Label Leakage and Protection in Two-party Split Learning

Abstract

Two-party split learning is a popular technique for learning a model across feature-partitioned data. In this work, we explore whether it is possible for one party to steal the private label information from the other party during split training, and whether there are methods that can protect against such attacks. Specifically, we first formulate a realistic threat model and propose a privacy loss metric to quantify label leakage in split learning. We then show that there exist two simple yet effective methods within the threat model that can allow one party to accurately recover private ground-truth labels owned by the other party. To combat these attacks, we propose several random perturbation techniques, including $\text{Marvell}$, an approach that strategically finds the structure of the noise perturbation by minimizing the amount of label leakage (measured through our quantification metric) of a worst-case adversary. We empirically demonstrate the effectiveness of our protection techniques against the identified attacks, and show that $\text{Marvell}$ in particular has improved privacy-utility tradeoffs relative to baseline approaches.

arXiv commentAccepted to ICLR 2022 (https://openreview.net/forum?id=cOtBRgsf2fO)

References in corpus (15)

• Deep Learning with Differential Privacy
• Communication-Efficient Learning of Deep Networks from Decentralized Data
• Learning Differentially Private Recurrent Language Models
• Differentially Private Federated Learning: A Client Level Perspective
• iDLG: Improved Deep Leakage from Gradients
• Protection Against Reconstruction and Its Applications in Private Federated Learning
• Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware
• Threats to Federated Learning: A Survey
• Split learning for health: Distributed deep learning without sharing raw patient data
• Measuring Calibration in Deep Learning
• Differentially Private Learning with Adaptive Clipping
• AdaCliP: Adaptive Clipping for Private SGD
• QUOTIENT: Two-Party Secure Neural Network Training and Prediction
• A Review of Homomorphic Encryption Libraries for Secure Computation
• Deep Learning with Label Differential Privacy

Cited by in corpus (8)

• Vertical Federated Learning: Concepts, Advances and Challenges
• Survey on Federated Learning Threats: concepts, taxonomy on attacks and defences, experimental study and challenges
• BlindFL: Vertical Federated Machine Learning without Peeking into Your Data
• Privacy-preserving Federated Adversarial Domain Adaption over Feature Groups for Interpretability
• Vertical Federated Learning without Revealing Intersection Membership
• KDk: A Defense Mechanism Against Label Inference Attacks in Vertical Federated Learning
• Residue-based Label Protection Mechanisms in Vertical Logistic Regression
• Privacy-Aware Document Visual Question Answering