Attribute Privacy

Attribute privacy focuses on protecting sensitive individual characteristics (e.g., age, gender) from inference attacks on machine learning models, particularly in text and image data. Current research emphasizes developing techniques like adversarial attacks and differentially private algorithms to mitigate this leakage, while also exploring the complex trade-offs between privacy, fairness, and model utility. This work is crucial for building trustworthy AI systems, ensuring responsible data usage, and safeguarding individual privacy in various applications, including social media and computer vision.

Papers

July 3, 2024

IncogniText: Privacy-enhancing Conditional Text Anonymization via LLM-based Private Attribute Randomization
Ahmed Frikha, Nassim Walha, Krishna Kanth Nakka, Ricardo Mendes, Xue Jiang, Xuebing Zhou
Text Anonymization Anonymised Data Private Feature Attribute Privacy Attribute Leakage

June 4, 2023

Adversary for Social Good: Leveraging Adversarial Attacks to Protect Personal Attribute Privacy
Xiaoting Li, Lingwei Chen, Dinghao Wu
Adversarial Attack Inference Attack Adversary Agent Attribute Inference Attack Attribute Privacy

February 15, 2023

Evaluating Trade-offs in Computer Vision Between Attribute Privacy, Fairness and Utility
William Paul, Philip Mathew, Fady Alajaji, Philippe Burlina
Computer Vision Procedural Fairness Task Utility Adversarial Method Attribute Inference Attack Attribute Privacy

November 18, 2022

On the Alignment of Group Fairness with Attribute Privacy
Jan Aalmoes, Vasisht Duddu, Antoine Boutet
Alignment Problem Group Fairness Inference Attack Adversarial Debiasing Attribute Privacy

September 8, 2022

Algorithms with More Granular Differential Privacy Guarantees
Badih Ghazi, Ravi Kumar, Pasin Manurangsi, Thomas Steinke
Differential Privacy Practical Algorithm Privacy Parameter Attribute Privacy

July 18, 2022

Protecting Global Properties of Datasets with Distribution Privacy Mechanisms
Michelle Chen, Olga Ohrimenko
Data Set Sensitive Data Privacy Utility Attribute Privacy

February 4, 2022

Dikaios: Privacy Auditing of Algorithmic Fairness via Attribute Inference Attacks
Jan Aalmoes, Vasisht Duddu, Antoine Boutet
Algorithmic Fairness Inference Attack Empirical Privacy Attribute Inference Attack Fairness Audit Attribute Privacy

Attribute Privacy

Papers

IncogniText: Privacy-enhancing Conditional Text Anonymization via LLM-based Private Attribute Randomization

Adversary for Social Good: Leveraging Adversarial Attacks to Protect Personal Attribute Privacy

Evaluating Trade-offs in Computer Vision Between Attribute Privacy, Fairness and Utility

On the Alignment of Group Fairness with Attribute Privacy

Algorithms with More Granular Differential Privacy Guarantees

Protecting Global Properties of Datasets with Distribution Privacy Mechanisms

Dikaios: Privacy Auditing of Algorithmic Fairness via Attribute Inference Attacks