consistently achieve success rates exceeding 50%, surpassing the baseline of random guessing and conclusively demonstrating that models memorize training labels, even when these labels are deliberately uncorrelated with the features.
format Preprint
id arxiv_https___arxiv_org_abs_2503_12801
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BLIA: Detect model memorization in binary classification model through passive Label Inference attack
Khan, Mohammad Wahiduzzaman
Chen, Sheng
Mironov, Ilya
Zhang, Leizhen
Noor, Rabib
Machine Learning
Cryptography and Security
Model memorization has implications for both the generalization capacity of machine learning models and the privacy of their training data. This paper investigates label memorization in binary classification models through two novel passive label inference attacks (BLIA). These attacks operate passively, relying solely on the outputs of pre-trained models, such as confidence scores and log-loss values, without interacting with or modifying the training process. By intentionally flipping 50% of the labels in controlled subsets, termed "canaries," we evaluate the extent of label memorization under two conditions: models trained without label differential privacy (Label-DP) and those trained with randomized response-based Label-DP. Despite the application of varying degrees of Label-DP, the proposed attacks consistently achieve success rates exceeding 50%, surpassing the baseline of random guessing and conclusively demonstrating that models memorize training labels, even when these labels are deliberately uncorrelated with the features.
title BLIA: Detect model memorization in binary classification model through passive Label Inference attack
topic Machine Learning
Cryptography and Security
url https://arxiv.org/abs/2503.12801