[Paper Review] Chameleon: A Hybrid Secure Computation Framework for Machine Learning Applications

M. Sadegh Riazi, Christian Weinert+4 more|arXiv (Cornell University)|Jan 10, 2018

Cryptography and Data Security74 references51 citations

TL;DR

Chameleon is a fast two-party secure computation framework that hybridizes GC, GMW, and additive secret sharing with an offline semi-honest third party to enable private machine learning, including efficient vector dot products and neural networks.

ABSTRACT

We present Chameleon, a novel hybrid (mixed-protocol) framework for secure function evaluation (SFE) which enables two parties to jointly compute a function without disclosing their private inputs. Chameleon combines the best aspects of generic SFE protocols with the ones that are based upon additive secret sharing. In particular, the framework performs linear operations in the ring $\mathbb{Z}_{2^l}$ using additively secret shared values and nonlinear operations using Yao's Garbled Circuits or the Goldreich-Micali-Wigderson protocol. Chameleon departs from the common assumption of additive or linear secret sharing models where three or more parties need to communicate in the online phase: the framework allows two parties with private inputs to communicate in the online phase under the assumption of a third node generating correlated randomness in an offline phase. Almost all of the heavy cryptographic operations are precomputed in an offline phase which substantially reduces the communication overhead. Chameleon is both scalable and significantly more efficient than the ABY framework (NDSS'15) it is based on. Our framework supports signed fixed-point numbers. In particular, Chameleon's vector dot product of signed fixed-point numbers improves the efficiency of mining and classification of encrypted data for algorithms based upon heavy matrix multiplications. Our evaluation of Chameleon on a 5 layer convolutional deep neural network shows 133x and 4.2x faster executions than Microsoft CryptoNets (ICML'16) and MiniONN (CCS'17), respectively.

Motivation & Objective

• Motivate efficient secure function evaluation (SFE) for private machine learning and MLaaS scenarios.
• Enable two parties to compute functions with private inputs using a hybrid protocol mix to improve scalability and performance.
• Extend ABY to support sequential circuits and signed fixed-point numbers for ML tasks.
• Reduce online cryptographic work by offloading heavy operations to an offline phase via a semi-honest third party.

Proposed method

• Combine Yao’s Garbled Circuits, Goldreich–Micali–Wigderson protocols, and additive secret sharing in a hybrid framework.
• Use an offline semi-honest third party (STP) to precompute OT extensions and multiplication triples and seed expansion for efficiency.
• Represent functions as a mix of sequential circuits (for GC/GMW) and linear operations in Z2l with additive shares.
• Implement a fast vector dot product protocol based on the Du-Atallah multiplication for efficient matrix multiplications.
• Allow switching between protocols at runtime to optimize for circuit depth and online communication, with share-type translations inspired by ABY.

Experimental results

Research questions

• RQ1 How can a two-party secure computation framework achieve high efficiency for ML workloads by mixing GC, GMW, and additive secret sharing?
• RQ2 What offloading strategies (offline STP, seed expansion, precomputation) enable practical online phase performance for secure ML tasks?
• RQ3 Can Chameleon support sequential circuits and signed fixed-point arithmetic to better accommodate deep learning and ML applications?
• RQ4 What are the comparative performance gains over existing SFE frameworks and HE-based approaches for ML workloads?

Key findings

• Chameleon achieves up to 321x and 256x less online communication for arithmetic and Boolean multiplication triples compared to ABY.
• Compared to Microsoft CryptoNets, Chameleon delivers 133x faster performance on a 5-layer CNN.
• Compared to MiniONN, Chameleon delivers 4.2x faster performance under a comparable configuration.
• The framework supports signed fixed-point arithmetic with 16, 32, and 64 bit representations, enabling ML tasks that rely on matrix multiplications.
• Offline precomputation and STP-based protocols substantially reduce online cryptographic workload and communication.
• Chameleon provides a practical two-party secure computation model with an offline third party generating correlated randomness and seeds.