Chameleon: A Hybrid Secure Computation Framework for Machine Learning\n Applications

2018/01/09 by M. Sadegh Riazi, Christian Weinert, Riazi, M. Sadegh +9 · 6 citations
Computer Science · #Cryptography and Data Security #Quantum Computing Algorithms and Architecture #Coding theory and cryptography

paper · pdf · doi:10.48550/arxiv.1801.03239

Abstract

We present Chameleon, a novel hybrid (mixed-protocol) framework for secure\nfunction evaluation (SFE) which enables two parties to jointly compute a\nfunction without disclosing their private inputs. Chameleon combines the best\naspects of generic SFE protocols with the ones that are based upon additive\nsecret sharing. In particular, the framework performs linear operations in the\nring \ℤ_2^l using additively secret shared values and nonlinear\noperations using Yao's Garbled Circuits or the Goldreich-Micali-Wigderson\nprotocol. Chameleon departs from the common assumption of additive or linear\nsecret sharing models where three or more parties need to communicate in the\nonline phase: the framework allows two parties with private inputs to\ncommunicate in the online phase under the assumption of a third node generating\ncorrelated randomness in an offline phase. Almost all of the heavy\ncryptographic operations are precomputed in an offline phase which\nsubstantially reduces the communication overhead. Chameleon is both scalable\nand significantly more efficient than the ABY framework (NDSS'15) it is based\non. Our framework supports signed fixed-point numbers. In particular,\nChameleon's vector dot product of signed fixed-point numbers improves the\nefficiency of mining and classification of encrypted data for algorithms based\nupon heavy matrix multiplications. Our evaluation of Chameleon on a 5 layer\nconvolutional deep neural network shows 133x and 4.2x faster executions than\nMicrosoft CryptoNets (ICML'16) and MiniONN (CCS'17), respectively.\n

Citations

• TensorFlow: A system for large-scale machine learning

Cited by

• DeepShare: Sharing ReLU Across Channels and Layers for Efficient Private Inference
• FALCON: A Fourier Transform Based Approach for Fast and Secure Convolutional Neural Network Predictions
• Prismo: A Decision Support System for Privacy-Preserving ML Framework Selection
• Optimizing Privacy-Preserving Primitives to Support LLM-Scale Applications
• CryptGNN: Enabling Secure Inference for Graph Neural Networks
• Efficient and Verifiable Privacy-Preserving Convolutional Computation for CNN Inference with Untrusted Clouds

Related

• CrypTen: Secure Multi-Party Computation Meets Machine Learning
• Morse-STF: Improved Protocols for Privacy-Preserving Machine Learning
• S++: A Fast and Deployable Secure-Computation Framework for Privacy-Preserving Neural Network Training
• Revisiting Secure Computation Using Functional Encryption: Opportunities and Research Directions
• Partially Encrypted Machine Learning using Functional Encryption