Protection Against Reconstruction and Its Applications in Private Federated Learning

Abstract

In large-scale statistical learning, data collection and model fitting are moving increasingly toward peripheral devices---phones, watches, fitness trackers---away from centralized data collection. Concomitant with this rise in decentralized data are increasing challenges of maintaining privacy while allowing enough information to fit accurate, useful statistical models. This motivates local notions of privacy---most significantly, local differential privacy, which provides strong protections against sensitive data disclosures---where data is obfuscated before a statistician or learner can even observe it, providing strong protections to individuals' data. Yet local privacy as traditionally employed may prove too stringent for practical use, especially in modern high-dimensional statistical and machine learning problems. Consequently, we revisit the types of disclosures and adversaries against which we provide protections, considering adversaries with limited prior information and ensuring that with high probability, ensuring they cannot reconstruct an individual's data within useful tolerances. By reconceptualizing these protections, we allow more useful data release---large privacy parameters in local differential privacy---and we design new (minimax) optimal locally differentially private mechanisms for statistical learning problems for \emph{all} privacy levels. We thus present practicable approaches to large-scale locally private model training that were previously impossible, showing theoretically and empirically that we can fit large-scale image classification and language models with little degradation in utility.

References in corpus (15)

• Deep Learning in Neural Networks: An Overview
• Distributed Representations of Words and Phrases and their Compositionality
• Deep Learning with Differential Privacy
• Communication-Efficient Learning of Deep Networks from Decentralized Data
• NLTK: The Natural Language Toolkit
• The Sixth Data Release of the Sloan Digital Sky Survey
• Searching for Exotic Particles in High-Energy Physics with Deep Learning
• Renyi Differential Privacy
• Learning Differentially Private Recurrent Language Models
• Stochastic (Approximate) Proximal Point Methods: Convergence, Optimality, and Adaptivity
• Local Privacy, Data Processing Inequalities, and Statistical Minimax Rates
• Online Importance Weight Aware Updates
• The importance of better models in stochastic optimization
• Privacy and Statistical Risk: Formalisms and Minimax Bounds
• Lower Bounds for Locally Private Estimation via Communication Complexity

Cited by in corpus (67)

• A Survey on Federated Learning Systems: Vision, Hype and Reality for Data Privacy and Protection
• Survey on Federated Learning Threats: concepts, taxonomy on attacks and defences, experimental study and challenges
• Threats to Federated Learning: A Survey
• FMore: An Incentive Scheme of Multi-dimensional Auction for Federated Learning in MEC
• How to DP-fy ML: A Practical Guide to Machine Learning with Differential Privacy
• A Comprehensive Survey on Local Differential Privacy Toward Data Statistics and Analysis
• Privacy in Deep Learning: A Survey
• Evaluating Differentially Private Machine Learning in Practice
• FedSplit: An algorithmic framework for fast federated optimization
• Federated Learning for Healthcare Informatics
• Label Leakage and Protection in Two-party Split Learning
• Robust Federated Learning: The Case of Affine Distribution Shifts
• The Internet of Federated Things (IoFT): A Vision for the Future and In-depth Survey of Data-driven Approaches for Federated Learning
• Enhanced Security and Privacy via Fragmented Federated Learning
• Natural Language Understanding with Privacy-Preserving BERT
• LDP-FL: Practical Private Aggregation in Federated Learning with Local Differential Privacy
• Breaking the Communication-Privacy-Accuracy Trilemma
• Lower Bounds for Locally Private Estimation via Communication Complexity
• Privacy Preservation in Federated Learning: An insightful survey from the GDPR Perspective
• A Federated Learning Approach for Mobile Packet Classification
• A Systematic Literature Review on Federated Learning: From A Model Quality Perspective
• Enhancing the Privacy of Federated Learning with Sketching
• Differentially Private Meta-Learning
• SPEED: Secure, PrivatE, and Efficient Deep learning
• DFL: High-Performance Blockchain-Based Federated Learning
• Enabling Fast Differentially Private SGD via Just-in-Time Compilation and Vectorization
• Self-supervised Cross-silo Federated Neural Architecture Search
• Client Selection and Bandwidth Allocation in Wireless Federated Learning Networks: A Long-Term Perspective
• FedCon: A Contrastive Framework for Federated Semi-Supervised Learning
• Estimating Smooth GLM in Non-interactive Local Differential Privacy Model with Public Unlabeled Data
• Aggregation and Transformation of Vector-Valued Messages in the Shuffle Model of Differential Privacy
• Shuffled Model of Federated Learning: Privacy, Communication and Accuracy Trade-offs
• Efficient Sparse Secure Aggregation for Federated Learning
• Network Shuffling: Privacy Amplification via Random Walks
• BlockFLow: An Accountable and Privacy-Preserving Solution for Federated Learning
• On the Practicality of Differential Privacy in Federated Learning by Tuning Iteration Times
• Instance-optimal Mean Estimation Under Differential Privacy
• FedSiam: Towards Adaptive Federated Semi-Supervised Learning
• Element Level Differential Privacy: The Right Granularity of Privacy
• PRECAD: Privacy-Preserving and Robust Federated Learning via Crypto-Aided Differential Privacy
• Privacy For Free: Wireless Federated Learning Via Uncoded Transmission With Adaptive Power Control
• FedSel: Federated SGD under Local Differential Privacy with Top-k Dimension Selection
• SoK: Training Machine Learning Models over Multiple Sources with Privacy Preservation
• Efficient Per-Example Gradient Computations in Convolutional Neural Networks
• Voting-based Approaches For Differentially Private Federated Learning
• Renyi Differential Privacy of the Subsampled Shuffle Model in Distributed Learning
• Source Inference Attacks in Federated Learning
• Exactly Optimal and Communication-Efficient Private Estimation via Block Designs
• Federated learning with differential privacy and an untrusted aggregator
• Robust Federated Learning with Attack-Adaptive Aggregation
• Federated Learning is Better with Non-Homomorphic Encryption
• High Dimensional Restrictive Federated Model Selection with multi-objective Bayesian Optimization over shifted distributions
• On the Privacy Risks of Deploying Recurrent Neural Networks in Machine Learning Models
• Separation of Powers in Federated Learning
• Minimax Bounds for Distributed Logistic Regression
• Federated Extra-Trees with Privacy Preserving
• DPCOVID: Privacy-Preserving Federated Covid-19 Detection
• FedPower: Privacy-Preserving Distributed Eigenspace Estimation
• Utility Preserving Secure Private Data Release
• Federated Unbiased Learning to Rank
• Improving Utility and Security of the Shuffler-based Differential Privacy
• Optimizing the Numbers of Queries and Replies in Federated Learning with Differential Privacy
• Efficient-FedRec: Efficient Federated Learning Framework for Privacy-Preserving News Recommendation
• Optimal Compression of Locally Differentially Private Mechanisms
• Causally Constrained Data Synthesis for Private Data Release
• Topology-aware Differential Privacy for Decentralized Image Classification
• Design and Analysis of Uplink and Downlink Communications for Federated Learning