Oblivious Multi-Party Machine Learning on Trusted Processors

By O. Ohrimenko, Felix Schuster, Cédric Fournet, Aastha Mehta, Sebastian Nowozin, Kapil Vaswani, Manuel Costa. 2016

602 citations recorded by the source corpus.

Subfields: Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Safety, Robustness & Interpretability

Citations in the corpus

Towards the Science of Security and Privacy in Machine Learning (2016, background): “For instance, we do not cover trusted computing platforms for ML [2].”
Deep Models Under the GAN: Information Leakage from Collaborative Deep Learning (2017, background): “In [49], the authors developed data-oblivious machine learning algorithms trained”
Mitigating Sybils in Federated Learning Poisoning (2018, background): “Trusted execution environments, such as SGX are an alternative solution [30].”
Slalom: Fast, Verifiable and Private Execution of Neural Networks in Trusted Hardware (2018, methodology): “Yet, SGX’s security comes at a performance cost, and there remains a large gap between TEEs and untrusted devices.”
Ekiden: A Platform for Confidentiality-Preserving, Trustworthy, and Performant Smart Contracts (2018, neutral): “[68] leverage SGX to offer privacy-preserving data analytics and machine learning with various security guarantees, Ryoan [41] is a distributed sandbox platform using SGX to confine privacy leakage from untrusted applications that process sensitive data.”
Contamination Attacks and Mitigation in Multi-Party Machine Learning (2018, methodology): “Secure MPC can be enabled with cryptographic techniques [6, 11, 18, 29, 32], and systems based on trusted processors such as Intel SGX [3, 5, 33].”
Machine Learning Models that Remember Too Much (2017, methodology): “Advances in data analytics frameworks based on trusted hardware such as SGX [7, 61, 66] and cryptographic protocols based on secure multi-party computation (see Section 8) may also serve as the basis for secure ML platforms.”
Machine Unlearning (2019, background): “Additionally, such schemes [33]–[35] have a large computational overhead ( G6 ) and are often used only for inference and not training.”
SoK: Security and Privacy in Machine Learning (2018, background): “For instance, we do not cover trusted computing platforms for ML [4].”
Machine Learning with Membership Privacy using Adversarial Regularization (2018, methodology): “These methods leverage various techniques including homomorphic encryption, garbled circuits, and secure multi-party computation for private machine learning on encrypted data [8, 20, 33, 37], as well as private computation using trusted hardware (e.g., Intel SGX) [26, 39].”
Making AI Forget You: Data Deletion in Machine Learning (2019, background): “cryptography [63, 16, 14, 13, 62, 31], and differential privacy [30, 21, 20, 64, 1] — do not lead to efficient data deletion, but rather attempt to make data private or non-identifiable.”
A Study of Face Obfuscation in ImageNet (2021, methodology): “To address hardware/software vulnerabilities, researchers have used enclaves —a hardware mechanism for protecting a memory region from unauthorized access—to execute machine learning workloads (Ohrimenko et al., 2016; Tramer & Boneh, 2018).”
The Limitations of Federated Learning in Sybil Settings (2020, background): “In the general ML setting, defenses rely on access to the training data [4] or access to the training process itself [7, 25, 45] to defend against adversaries.”
Privacy-preserving Machine Learning through Data Obfuscation (2018, methodology): “[26] proposed to use Intel SGX-processors to protect the training tasks against privileged adversaries in the host environment.”
MLCapsule: Guarded Offline Deployment of Machine Learning as a Service (2018, background): “[33] investigated oblivious data access patterns for a range of ML algorithms applied in SGX.”
Neural Network Inversion in Adversarial Setting via Background Knowledge Alignment (2019, background): “Secure & Privacy-Preserving ML In the wake of security and privacy threats posed to ML techniques, much research has been devoted to provisioning secure and privacy-preserving training of ML models [1, 11, 56, 61, 64] The threat models assumed by these techniques are to protect privacy of users’ dat”
The relationship between trust in AI and trustworthy machine learning technologies (2019, background): “• Safety Technologies: technologies focused on ensuring the operation of the model as intended in presence of active or passive malicious attacker [14, 73, 16, 15, 78, 56, 79, 71].”
EnclaveDB: A Secure Database Using SGX (2018, background): “Our microbenchmarks consists of a simple key-value store [31] and a set of machine learning applications [32].”
Opacus: User-Friendly Differential Privacy Library in PyTorch (2021, methodology): “Finally, we mention alternative approaches to ML training under different notions of privacy and security: using secure hardware to support oblivious training over encrypted data [14], or relying on secure multi-party computation techniques (SMPC) to train over data jointly held by several protocol ”
A Survey on Federated Learning Systems: Vision, Hype and Reality for Data Privacy and Protection (2019, methodology): “ware-based approach is to use trusted execution environment (TEE) such as Intel SGX processors [132], [146], which can guarantee that the code and data loaded inside are protected.”