Private Machine Learning

Private machine learning (PML) aims to train machine learning models on sensitive data while rigorously guaranteeing individual privacy, typically using differential privacy (DP) mechanisms. Current research focuses on improving the accuracy of DP models by enhancing algorithms like DP-SGD, exploring techniques like noise-tolerant pre-training and adaptive hyperparameter optimization, and leveraging public data to reduce the impact of privacy-preserving noise. These advancements are crucial for enabling the responsible use of sensitive data in various applications, addressing the inherent trade-off between privacy and model utility, and fostering trust in data-driven technologies.

Papers

October 4, 2022

Recycling Scraps: Improving Private Learning by Leveraging Intermediate Checkpoints
Virat Shejwalkar, Arun Ganesh, Rajiv Mathews, Yarong Mu, Shuang Song, Om Thakkar, Abhradeep Thakurta, Xinyi Zheng
Private Learning Model Checkpoint Private Machine Learning Advanced Textile Recycling Intermediate Checkpoint

August 26, 2022

DiVa: An Accelerator for Differentially Private Machine Learning
Beomsik Park, Ranggi Hwang, Dongho Yoon, Yoonhyuk Choi, Minsoo Rhu
Differential Privacy Differential Privacy SGD Private Machine Learning Phantom 2D Accelerator Diva 360

June 29, 2022

Matryoshka: Stealing Functionality of Private ML Data by Hiding Models in Model
Xudong Pan, Yifan Yan, Shengyao Zhang, Mi Zhang, Min Yang
Full Model Private Machine Learning Non Functional Property Information Hiding Model Confidentiality

May 27, 2022

Auditing Differential Privacy in High Dimensions with the Kernel Quantum R\'enyi Divergence
Carles Domingo-Enrich, Youssef Mroueh
Differential Privacy High Dimension Private Machine Learning DP Algorithm Privacy Model R\'enyi Divergence

May 13, 2022

On the Importance of Architecture and Feature Selection in Differentially Private Machine Learning
Wenxuan Bao, Luke A. Bauer, Vincent Bindschaedler
Differential Privacy Feature Selection Privacy Preserving Importance Aware Architecture Design Private Machine Learning Private Learning Algorithm

April 2, 2022

Production of Categorical Data Verifying Differential Privacy: Conception and Applications to Machine Learning
Héber H. Arcolezi
Machine Learning Financial Application Differential Privacy Privacy Preserving Production Incident Private Machine Learning Privacy Utility Trade Novel Concept

March 7, 2022

Continual and Sliding Window Release for Private Empirical Risk Minimization
Lauren Watson, Abhirup Ghosh, Benedek Rozemberczki, Rik Sarkar
Differential Privacy Stochastic Gradient Descent Empirical Risk Minimization Federat\textbf{T}ed Cl\textbf{A}ss Continual Lea\textbf{R}nin\textbf{G Private Machine Learning Sliding Window Private Empirical Risk Minimization

February 25, 2022

Does Label Differential Privacy Prevent Label Inference Attacks?
Ruihan Wu, Jin Peng Zhou, Kilian Q. Weinberger, Chuan Guo
Differential Privacy Private Machine Learning Label Inference Attack Private Label

February 21, 2022

Individualized PATE: Differentially Private Machine Learning with Individual Privacy Guarantees
Franziska Boenisch, Christopher Mühl, Roy Rinberg, Jannis Ihrig, Adam Dziedzic
Differential Privacy Privacy Guarantee Private Machine Learning Privacy Utility Trade User Configurable Privacy Defense

December 1, 2021

Improving Differentially Private SGD via Randomly Sparsified Gradients
Junyi Zhu, Matthew B. Blaschko
Random Sparsification Private Stochastic Private Machine Learning Dp SGD Gradient Sparsification

November 24, 2021

Differentially Private Nonparametric Regression Under a Growth Condition
Noah Golowich
Private Algorithm Private Machine Learning Hypothesis Class Private Linear Regression Growth Monitoring

November 9, 2021

The Role of Adaptive Optimizers for Honest Private Hyperparameter Selection
Shubhankar Mohapatra, Sajin Sasy, Xi He, Gautam Kamath, Om Thakkar
Integral Role Hyperparameter Optimization Hyperparameter Tuning Private Machine Learning Hyperparameter Selection Adaptive Optimizers Private Hyperparameter