Group Privacy

Group privacy in machine learning focuses on protecting the sensitive information of individuals within a dataset while still enabling collaborative model training. Current research emphasizes developing techniques like differential privacy and its variants (e.g., metric privacy) to provide formal privacy guarantees, particularly within federated learning frameworks. These methods often involve sophisticated algorithms, such as those based on prompt decomposition for LLMs or user selection strategies guided by reinforcement learning, to balance privacy preservation with model accuracy and fairness. This research area is crucial for enabling the responsible use of sensitive data in various applications, from personalized medicine to large language model deployment, while mitigating privacy risks.

Papers

December 30, 2023

ConfusionPrompt: Practical Private Inference for Online Large Language Models
Peihua Mai, Youjia Yang, Ran Yan, Rui Ye, Yan Pang
Large Language Model Privacy Amplification Black Box Large Language Model Private Inference Group Privacy

September 1, 2023

Advancing Personalized Federated Learning: Group Privacy, Fairness, and Beyond
Filippo Galli, Kangsoo Jung, Sayan Biswas, Catuscia Palamidessi, Tommaso Cucinotta
Differential Privacy Procedural Fairness Privacy Preserving Machine Learning Differential Privacy Guarantee Group Privacy

August 23, 2023

ULDP-FL: Federated Learning with Across Silo User-Level Differential Privacy
Fumiyuki Kato, Li Xiong, Shun Takagi, Yang Cao, Masatoshi Yoshikawa
Private Federated Learning Privacy Utility User Level Differential Privacy Group Privacy DP Guarantee

May 15, 2023

Privacy Auditing with One (1) Training Run
Thomas Steinke, Milad Nasr, Matthew Jagielski
Differential Privacy Modeling Assumption Empirical Privacy Privacy Preserving Computation Training Run Group Privacy

February 27, 2023

On Differentially Private Online Predictions
Haim Kaplan, Yishay Mansour, Shay Moran, Kobbi Nissim, Uri Stemmer
Information Theoretic Private Prediction Group Privacy

June 7, 2022

Group privacy for personalized federated learning
Filippo Galli, Sayan Biswas, Kangsoo Jung, Tommaso Cucinotta, Catuscia Palamidessi
LeArning Abstract Differential Privacy Local Differential Privacy Differential Privacy Guarantee Group Privacy

November 7, 2021

DQRE-SCnet: A novel hybrid approach for selecting users in Federated Learning with Deep-Q-Reinforcement Learning based on Spectral Clustering
Mohsen Ahmadi, Ali Taghavirashidizadeh, Danial Javaheri, Armin Masoumian, Saeid Jafarzadeh Ghoushchi, Yaghoub Pourasad
Federated Learning Decentralized Learning Spectral Clustering Hybrid Approach Data Sharing Deep Q Group Privacy