Protecting Sensitive Attributes by Adversarial Training Through Class-Overlapping Techniques

Lin, T. H.; Lee, Ying-Shuo; Chang, Fu-Chieh; Chang, J. Morris; Wu, Pei-Yuan

doi:10.1109/tifs.2023.3236180

Cited by 4 publications

(1 citation statement)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bayesian inference and individual privacy diference rules have been adopted to deduce user privacy [33]. Adversarial training techniques, such as overlapping technology, have been applied to deduce and protect the sensitive information of users [34]. Tese reasoning techniques can only detect specifc types of privacy information, not multiple types of privacy leakage reasoning.…”

Section: Introductionmentioning

confidence: 99%

An Automatic Privacy‐Aware Framework for Text Data in Online Social Network Based on a Multi‐Deep Learning Model

Liu,

Sun,

et al. 2023

International Journal of Intelligent Systems

View full text Add to dashboard Cite

With the increasing severity of user privacy leaks in online social networks (OSNs), existing privacy protection technologies have difficulty meeting the diverse privacy protection needs of users. Therefore, privacy-aware (PA) for the text data that users post on OSNs has become a current research focus. However, most existing PA algorithms for OSN users only provide the types of privacy disclosures rather than the specific locations of disclosures. Furthermore, although named entity recognition (NER) technology can extract specific locations of privacy text, it has poor recognition performance for nested and interest privacy. To address these issues, this paper proposes a PA framework based on the extraction of OSN privacy information content. The framework can automatically perceive the privacy information shared by users in OSNs and accurately locate which parts of the text are leaking sensitive information. Firstly, we combine the roformerBERT model, BI_LSTM model, and global_pointer algorithm to construct a direct privacy entity recognition (DPER) model for solving the specific privacy location recognition and entity nesting problems. Secondly, we use the roformerBERT model and UniLM framework to construct an interest privacy inference (IPI) model for interest recognition and to generate interpretable text that supports this interest. Finally, we constructed a dataset of 13,000 privacy-containing texts for experimentation. Experimental results show that the overall accuracy of the DPER model can reach 91.80%, while that of the IPI model can reach 98.3%. Simultaneously, we compare the proposed model with recent methods. The analysis of the results indicates that the proposed model exhibits better performance than previous methods.

show abstract

Section: Introductionmentioning

confidence: 99%

An Automatic Privacy‐Aware Framework for Text Data in Online Social Network Based on a Multi‐Deep Learning Model

Liu,

Sun,

et al. 2023

International Journal of Intelligent Systems

View full text Add to dashboard Cite

show abstract

Privacy Attacks and Defenses in Machine Learning: A Survey

Liu,

Han,

2024

Lecture Notes in Electrical Engineering

View full text Add to dashboard Cite

Learning Privacy-Preserving Embeddings for Image Data to Be Published

Li,

Lin

2023

ACM Trans. Intell. Syst. Technol.

View full text Add to dashboard Cite

Deep learning has its superiority on learning feature representations that bring promising performance in various application domains. Recent advances have shown that privacy attributes of users and patients (e.g., identity, gender and race) can be accurately inferred from image data. To avoid the risk of privacy leaking, data owners can resort to release the embeddings, rather than the original images. In this paper, we aim at learning to generate privacy-preserving embeddings from image data. The obtained embeddings are required to maintain the data utility (e.g., keep the performance of the main task like disease prediction), and to simultaneously prevent the private attributes of data instances from being accurately inferred. In addition, we also want the embeddings hard to be successfully used to reconstruct the original images. We propose a hybrid method based on multi-task learning to reach the goal. The key idea is two-fold. One is to learn the feature encoder that can benefit the main task and fool the sensitive task at the same time via iterative training and feature disentanglement. The other is to incorporate the learning of adversarial examples to mislead the sensitive attribute classification’s performance. Experiments conducted on Multi-Attribute Facial Landmark (MAFL) and NIH Chest X-rays datasets exhibit the effectiveness of our hybrid method. A set of advanced studies also show the usefulness of each model component, the difficulty in data reconstruction, and the performance impact of task correlation.

show abstract

Protecting Sensitive Attributes by Adversarial Training Through Class-Overlapping Techniques

Cited by 4 publications

References 27 publications

An Automatic Privacy‐Aware Framework for Text Data in Online Social Network Based on a Multi‐Deep Learning Model

An Automatic Privacy‐Aware Framework for Text Data in Online Social Network Based on a Multi‐Deep Learning Model

Privacy Attacks and Defenses in Machine Learning: A Survey

Learning Privacy-Preserving Embeddings for Image Data to Be Published

Contact Info

Product

Resources

About